{"loom-hero-glyph.549dcff4e1":"POPCNT","loom-hero-glyph.aebc1afa47":"XNOR","loom-hero-glyph.d89a86f5fe":"integer and bitwise","loom-hero-glyph.c22f728493":"SPARSE","loom-hero-glyph.1cd11a57b4":"answer + trace, attached by default","loom-hero-glyph.a236202512":"weave trace","loom-hero-glyph.b6b2f392b5":"new specialists join additively","loom-hero-glyph.e9fe610d86":"separately versioned and evaluated","loom-hero-glyph.f1524b9d14":"Clusters","loom-hero-glyph.f9fc695619":"6 of 528 active","loom-hero-glyph.5a24102340":"Catalogue","loom-hero-glyph.6a83b7095f":"active specialists","loom-hero-glyph.e4cf280468":"528 domain specialists, 4 to 8 active per query, deterministic top-k routing","loom-hero-glyph.7d15dd1bec":"Routing","loom-hero-glyph.076546091f":", sparse domain-specialist layer","loom-hero-glyph.ac15bcacac":"Dweve Loom [528 domain specialists]","loom-hero-glyph.de4c77b587":"Dweve Loom","loom-hero-glyph.705c87154a":"does this clause meet the retention rule?","loom-hero-glyph.b4ce2e3745":"multimodal, domain, meta, verify, transfer","loom-hero-glyph.d157c126bc":"Specialised","loom-hero-glyph.c00e187430":"reasoning, maths, science, code, language","loom-hero-glyph.e211a3681b":"Foundational","loom-hero-glyph.e8cce7d41c":"deterministic replay ready","loom-hero-glyph.f351426e58":"proof certificate attached","loom-hero-glyph.7fa3f7ee35":"integer and bitwise, no float","loom-hero-glyph.c3ee84e69b":"4 to 8 active per query","loom-hero-glyph.35471b951f":"PAP, gate subsets, top-k","loom-hero-glyph.40de2a1bcf":"Transfer and adaptation","loom-hero-glyph.03128bed90":"Verification","loom-hero-glyph.89b86ab0e6":"Language","loom-hero-glyph.77e774bacb":"Mathematics","loom-hero-glyph.0d37e7209e":"Domain industries","loom-hero-glyph.9b57338db1":"Core reasoning","loom-hero-glyph.47c75f51ad":"Permuted Agreement","loom-hero-glyph.6f6cfd8253":"Per-specialist gate subsets","loom-hero-glyph.93b5032808":"Approximate neighbours","loom-hero-glyph.03471cdb5a":"Hypervector signature","loom-consumer-glyph.cdfef203c4":"Made in Europe, for everyday use","loom-consumer-glyph.818f1c6467":"Answered by your Health specialist. You can see exactly who helped.","loom-consumer-glyph.a0dfaafc00":"Here is your answer","loom-consumer-glyph.3066b4b86e":"Only the domain specialists who know step forward. The rest stay quiet\n            rather than guess.","loom-consumer-glyph.4e51531bb5":"Not my area","loom-consumer-glyph.030055a0ee":"I know this","loom-consumer-glyph.d061a803a4":"The team answers","loom-consumer-glyph.62c4b1ba84":"Is this medicine safe to take with my other ones?","loom-consumer-glyph.7ed149da71":"You ask","loom-consumer-glyph.a2c5670f8b":"No made-up facts","loom-consumer-glyph.c25a635070":"Same question, same answer","loom-consumer-glyph.4aab1585d6":"You see who answered","loom-consumer-glyph.70f8bb9a8a":"Home","loom-consumer-glyph.adc0752adb":"Cooking","loom-consumer-glyph.22fb76e46f":"Travel","loom-consumer-glyph.3703cd2168":"Health","loom-consumer-glyph.1dc043bb8e":"Money","aura-ask-grid.af0854de67":"Aura","aura-business-sides.a0853da24b":"Spec sheet","aura-business-visuals.40417391c5":"Aura · resolves","aura-business-visuals.271973aefe":"xMidYMid meet","aura-business-visuals.3e91add48a":"BOARD","aura-showcase.eyebrow":"Dweve platform","aura-showcase.subtitle":"Coding agent and operator assistance. Demo metrics are illustrative.","aura-showcase.pillTools":"25+ tools","aura-showcase.pillRunning":"running","aura-showcase.pillAnalysing":"analysing codebase","aura-showcase.capabilityToolsTitle":"25+ tools","aura-showcase.capabilityToolsBody":"Terminal, search, lint, test, git, and more.","aura-showcase.capabilityMemoryTitle":"Semantic memory","aura-showcase.capabilityMemoryBody":"Remembers your codebase and team context.","aura-showcase.capabilityAgentsTitle":"Multi-agent orchestration","aura-showcase.capabilityAgentsBody":"Specialised agents collaborate across concerns.","aura-showcase.guaranteeAuditTitle":"Audit trail","aura-showcase.guaranteeAuditBody":"Every step is recorded with timestamps.","aura-showcase.guaranteeGuardrailsTitle":"Guardrails","aura-showcase.guaranteeGuardrailsBody":"Policies, checks, and tests always run.","aura-showcase.guaranteeApprovalTitle":"Code changes with approval","aura-showcase.guaranteeApprovalBody":"Review diffs, request changes, final sign-off.","aura-showcase.guaranteeReplayTitle":"Deterministic replay","aura-showcase.guaranteeReplayBody":"Replay any session bit-for-bit when something needs review.","aura-showcase.workflowUnderstand":"Understand","aura-showcase.workflowUnderstandCaption":"Read retry.ts, traced timeout bug","aura-showcase.workflowPlan":"Plan","aura-showcase.workflowPlanCaption":"Extract guard + cap retries","aura-showcase.workflowImplement":"Implement","aura-showcase.workflowImplementCaption":"Patched retry.ts + types.ts","aura-showcase.workflowTest":"Test","aura-showcase.workflowTestCaption":"28 tests, 94% coverage","aura-showcase.workflowVerify":"Verify","aura-showcase.workflowVerifyCaption":"Running lint & typecheck...","aura-showcase.workflowAudit":"Audit","aura-showcase.workflowAuditCaption":"Awaiting summary write-up","aura-showcase.footer":"Autonomous agents that write code and keep receipts","aura-showcase.testsPassing":"passing","aura-showcase.testsFailing":"failing","aura-showcase.modified":"modified","aura-showcase.af0854de67":"Aura","aura-showcase.73989d9c59":"Running","aura-showcase.f9e94338d0":"Analyzing codebase","aura-showcase.d7a484140f":"Workflow","aura-showcase.287fda8114":"Unified","aura-showcase.80e13549b4":"Coverage","aura-showcase.b71cc78b34":"Extracted isRetryableError","aura-showcase.c37a8a9e2b":"to handle\n                    network timeouts, 5xx responses, and idempotent-safe\n                    conditions. Pure helper, fully unit-tested.","aura-tech-middles.7fba7d244f":"Hot path","aura-tech-middles.1be47ef71c":"Cross-cutting","aura-tech-middles.92096d78df":"Ingress","aura-tech-middles.8e05a902ed":"Egress","aura-tech-middles.49a66c994d":"Builtin agents","aura-tech-middles.a60792e996":"Outer loop","aura-tech-middles.0dfd4db698":"Inner loop","aura-tech-middles.bbd88bff92":"Incoming request","aura-tech-middles.a618b4be8d":"Query","aura-tech-middles.a4942c26d0":"SELECT","aura-tech-middles.2cb479aa15":"* FROM","aura-tech-middles.1c255b91e7":"WHERE","aura-tech-middles.2164bde64d":"ORDER BY","aura-tech-middles.82dd2cdf36":"Block","aura-tech-middles.271973aefe":"xMidYMid meet","core-extras.96d330537a":"Base IR","core-extras.bee831a169":"Substrate","core-extras.61a0822826":"Apps on Core","core-extras.15438cc098":"not supported","core-extras.ad565d9d01":"Feature","core-extras.0155bf03e5":"Segment legend","core-extras.8b7110078e":"Algorithm groups by data type","core-extras.25ab4276a6":"Scrollable capability matrix","loom-business-glyph.590c236a32":", one Core","loom-business-glyph.ff221d4752":"CPU to GPU:","loom-business-glyph.15010df884":"the active weave","loom-business-glyph.fe4dd528ba":"By design, not bolted on","loom-business-glyph.abfdec56ba":"Art. 25","loom-business-glyph.a5b7d57a85":"GDPR","loom-business-glyph.22c2234e11":"Made in Europe","loom-business-glyph.b7a3f27883":"Risk removed","loom-business-glyph.0fed3b96cc":"Loom","loom-business-glyph.9e0f79df9d":"Chat tab","loom-business-glyph.8ac7aa173f":"What you get","loom-business-glyph.7a991efc81":"A weave trace on every output. Built for procurement, audit, and\n            regulated workflows inside Europe.","loom-business-glyph.adc8c4f82b":"An answer you can","loom-business-glyph.challengeAccent":"challenge","loom-business-glyph.950740468b":"The management story","loom-business-glyph.41b81eb8db":"Approved","loom-business-glyph.0708e88794":"Procurement view","loom-business-glyph.de4c77b587":"Dweve Loom","loom-business-glyph.0989179299":"No network at all","loom-business-glyph.5df2ebe7c0":"Air-gapped","loom-business-glyph.f6c83ef308":"Infrastructure you control","loom-business-glyph.ca84f59176":"On-prem","loom-business-glyph.7eaf4d4f1f":"Operated from Arnhem","loom-business-glyph.9b78aaba24":"Managed Mesh","loom-business-glyph.4e048161ef":"Through Fabric, or directly under a licence","loom-business-glyph.96f32ed98a":"Cloud lock-in","loom-business-glyph.dbe0a00e6c":"Provenance and replay","loom-business-glyph.458e74f6b2":"Nothing for the auditor","loom-business-glyph.c3ee84e69b":"4 to 8 active per query","loom-business-glyph.5635a44134":"Runaway query cost","loom-business-glyph.52da239cf7":"Trace on every answer","loom-business-glyph.45be69208e":"Unexplainable output","loom-business-glyph.8db48970f0":"Complete state pinned","loom-business-glyph.a110ee6543":"Changes tomorrow","loom-business-glyph.c0f85d6679":"Replay","loom-business-glyph.077d131ec0":"Rule that fired","loom-business-glyph.7257c63b8c":"No version","loom-business-glyph.bb9cf14180":"Policy","loom-business-glyph.ddcffc8102":"Specialist named","loom-business-glyph.b2f2dcd4e5":"One black box","loom-business-glyph.a4d81396d4":"Attribution","loom-business-glyph.d6079fc753":"A record you can cite","loom-business-glyph.2da4ec5795":"A paragraph","loom-business-glyph.a16a4eda7c":"Answer","loom-hidden-jobs.e55685f945":"Visible work","loom-hidden-jobs.d036534321":"exposed","loom-hidden-jobs.cf37a73514":"hidden versus owned","loom-hidden-jobs.1ba1e61440":"The monolith","loom-hidden-jobs.1e68cce129":"eight unnamed jobs, one paragraph out","loom-hidden-jobs.6144cd5e38":"classify","loom-hidden-jobs.e4f23495a1":"extract","loom-hidden-jobs.413bc6164a":"retrieve","loom-hidden-jobs.9f0d66272a":"rank","loom-hidden-jobs.a8f700e7c0":"calculate","loom-hidden-jobs.f544e4c95b":"solve","loom-hidden-jobs.86a4261d84":"verify","loom-hidden-jobs.f7c1380a0d":"explain","loom-hidden-jobs.95d5dfdad2":"The Loom version","loom-hidden-jobs.a2280d8b82":"the same jobs as typed nodes with distinct outputs","loom-hidden-jobs.8d767bf5b7":"class","loom-hidden-jobs.6325f8c630":"span","loom-hidden-jobs.100ec4462d":"evidence","loom-hidden-jobs.48a3661d84":"status","loom-hidden-jobs.95dea6c1d5":"hidden jobs","loom-hidden-jobs.45dc634f42":"owned responsibilities","loom-four-weaves.e0f45b8843":"Task patterns","loom-four-weaves.87b6ebf9c6":"woven","loom-four-weaves.83055097ee":"no repeated topology","loom-four-weaves.d6c3609d04":"label a ticket","loom-four-weaves.212879992e":"1 thread lit","loom-four-weaves.391a8866cb":"find the date","loom-four-weaves.7d7d12b881":"2 threads lit","loom-four-weaves.28a3d48e6c":"find the policy","loom-four-weaves.ac2a66af85":"5 threads lit","loom-four-weaves.082257ee0b":"explain the outcome","loom-four-weaves.ae906d029c":"deep weave: reasoning, solver, specialists","loom-four-weaves.710619412e":"the task is the pattern","loom-four-weaves.2522cf0b58":"the loom never repeats itself","loom-perception-stations.c4645ad1b6":"Perception stations","loom-perception-stations.229670d592":"seen","loom-perception-stations.2eef901046":"addressable perception","loom-perception-stations.372ea08cab":"text","loom-perception-stations.ec96667ac8":"documents","loom-perception-stations.19f49d8526":"images","loom-perception-stations.a06a492959":"audio","loom-perception-stations.ffbaf58f12":"video","loom-perception-stations.52a6391f3d":"structured","loom-perception-stations.e666e847a0":"Typed observations","loom-perception-stations.057fe37727":"span, page 4, clause 12.1","loom-perception-stations.4786f8583c":"table cell, B7","loom-perception-stations.b4d829f1a4":"entity, counterparty","loom-perception-stations.e0d5b0ab7e":"event, at 00:41","loom-perception-stations.9ab8e7dbbf":"a later constraint check points back to the exact span and cell","loom-perception-stations.24352ba2e5":"not one embedding cloud","loom-perception-stations.9a151ff43f":"state the weave can cite","loom-memory-feeds.9b2beaac02":"Memory flow","loom-memory-feeds.6543f14e1a":"scoped","loom-memory-feeds.8bc68764fd":"access by responsibility","loom-memory-feeds.8178ea208a":"working","loom-memory-feeds.83934b649f":"feeds the active task state","loom-memory-feeds.bb7c31c52b":"episodic","loom-memory-feeds.99e3850f49":"feeds the domain specialist its cases","loom-memory-feeds.94fb17e32b":"semantic","loom-memory-feeds.45733e57e1":"feeds the solver facts and constraints","loom-memory-feeds.721ce72cde":"procedural","loom-memory-feeds.539ee5bee5":"feeds reusable ways of working","loom-memory-feeds.072f0738bc":"unrelated history stays visibly closed to the classifier","loom-memory-feeds.141f40846d":"recall is scoped","loom-memory-feeds.a4cd1c60cb":"no unbounded transcript","loom-search-funnel.17e214ac4f":"staged","loom-search-funnel.1dd98960b9":"search as a weave","loom-search-funnel.201b513c76":"the final evidence stays linked to its source passages","loom-search-funnel.2f6d3eca5a":"ordered","loom-search-funnel.413bc6164a":"retrieve","loom-search-funnel.47723f290d":"Cheap representation up front, deep judgement only at the tail.","loom-search-funnel.69f494bfbc":"rerank","loom-search-funnel.75ebcb361c":"score","loom-search-funnel.7cf9c298ed":"Search stages","loom-search-funnel.7e6a97c564":"represent","loom-search-funnel.85f2f27b86":"Cognition cost across the funnel","loom-search-funnel.86a4261d84":"verify","loom-search-funnel.9939b4cf88":"the frontier only","loom-search-funnel.9f0d66272a":"rank","loom-search-funnel.cf46065b2d":"candidate set","loom-search-funnel.d5525495e0":"six separate judgements","loom-search-funnel.f019fac0df":"binary embeddings","loom-search-funnel.f4e91f3e65":"relevance","loom-search-funnel.f62ec6dd4b":"support checked","loom-search-funnel.ffc9d249e6":"depth only at the tail","loom-solver-switchboard.a4332defe7":"Native solvers","loom-solver-switchboard.0c0b42a5dc":"docked","loom-solver-switchboard.120f8be6f7":"native methods","loom-solver-switchboard.d20602f525":"SAT","loom-solver-switchboard.11e9ba26c9":"schedule","loom-solver-switchboard.085ac9a633":"LP","loom-solver-switchboard.42d9d2622f":"temporal","loom-solver-switchboard.29a184b65c":"graph","loom-solver-switchboard.06f9b0facd":"symbolic","loom-solver-switchboard.6c943beb4d":"synthesis","loom-solver-switchboard.d9965722ca":"each card docks into its solver family","loom-solver-switchboard.aea9b3bfac":"UNSAT","loom-solver-switchboard.4848901b39":"optimal","loom-solver-switchboard.88c0095873":"infeasible","loom-solver-switchboard.50d8b4a941":"unknown","loom-solver-switchboard.ef64363d23":"proof","loom-solver-switchboard.eb9ceb1149":"counterexample","loom-solver-switchboard.a502a38a60":"outputs stay solver outcomes, never prose confidence","loom-solver-switchboard.16edaa9bcc":"the problem has a shape","loom-solver-switchboard.9b699291a7":"the answer keeps its status","loom-solver-switchboard.a5a0d15d4a":"problem","loom-solver-switchboard.1822f249c0":"solver family","loom-solver-switchboard.48a3661d84":"status","loom-solver-switchboard.caeff09868":"third-party backends, integrated","loom-reasoning-branches.b48f4fea48":"Reasoning paths","loom-reasoning-branches.bc1f2fe54a":"dispatched","loom-reasoning-branches.e2c6116891":"reasoning shapes","loom-reasoning-branches.701f7576c7":"one observation","loom-reasoning-branches.1ae7d5617d":"deductive","loom-reasoning-branches.67f06dc656":"a derivation","loom-reasoning-branches.4b660a7844":"abductive","loom-reasoning-branches.ac97269685":"ranked hypotheses","loom-reasoning-branches.88eb829d4d":"causal","loom-reasoning-branches.6e59de43c1":"a dependency path","loom-reasoning-branches.e4c21fa1d9":"counterfactual","loom-reasoning-branches.d25c99bbdf":"a changed outcome","loom-reasoning-branches.017b1c81c6":"a metacognitive selector compares the paths","loom-reasoning-branches.9ffb19319d":"different artefacts","loom-reasoning-branches.99171cb7e8":"not one confidence score","loom-reasoning-branches.05bda8a9de":"pruned","loom-reasoning-branches.1e61fe1e47":"kept","loom-reasoning-branches.2fa9505626":"artefact","loom-reasoning-branches.3d46f1a9a5":"branches that fail their own check are pruned and left visible","loom-expert-backdrop.7f7294b008":"Specialist catalogue","loom-expert-backdrop.d8f7d26b00":"routed","loom-expert-backdrop.ee278d7fad":"528, sparsely lit","loom-expert-backdrop.6ca12fdc19":"the wide catalogue, dim","loom-expert-backdrop.87d0a0fc94":"the router lights a bounded set","loom-expert-backdrop.df040a130f":"contract law","loom-expert-backdrop.2b3760ade4":"temporal reasoning","loom-expert-backdrop.65279e1ecf":"verification","loom-expert-backdrop.9b69241c1a":"the selected specialists feed the existing graph","loom-expert-backdrop.ec1414ee5f":"the graph stays the hero","loom-expert-backdrop.eb9171c3c9":"the catalogue stays the backdrop","loom-expert-backdrop.6a7b34bf54":"528 domain specialists","loom-expert-backdrop.e9f85dc9ea":"4 to 8 active, up to 20","loom-expert-backdrop.78e596b662":"the rest dormant","loom-expert-backdrop.ece99c87f1":"the existing graph","loom-expression-sockets.06d5fde86a":"The exact location comes from extraction, unchanged.","loom-expression-sockets.1e54b9ec62":"the transformer speaks","loom-expression-sockets.2ef88f93b1":"facts enter through hard sockets, not a prompt summary","loom-expression-sockets.3108274514":"It does not decide the status","loom-expression-sockets.31961584a3":"exact span","loom-expression-sockets.3d91f8b4d5":"constrained","loom-expression-sockets.64e4d74ea9":"verified evidence","loom-expression-sockets.655c6f28e9":"solver status","loom-expression-sockets.90298d1506":"What the transformer never does","loom-expression-sockets.965ea80636":"It does not fabricate the evidence","loom-expression-sockets.b6cf1d154c":"the explanation, expressed last","loom-expression-sockets.b77632c2d9":"It does not invent the span","loom-expression-sockets.dca2059d49":"The cited document arrives verified, with provenance.","loom-expression-sockets.ea5e3a495e":"The SAT or UNSAT result comes from the solver.","loom-expression-sockets.f383235b08":"the weave decides what is true","loom-expression-sockets.fd98dbfb01":"language as one thread","loom-expression-sockets.ff297c71bb":"Structured expression","loom-thread-spool.1e4fa1c675":"Native paths","loom-thread-spool.6f5b01580c":"native","loom-thread-spool.8ba3ed9a84":"each material its own","loom-thread-spool.5d5638b12a":"Bipolar learned path","loom-thread-spool.20ef76a6e5":"models, embeddings, routers, transformers, specialists","loom-thread-spool.ec90cd62a5":"trained binary and integer, run one-bit","loom-thread-spool.d9ca5c93f7":"Native symbolic materials","loom-thread-spool.b08df4dccc":"graphs, intervals, constraints, proofs","loom-thread-spool.36ea364262":"kept in their own representations","loom-thread-spool.71ca1eee01":"the weave combines them without pretending they are identical","loom-thread-spool.027b7bc32a":"no float model behind a quantiser","loom-thread-spool.3b11e67466":"no neural costume on a solver","loom-cost-meter.09f3d23f56":"You are billed for","loom-cost-meter.0c150c2cc4":"The rest of the catalogue, which stays available at no added cost.","loom-cost-meter.285c23ca81":"cost follows lit nodes","loom-cost-meter.29cd95aa72":"deep weave","loom-cost-meter.33620369b5":"The bill","loom-cost-meter.4cd8dcf678":"Work and cost","loom-cost-meter.637b77774e":"catalogue size is fixed cost zero","loom-cost-meter.7979249e4b":"reasoning, solver, specialists lit","loom-cost-meter.a091a0d6f4":"search stages lit","loom-cost-meter.a0d6b6d0d9":"one classifier lit","loom-cost-meter.c3d4d5c216":"tiny weave","loom-cost-meter.d42c8a8ffe":"metered","loom-cost-meter.d8e07cc121":"The nodes the weave actually lit for this query.","loom-cost-meter.dd789acfd7":"medium weave","loom-cost-meter.e176ea327c":"You are not billed for","loom-cost-meter.e55c84f5df":"activation is the bill","loom-cost-meter.f58d4c1f63":"the meter follows lit nodes, never the full catalogue","loom-clip-in.33469d3faa":"pinned","loom-clip-in.d4bdfd88da":"the weave","loom-clip-in.a109f0e810":"patterns pinned","loom-clip-in.d52a19b117":"thread clipped in","loom-clip-in.e670038522":"retrained","loom-clip-in.f4ebbef538":"Add a thread","loom-clip-in.172415f02e":"added","loom-clip-in.9bf62de24f":"additive growth","loom-clip-in.60565443fa":"new solver thread","loom-clip-in.28b4542737":"arrives with its own tests and version","loom-clip-in.e2d3389a59":"appears in one new task pattern","loom-clip-in.7efb97f118":"earlier patterns remain pinned and unchanged","loom-clip-in.7f4b6ed2e4":"the weave grows richer","loom-clip-in.0ff59419f2":"nothing else is retrained","loom-red-node.c830f6d304":"Local correction","loom-red-node.17a3a24e36":"repaired","loom-red-node.fc7e0ebd1b":"corrections addressed","loom-red-node.05fcd05070":"The failed trace","loom-red-node.a001cafb57":"perception","loom-red-node.3a93bc3e5a":"retrieval","loom-red-node.26265f6c77":"ranking","loom-red-node.b93db90c74":"constraints","loom-red-node.ec9814396b":"expression","loom-red-node.26bfdb383b":"ranking: the weak thread, in red","loom-red-node.fc368f6178":"the repair lands on that node","loom-red-node.c4248008bd":"replay turns green, neighbours stay pinned","loom-red-node.7316e0412d":"no global retraining project","loom-red-node.aa03122df8":"one owner, one fix","loom-weave-map.24312d9b63":"Weave map","loom-weave-map.1d28e511b1":"readable","loom-weave-map.96c810fc76":"observable cognition","loom-weave-map.aa8ad02035":"01 perception","loom-weave-map.f50b093cf9":"02 memory","loom-weave-map.721d05cb59":"03 retrieval","loom-weave-map.718316c269":"04 reasoning","loom-weave-map.4c0a5f8cb3":"05 solver","loom-weave-map.e67a3222e6":"06 verify","loom-weave-map.eb7cd04564":"07 express","loom-weave-map.2fca1503dc":"every handoff numbered, every node versioned","loom-weave-map.1c05667e32":"deliberately excluded: private model thoughts","loom-weave-map.07fdb93691":"the map of what ran","loom-weave-map.8a4b0203a5":"nothing else claimed","loom-refusal-gate.01038d99f4":"refusal as outcome","loom-refusal-gate.0a047b1a3f":"the fifth refuses, typed","loom-refusal-gate.18cd42d86d":"gated","loom-refusal-gate.37a5301a88":"result","loom-refusal-gate.3a93bc3e5a":"retrieval","loom-refusal-gate.3d9d4474a2":"four answers leave","loom-refusal-gate.47b73d4f0c":"Refusal gate","loom-refusal-gate.65279e1ecf":"verification","loom-refusal-gate.6c3e633a6c":"the failed verification node and the missing input stay visible","loom-refusal-gate.81c5d49be9":"memory","loom-refusal-gate.9662a505ed":"reasoning","loom-refusal-gate.a001cafb57":"perception","loom-refusal-gate.c1f9ff509b":"insufficient evidence","loom-posture-stamps.170a41e339":"It stops on the same conditions.","loom-posture-stamps.2f0d4f153f":"on-premises","loom-posture-stamps.33489ad5a5":"the boundary changes","loom-posture-stamps.343b6d42aa":"WEAVE","loom-posture-stamps.39f7f4b795":"The weave means the same thing everywhere.","loom-posture-stamps.461ed326bf":"The same run record is produced.","loom-posture-stamps.52ea612e1c":"The same catalogue stays addressable.","loom-posture-stamps.53354e1327":"air-gapped","loom-posture-stamps.6881023a05":"managed Mesh","loom-posture-stamps.69bfb7745e":"Deployment boundaries","loom-posture-stamps.7458a9ee82":"local-first sockets","loom-posture-stamps.77660df195":"the cognitive contract does not","loom-posture-stamps.7a80c9860d":"Refusal rules","loom-posture-stamps.809c9d1822":"external sockets open, controlled","loom-posture-stamps.889080aff0":"Unchanged in every posture","loom-posture-stamps.88fac1043e":"Specialist catalogue","loom-posture-stamps.8c305a07c6":"consistent","loom-posture-stamps.94723eeb50":"Cognitive contract","loom-posture-stamps.a0f2c69a18":"edge","loom-posture-stamps.dbc76d36a7":"sockets controlled by you","loom-posture-stamps.de9a0de179":"deployment postures","loom-posture-stamps.e1f0ffe317":"Execution trace","loom-posture-stamps.eb09df8c26":"no outbound sockets at all","loom-stack-loom.14f03f58af":"The Loom stack","loom-stack-loom.f4a655b717":"placed","loom-stack-loom.f1242543e5":"clear responsibilities","loom-stack-loom.a010de5c61":"Fabric","loom-stack-loom.3194c8b1eb":"people ask, inspect, use","loom-stack-loom.485a04fef4":"Nexus","loom-stack-loom.32e10c1777":"organises work with agents and authority","loom-stack-loom.0fed3b96cc":"Loom","loom-stack-loom.98375d4a6f":"weaves the cognition","loom-stack-loom.4a86080d67":"Spindle","loom-stack-loom.4e76d6548e":"governed, lossless knowledge","loom-stack-loom.68836c550e":"Core","loom-stack-loom.c519a526c4":"executes every operation","loom-stack-loom.873132a799":"Mesh","loom-stack-loom.b4222f17be":"distributed compute underneath","loom-stack-loom.a94a976303":"cognition in the middle","loom-stack-loom.2ee19976e4":"the stack carries the rest","loom-e-graph-compiler.9dda1fc214":"Request compiler","loom-e-graph-compiler.2bb6b986c5":"active","loom-e-graph-compiler.79e47b59ba":"classify to emit","loom-e-graph-compiler.9af1dba608":"classify task","loom-e-graph-compiler.a1a8140075":"infer graph","loom-e-graph-compiler.bb12462363":"resolve versions","loom-e-graph-compiler.dd1ee96191":"activate nodes","loom-e-graph-compiler.101a19f1bb":"run parallel branches","loom-e-graph-compiler.86a4261d84":"verify","loom-e-graph-compiler.ee682dca1e":"emit result plus trace","loom-e-graph-compiler.23f2880adb":"> loom compile task:ticket-triage","loom-e-graph-compiler.a4a1121f6b":"only required nodes activate","loom-e-graph-compiler.a5195afe76":"one result, one weave trace","loom-e-thread-contract.f2996bdb72":"Thread descriptor","loom-e-thread-contract.07235a8030":"signed","loom-e-thread-contract.b3e62f57c8":"typed end to end","loom-e-thread-contract.0f03b7ccf1":"input shape","loom-e-thread-contract.c90398d7a7":"output shape","loom-e-thread-contract.fa6838ad71":"capability signature","loom-e-thread-contract.f8393338a2":"version hash","loom-e-thread-contract.9777b44d4a":"cost model","loom-e-thread-contract.21eee0794a":"determinism contract","loom-e-thread-contract.5fb6293f49":"failure modes","loom-e-thread-contract.8f1ec90942":"evidence surface","loom-e-thread-contract.2e322bdd2f":"sibling descriptors carry different output types","loom-e-thread-contract.3b5b9cc799":"not an untyped callback","loom-e-thread-contract.008fb0f052":"a contract the planner composes","loom-e-graph-planner.624e1172f1":"Task graphs","loom-e-graph-planner.80e61027a7":"planned","loom-e-graph-planner.5f10199d22":"shallow or deep","loom-e-graph-planner.8f3a646866":"Deep graph","loom-e-graph-planner.cf9662d988":"perception fan-out, memory join, solver branch, specialist branch, verification gate, expression","loom-e-graph-planner.808f3699a9":"Shallow graph","loom-e-graph-planner.03b2e86aeb":"one classifier node","loom-e-graph-planner.5d2a234b62":"no fixed pipeline","loom-e-graph-planner.f6f7031af0":"the topology fits the task","loom-e-binary-space.81719f107f":"Binary space","loom-e-binary-space.d18aac96b9":"shared","loom-e-binary-space.6fe55a45fd":"the bipolar plane","loom-e-binary-space.08541c6d8b":"XNOR similarity","loom-e-binary-space.b0d7d5229d":"population count","loom-e-binary-space.85c69c067b":"XOR bind","loom-e-binary-space.e000be7aac":"majority bundle","loom-e-binary-space.fe9cc9f59b":"passes through model, memory, retrieval, and routing nodes","loom-e-binary-space.bc84aa38d6":"a boundary marks conversion to native CNF, graph, interval, or proof","loom-e-binary-space.31c50967be":"learned threads share the plane","loom-e-binary-space.fe546419fe":"symbolic structures stay native","loom-e-perception-fabric.ad18cc3477":"Perception pipeline","loom-e-perception-fabric.2ac449ab52":"converged","loom-e-perception-fabric.4922bd4046":"addressable fields","loom-e-perception-fabric.372ea08cab":"text","loom-e-perception-fabric.ec96667ac8":"documents","loom-e-perception-fabric.19f49d8526":"images","loom-e-perception-fabric.a06a492959":"audio","loom-e-perception-fabric.ffbaf58f12":"video","loom-e-perception-fabric.52a6391f3d":"structured","loom-e-perception-fabric.6f7e01e79e":"converge into a typed observation graph","loom-e-perception-fabric.ee7aaae2f5":"downstream nodes consume exact fields and spans","loom-e-perception-fabric.ad630ca6df":"not generic prompt context","loom-e-perception-fabric.6b2a69e8af":"observations the weave can cite","loom-e-memory-fabric.4e4a49aef1":"Memory tiers","loom-e-memory-fabric.d8f7d26b00":"routed","loom-e-memory-fabric.bb01ed309a":"lifecycle events","loom-e-memory-fabric.8178ea208a":"working","loom-e-memory-fabric.bb7c31c52b":"episodic","loom-e-memory-fabric.94fb17e32b":"semantic","loom-e-memory-fabric.721ce72cde":"procedural","loom-e-memory-fabric.66ebb75e38":"a consolidation event promotes an episode into semantic state","loom-e-memory-fabric.95cb6fd1e6":"a decay event removes low-value working state","loom-e-memory-fabric.574b335aec":"content-addressed recall","loom-e-memory-fabric.6661c976a1":"bounded by lifecycle","loom-e-memory-fabric.8d5d1b1b15":"Memory cell record","loom-e-memory-fabric.b7b30eaccb":"content hash key","loom-e-memory-fabric.292fffc736":"one of four tiers","loom-e-memory-fabric.a5d17d4cc4":"stored value","loom-e-memory-fabric.7b80c660fe":"recency tick","loom-e-memory-fabric.fbb81eb805":"salience score","loom-e-memory-fabric.8121ffef88":"consolidate or decay","loom-e-memory-fabric.98e2670cbb":"Content-addressed, bounded by lifecycle","loom-e-micro-models.64d85431d8":"Micro-models","loom-e-micro-models.c072e8329e":"bounded","loom-e-micro-models.b707ce6286":"one responsibility each","loom-e-micro-models.49061e96f4":"classifier","loom-e-micro-models.0bfb6034e4":"span extractor","loom-e-micro-models.5f2584c4be":"ranker","loom-e-micro-models.eb9424eb30":"reranker","loom-e-micro-models.1ff820cf32":"embedder","loom-e-micro-models.3c5b492776":"gate","loom-e-micro-models.fff839dbb7":"a gate model decides whether a deeper branch activates","loom-e-micro-models.44a8e3bd89":"small because narrow","loom-e-micro-models.cf9f0e08ba":"not because secondary","loom-e-micro-models.91e30e3897":"Unit invocations","loom-e-micro-models.96f0ea849f":"Narrow units, one typed contract each","loom-e-micro-models.63b6594ddf":"each unit returns one typed contract","loom-e-micro-models.f400a3b403":"deeper branch skipped","loom-e-search-graph.258bc84373":"Search frontier","loom-e-search-graph.17e214ac4f":"staged","loom-e-search-graph.8db529dc2a":"binary space","loom-e-search-graph.a8923bb6da":"encode","loom-e-search-graph.413bc6164a":"retrieve","loom-e-search-graph.75ebcb361c":"score","loom-e-search-graph.9f0d66272a":"rank","loom-e-search-graph.69f494bfbc":"rerank","loom-e-search-graph.86a4261d84":"verify","loom-e-search-graph.c5f2481ee3":"source-addressed evidence leaves the final verifier","loom-e-search-graph.1b70605f4c":"candidate generation stays compact","loom-e-search-graph.0a62d6b10d":"reranking touches the frontier only","loom-e-bipolar-transformer.f293fe64c1":"Typed transformer","loom-e-bipolar-transformer.3d91f8b4d5":"constrained","loom-e-bipolar-transformer.06fb2de45d":"native one-bit","loom-e-bipolar-transformer.26a8e4d9ff":"span offsets","loom-e-bipolar-transformer.3a6bb905fc":"solver model","loom-e-bipolar-transformer.c0ea6dfc6a":"evidence set","loom-e-bipolar-transformer.4a8c07dd62":"specialist findings","loom-e-bipolar-transformer.b990ceea7d":"emits structured text","loom-e-bipolar-transformer.5989e92235":"no hidden float master","loom-e-bipolar-transformer.d9e47cf871":"facts enter through hard ports","loom-e-reasoning-dispatch.94414ffcf6":"Reasoning dispatch","loom-e-reasoning-dispatch.bc1f2fe54a":"dispatched","loom-e-reasoning-dispatch.b33fbe8c4a":"mode recorded","loom-e-reasoning-dispatch.269cdc0042":"query A to causal, counterfactual and probabilistic","loom-e-reasoning-dispatch.23f902707b":"query B to deductive, temporal and deontic","loom-e-reasoning-dispatch.4c680c203f":"outputs remain typed, the mode joins the trace","loom-e-reasoning-dispatch.35c5aa8bc7":"fifteen named forms","loom-e-reasoning-dispatch.99171cb7e8":"not one confidence score","loom-e-reasoning-dispatch.query":"query","loom-e-reasoning-dispatch.classifier":"classifier","loom-e-reasoning-dispatch.forms":"reasoning forms","loom-e-reasoning-dispatch.typedOut":"typed out","loom-e-reasoning-dispatch.traceJoined":"mode joined to trace","loom-e-solver-atlas.817dec2f41":"Solver atlas","loom-e-solver-atlas.e7290f6ca5":"grouped","loom-e-solver-atlas.17a76c0913":"24 regions","loom-e-solver-atlas.38dd669280":"logical","loom-e-solver-atlas.ba42eddff9":"SMT","loom-e-solver-atlas.b93db90c74":"constraints","loom-e-solver-atlas.50a1dca141":"optimisation","loom-e-solver-atlas.29a184b65c":"graph","loom-e-solver-atlas.0155752dfb":"automata","loom-e-solver-atlas.06f9b0facd":"symbolic","loom-e-solver-atlas.6c943beb4d":"synthesis","loom-e-solver-atlas.b5b81a8207":"classification routes the problem into the matching region; the native artefact comes back out","loom-e-solver-atlas.3cb010e8ea":"recognise the problem shape","loom-e-solver-atlas.d9e4b9bd2c":"the status stays exact","loom-e-solver-atlas.problemToken":"problem token","loom-e-solver-atlas.features":"features [ boolean, clauses, no arithmetic ]","loom-e-solver-atlas.target":"Boolean + quantified SAT","loom-e-solver-atlas.region.boolean":"Boolean + quantified SAT","loom-e-solver-atlas.region.maxsat":"MaxSAT + pseudo-Boolean","loom-e-solver-atlas.region.smt":"SMT + optimisation","loom-e-solver-atlas.region.constraints":"Constraint programming","loom-e-solver-atlas.region.logic":"Logic programming","loom-e-solver-atlas.region.proof":"Proof + ontology","loom-e-solver-atlas.region.linear":"Linear programming","loom-e-solver-atlas.region.convex":"Quadratic + convex","loom-e-solver-atlas.region.mixed":"Mixed + nonlinear","loom-e-solver-atlas.region.scheduling":"Scheduling + temporal","loom-e-solver-atlas.region.graph":"Graph + pathfinding","loom-e-solver-atlas.region.automata":"Automata + transducers","loom-e-solver-atlas.region.hybrid":"Timed + hybrid","loom-e-solver-atlas.region.model":"Model checking + Petri","loom-e-solver-atlas.region.games":"Equilibrium + games","loom-e-solver-atlas.region.probability":"Probability + planning","loom-e-solver-atlas.region.program":"Program + protocol","loom-e-solver-atlas.region.rewriting":"Rewriting + equivalence","loom-e-solver-atlas.region.algebra":"Symbolic algebra","loom-e-solver-atlas.region.roots":"Roots + intervals","loom-e-solver-atlas.region.differential":"Differential systems","loom-e-solver-atlas.region.matrix":"Matrices + geometry","loom-e-solver-atlas.region.synthesis":"Synthesis + search","loom-e-solver-atlas.region.evidence":"Solver evidence","loom-e-solver-dispatch.2693a4530e":"Solver selection","loom-e-solver-dispatch.0f4191a584":"scored","loom-e-solver-dispatch.d4bdddb7b5":"real status out","loom-e-solver-dispatch.efc1e4fe46":"three candidate solvers scored by theory, features, and budget","loom-e-solver-dispatch.ee221175e3":"the selected solver returns UNSAT plus core","loom-e-solver-dispatch.8c59c7391b":"a second branch times out to unknown, no fabricated answer","loom-e-solver-dispatch.5ee6c733e1":"native and FFI stay distinct","loom-e-solver-dispatch.98b2fc789a":"unknown remains unknown","loom-e-knowledge-verification.58cafba6b0":"Knowledge verification","loom-e-knowledge-verification.75e4aedce4":"checked","loom-e-knowledge-verification.f68f1db47b":"artefacts stay distinct","loom-e-knowledge-verification.a6e2752634":"a new fact enters the belief base","loom-e-knowledge-verification.cab647a20a":"it creates a conflict","loom-e-knowledge-verification.1c099af4d7":"revision resolves it","loom-e-knowledge-verification.7dd5ee7c71":"a proof checker validates the survivor","loom-e-knowledge-verification.bd2d167b6e":"AION attaches a certificate where the path supports one","loom-e-knowledge-verification.a5314cb102":"proof is not explanation","loom-e-knowledge-verification.c39f20104a":"inference is named as inference","loom-e-knowledge-verification.58dbfe3d65":"fact","loom-e-knowledge-verification.ae214513a0":"conflict","loom-e-knowledge-verification.eae79d6e0c":"revision","loom-e-knowledge-verification.ef64363d23":"proof","loom-e-knowledge-verification.03fbb0ef7b":"belief base","loom-e-knowledge-verification.2ff8dd800e":"incoming","loom-e-knowledge-verification.6cafe3553c":"retracted","loom-e-knowledge-verification.1e61fe1e47":"kept","loom-e-knowledge-verification.0fef23ceaa":"certificate issued","loom-e-expert-layer.9f4d357043":"Specialist selection","loom-e-expert-layer.d8f7d26b00":"routed","loom-e-expert-layer.0fd3113844":"sparse depth","loom-e-expert-layer.05d801c573":"full catalogue","loom-e-expert-layer.3f31694ec2":"candidate pool","loom-e-expert-layer.742998edaf":"gates","loom-e-expert-layer.6a83b7095f":"active specialists","loom-e-expert-layer.fa650cbb6c":"specialist outputs rejoin solver, constraint, and verifier nodes","loom-e-expert-layer.62a3bbf21d":"domain depth added","loom-e-expert-layer.ec1414ee5f":"the graph stays the hero","loom-e-expert-layer.e43e60346e":"solver","loom-e-expert-layer.9f454fd4e8":"constraint","loom-e-expert-layer.e15436a2bb":"verifier","loom-e-expert-layer.190982dc9f":"4 to 8 active","loom-e-expert-layer.6a7b34bf54":"528 domain specialists","loom-e-expert-layer.e7c89fbad0":"passed gate","loom-e-expert-layer.a466a9bf1d":"below threshold","loom-e-expert-layer.91d9c1df24":"foundational reasoning","loom-e-expert-layer.709d76ff08":"mathematics and science","loom-e-expert-layer.ac3c4fb2f9":"code and systems","loom-e-expert-layer.0e4b1b4092":"language and multimodal","loom-e-expert-layer.b0127f7ba4":"verification and transfer","loom-e-expert-layer.93edeb659c":"Activation map","loom-e-expert-layer.cb5ae4a831":"this query","loom-e-expert-layer.b780d252a1":"specialist","loom-e-expert-layer.9120580e94":"domain","loom-e-expert-layer.3c5b492776":"gate","loom-e-expert-layer.aa4a5f8125":"state","loom-e-expert-layer.9662a505ed":"reasoning","loom-e-expert-layer.f36e82fc55":"maths, science","loom-e-expert-layer.909d152604":"code, systems","loom-e-expert-layer.e11523c5ff":"language","loom-e-expert-layer.65279e1ecf":"verification","loom-e-expert-layer.34420cf5d6":"transfer","loom-e-expert-layer.9d4e1e23bd":"pass","loom-e-expert-layer.2bb6b986c5":"active","loom-e-expert-layer.2c03439596":"held","loom-e-expert-layer.4797abf913":"Four to eight wake per query, the rest stay dormant","loom-e-multilevel-routing.4d9e5ce169":"Routing levels","loom-e-multilevel-routing.a409e01673":"walked","loom-e-multilevel-routing.4c0e4fbef5":"a decision at each","loom-e-multilevel-routing.76c5f1172b":"task to graph","loom-e-multilevel-routing.311c4ca149":"modality to perception","loom-e-multilevel-routing.e546594cb3":"memory to tier","loom-e-multilevel-routing.6b10f835cc":"search to ranking","loom-e-multilevel-routing.7fbb37ec35":"structure to reasoner","loom-e-multilevel-routing.e003593634":"theory to solver","loom-e-multilevel-routing.63c2ccea07":"domain to specialist","loom-e-multilevel-routing.5e26446c6c":"a trace cursor prints the decision record at each level","loom-e-multilevel-routing.2744998d5a":"candidate set and scores","loom-e-multilevel-routing.625e7a35ae":"threshold, version, selected path","loom-e-multilevel-routing.3a1d9722ef":"decision record","loom-e-multilevel-routing.47544b43ae":"walk the path","loom-e-multilevel-routing.835f3b50e3":"selected","loom-e-multilevel-routing.100eedb02a":"candidates","loom-e-multilevel-routing.75ebcb361c":"score","loom-e-merge-semantics.ebbbd77e5f":"Merge semantics","loom-e-merge-semantics.bda4b1c3ce":"typed","loom-e-merge-semantics.31d15249ca":"different operators","loom-e-merge-semantics.67ad5a07a2":"union","loom-e-merge-semantics.9f0d66272a":"rank","loom-e-merge-semantics.9373a0d79c":"quorum","loom-e-merge-semantics.94a73074d2":"arbitration","loom-e-merge-semantics.b6555ab93b":"dominance","loom-e-merge-semantics.6309bf427e":"averaging SAT with model confidence is rejected","loom-e-merge-semantics.c69a805742":"merge is part of the contract","loom-e-merge-semantics.f17c79c910":"not a final coherence prompt","loom-e-merge-semantics.66b0c73b3b":"inputs","loom-e-merge-semantics.1029d67644":"output","loom-e-merge-semantics.d20602f525":"SAT","loom-e-merge-semantics.b5e7b5c3b7":"model confidence","loom-e-merge-semantics.6eba83c8d9":"types do not unify","loom-e-binary-training.4ace5ad358":"One-bit training","loom-e-binary-training.6f5b01580c":"native","loom-e-binary-training.975663e865":"no post-hoc quantiser","loom-e-binary-training.e8eefd6c94":"binary data representation","loom-e-binary-training.c6549324d0":"integer updates","loom-e-binary-training.cd80dd3a05":"one-bit artefact","loom-e-binary-training.132593a905":"no post-hoc quantiser stage appears in the path","loom-e-binary-training.dd0a9d2d07":"the target participates in training","loom-e-binary-training.c447079eb7":"families train independently","loom-e-binary-training.3207357a0b":"training path","loom-e-binary-training.0f89ac8072":"post-hoc quantiser","loom-e-binary-training.5f4561cd1e":"not in the path","loom-e-binary-training.a17c9aaa61":"data","loom-e-binary-training.0a25ba5991":"update","loom-e-binary-training.2fa9505626":"artefact","loom-e-binary-training.6b757023bc":"Export run","loom-e-binary-training.902df62e99":"One-bit is native, not quantised","loom-e-binary-training.d101bd986d":"no post-hoc quantiser stage","loom-e-binary-training.53c70625c1":"bipolar representation","loom-e-binary-training.01353ee1a4":"integer accumulators","loom-e-version-manifest.b99a24fabc":"Weave manifest","loom-e-version-manifest.33469d3faa":"pinned","loom-e-version-manifest.eda94fe216":"hashes per node","loom-e-version-manifest.03614a48ee":"perception v2","loom-e-version-manifest.4925e036fe":"memory snapshot","loom-e-version-manifest.6dcb07ea91":"ranker v3","loom-e-version-manifest.faaa08364f":"solver build","loom-e-version-manifest.54c0e7df2c":"specialist set","loom-e-version-manifest.59c8f8421b":"policy bundle","loom-e-version-manifest.472429e225":"a later weave swaps one ranker version, the diff isolated","loom-e-version-manifest.da9ff7ffff":"every boundary versioned","loom-e-version-manifest.4459d33d1a":"upgrades are explicit events","loom-e-version-manifest.f8e966d1e2":"node","loom-e-version-manifest.c692273deb":"version","loom-e-version-manifest.2346ad27d7":"hash","loom-e-version-manifest.51de2b835b":"before","loom-e-version-manifest.405906c9d5":"after","loom-e-version-manifest.75a0ee1ba9":"diff","loom-e-replay.18046fa3f7":"Deterministic replay","loom-e-replay.547ff52d6c":"controlled","loom-e-replay.064ca523b5":"node by node","loom-e-replay.ec95abc24a":"captured replay aligns every node","loom-e-replay.bca7129fa7":"live replay diverges at one external evidence window","loom-e-replay.7751524ce8":"the diff points exactly there","loom-e-replay.6042daeaa8":"graph, routes, state, versions","loom-e-replay.ffff1a9f48":"external evidence can change","loom-e-replay.d73ef92426":"original","loom-e-replay.341d953e69":"replay","loom-e-replay.ef5c844eab":"match","loom-e-replay.3936744cd9":"diverges here","loom-e-replay.a001cafb57":"perception","loom-e-replay.81c5d49be9":"memory","loom-e-replay.3559d7accf":"search","loom-e-replay.e43e60346e":"solver","loom-e-replay.86a4261d84":"verify","loom-e-replay.f3c62de455":"express","loom-e-trace.2645882678":"Weave trace","loom-e-trace.51e3d9c9d8":"observable","loom-e-trace.3b010f75a5":"no private channel","loom-e-trace.c8e85b863d":"nodes activated","loom-e-trace.26d529a156":"typed inputs and outputs","loom-e-trace.408320d8d4":"memories and evidence","loom-e-trace.af797cec7c":"solver calls and statuses","loom-e-trace.d1bf2aca7b":"verification results","loom-e-trace.9ea57526a6":"expression path","loom-e-trace.c761b1e576":"a private-thought channel is absent by design","loom-e-trace.0719b3e8db":"artefacts with IDs","loom-e-trace.5cf4edbb86":"never generated internal prose","loom-e-trace.2494ddb6a2":"run task","loom-e-trace.87ea5dfc8b":"id","loom-e-trace.bda4b1c3ce":"typed","loom-e-loom-versus-nexus.981c2e4549":"Loom and Nexus","loom-e-loom-versus-nexus.94d5cab6f5":"split","loom-e-loom-versus-nexus.cc72ddd224":"cognition versus organisation","loom-e-loom-versus-nexus.0fed3b96cc":"Loom","loom-e-loom-versus-nexus.9a8e265faf":"one woven cognitive graph, one result","loom-e-loom-versus-nexus.485a04fef4":"Nexus","loom-e-loom-versus-nexus.8fffdda20b":"agents with identity, authority, and tasks","loom-e-loom-versus-nexus.4849046ecd":"the shared faculties serve different responsibilities","loom-e-loom-versus-nexus.9d69d7ccb5":"a thread inside a graph","loom-e-loom-versus-nexus.c5960a7ef5":"a capability inside an agent","loom-e-loom-versus-nexus.484a1b4f3a":"one result","loom-e-loom-versus-nexus.c6855959e9":"shared foundation","loom-e-loom-versus-nexus.59e70a7bde":"shared faculties","loom-e-loom-versus-nexus.60c58ad2ef":"communication","loom-e-loom-versus-nexus.52934c29ca":"delegate","loom-e-loom-versus-nexus.cdb02571bf":"lead hands work down to agents","loom-e-loom-versus-nexus.a27297bde9":"report","loom-e-loom-versus-nexus.ee6a59c146":"agents return results up to the lead","loom-e-loom-versus-nexus.326ea1f25c":"message bus","loom-e-loom-versus-nexus.8f68426ad6":"agents coordinate over one channel","loom-e-loom-versus-nexus.lifecycle":"lifecycle","loom-e-loom-versus-nexus.role.authority":"authority","loom-e-loom-versus-nexus.role.delegate":"delegate","loom-e-loom-versus-nexus.role.task":"task","loom-e-loom-versus-nexus.life.spawn":"spawn","loom-e-loom-versus-nexus.life.act":"act","loom-e-loom-versus-nexus.life.report":"report","loom-e-loom-versus-nexus.life.retire":"retire","loom-e-loom-versus-nexus.faculty.perception":"perception","loom-e-loom-versus-nexus.faculty.memory":"memory","loom-e-loom-versus-nexus.faculty.models":"micro models","loom-e-loom-versus-nexus.faculty.solvers":"solvers","loom-e-loom-versus-nexus.faculty.trace":"trace","loom-e-core-underneath.59478a2ee1":"Lowering to Core","loom-e-core-underneath.a8aacba1d9":"executed","loom-e-core-underneath.7c1e084a7b":"operations underneath","loom-e-core-underneath.5306b6c365":"the Loom graph lowers into Core operation paths","loom-e-core-underneath.ff221d4752":"CPU","loom-e-core-underneath.a6a6318544":"GPU","loom-e-core-underneath.ef98362b8a":"browser","loom-e-core-underneath.dd3b67f3df":"deployment backends","loom-e-core-underneath.23d2571334":"selected deployments place Core on Kera","loom-e-core-underneath.b58bbcc943":"Loom chooses cognition","loom-e-core-underneath.48c5535667":"Core executes operations","loom-e-core-underneath.2373aeb187":"Loom graph","loom-e-core-underneath.c6875e7a5f":"Core operation paths","loom-e-core-underneath.3d36387a86":"backends","loom-e-core-underneath.a7e44d0843":"Backend targets","loom-e-core-underneath.8cab0006d9":"same graph","loom-e-core-underneath.d3a9d61785":"One Core, lowered to the metal underneath","loom-e-deploy-graph.b062c21f1e":"Deployment graph","loom-e-deploy-graph.85a01b85d1":"deployed","loom-e-deploy-graph.3d80aa6b05":"sockets by policy","loom-e-deploy-graph.6881023a05":"managed Mesh","loom-e-deploy-graph.2f0d4f153f":"on-premises","loom-e-deploy-graph.a0f2c69a18":"edge","loom-e-deploy-graph.c575f3e383":"distributed","loom-e-deploy-graph.53354e1327":"air-gapped","loom-e-deploy-graph.3aad267ac0":"external sockets enabled or sealed according to policy","loom-e-deploy-graph.35538c11e6":"the contract travels","loom-e-deploy-graph.330f8d474d":"the meaning does not change","loom-e-deploy-graph.d140aa6e7e":"one manifest","loom-e-deploy-graph.fd871c1d54":"sockets open","loom-e-deploy-graph.9c54b3914f":"sockets sealed","loom-e-deploy-graph.34df5da532":"Posture policy","loom-e-deploy-graph.2fbc9f0c9f":"posture","loom-e-deploy-graph.a17c9aaa61":"data","loom-e-deploy-graph.3b9c7a216a":"external sockets","loom-e-deploy-graph.41de761f53":"sharing","loom-e-deploy-graph.980c5d4a9d":"EU region","loom-e-deploy-graph.ed525355b3":"your hardware","loom-e-deploy-graph.dbebf5a146":"device-local","loom-e-deploy-graph.8c7667b3c2":"federated","loom-e-deploy-graph.1b84995dfa":"isolated","loom-e-deploy-graph.9f00fad98b":"policy","loom-e-deploy-graph.71f8e7976e":"none","loom-e-deploy-graph.5fc7e38bff":"open","loom-e-deploy-graph.5d2b90a4c4":"sealed","loom-e-deploy-graph.c2a3c4aaa7":"The manifest is identical, the policy differs","loom-e-close.9e3c1d591d":"Claim evidence","loom-e-close.100ec4462d":"evidence","loom-e-close.4beed4ad6f":"the engineer band","loom-e-close.0562209a4c":"multimodal perception","loom-e-close.e423af0143":"four-tier memory","loom-e-close.7636ae6426":"native micro models","loom-e-close.7d033c38c5":"fully bipolar transformers","loom-e-close.1c792f01cf":"reasoning modes","loom-e-close.bae16fec2d":"solver fabric","loom-e-close.286987b4f9":"528-domain-specialist sparse layer","loom-e-close.0183f47f43":"deterministic trace","loom-e-close.00add3b375":"distinct responsibilities","loom-e-close.0ea769a3d4":"only the graph the task needs","loom-c-different-thinking.bf1bb1dd74":"Question patterns","loom-c-different-thinking.87b6ebf9c6":"woven","loom-c-different-thinking.ef3e0a9294":"behind one surface","loom-c-different-thinking.476b6ad81d":"label this","loom-c-different-thinking.c2a8baeb65":"a label","loom-c-different-thinking.f30ede5724":"read this letter","loom-c-different-thinking.db84d0becc":"an exact source span","loom-c-different-thinking.870d61030a":"plan these appointments","loom-c-different-thinking.02ab92a8a1":"a solved plan","loom-c-different-thinking.56972bc22d":"why did this happen","loom-c-different-thinking.311f343043":"an answer with evidence","loom-c-different-thinking.f346272063":"all four pass through one Fabric conversation","loom-c-different-thinking.75ef241270":"you ask once","loom-c-different-thinking.b82651da31":"Loom assembles the thinking","loom-c-different-thinking.16f9012579":"you ask","loom-c-different-thinking.7c448a3a50":"it returns","loom-c-different-thinking.51621eea1d":"one Fabric surface","loom-c-different-thinking.6684db0c8a":"a quick sorter","loom-c-different-thinking.138ea271b9":"an exact finder","loom-c-different-thinking.f09d25fb50":"a real planner","loom-c-different-thinking.35c05d8471":"a cause tracer","loom-c-different-jobs.4809cfaec2":"Task fit","loom-c-different-jobs.7595d8220a":"mismatched","loom-c-different-jobs.bfd71abb80":"one size fits none","loom-c-different-jobs.ab80bfbe5d":"find a date","loom-c-different-jobs.1451e514c7":"understand a picture","loom-c-different-jobs.7d2de72b61":"search sources","loom-c-different-jobs.3fd2a5d732":"plan a schedule","loom-c-different-jobs.bc8e58ff35":"explain a cause","loom-c-different-jobs.49be9676d2":"write it up","loom-c-different-jobs.1ba6d0e76d":"a single generic chat bubble does not fit all six","loom-c-different-jobs.968f75aa40":"different jobs","loom-c-different-jobs.fc34019c70":"different thinking","loom-c-different-jobs.07b5e618cb":"date finder","loom-c-different-jobs.2ea872495a":"image reader","loom-c-different-jobs.3187278eec":"source search","loom-c-different-jobs.bca17606f6":"schedule planner","loom-c-different-jobs.8199d975d8":"cause tracer","loom-c-different-jobs.9dade87cd5":"the writer","loom-c-different-jobs.ba120d9714":"the matching tool","loom-c-different-jobs.2ffc270a42":"fits none of the six","loom-c-different-jobs.8fbb4cbd80":"a date on your calendar","loom-c-different-jobs.f6e817e946":"what the picture shows","loom-c-different-jobs.8dc5b301dc":"where it came from","loom-c-different-jobs.3badee49ca":"a plan you can keep","loom-c-different-jobs.a5d330e18f":"the reason why","loom-c-different-jobs.db1c4cfa40":"a draft you can send","loom-c-different-jobs.a78d12d854":"you get","loom-c-small-stays-small.ef1a6cd540":"A small task","loom-c-small-stays-small.b55e22fe78":"exact","loom-c-small-stays-small.fb4edc3d93":"small job, small weave","loom-c-small-stays-small.873e7e4a90":"the letter","loom-c-small-stays-small.090c17a586":"the highlighted date span","loom-c-small-stays-small.2723568800":"its source location","loom-c-small-stays-small.60c6fa6f09":"the answer","loom-c-small-stays-small.678e237aee":"everything else stays dim","loom-c-small-stays-small.a32e1ac65a":"a bounded job","loom-c-small-stays-small.7814627b99":"bounded intelligence","loom-c-small-stays-small.154767cadf":"Renewal notice","loom-c-small-stays-small.c8e28d9d58":"please reply by 14 March","loom-c-small-stays-small.f4c83d33a2":"page 1, line 8","loom-c-small-stays-small.50b2641094":"deadline 14 March","loom-c-small-stays-small.1d49cd053f":"the whole weave","loom-c-small-stays-small.18cc852c9a":"letter","loom-c-small-stays-small.f90cee9b5b":"date span","loom-c-small-stays-small.828d338a9b":"source","loom-c-small-stays-small.25dc282b5a":"answer","loom-c-see-input.93703c49d1":"What it sees","loom-c-see-input.229670d592":"seen","loom-c-see-input.9eeca359cf":"connected to the source","loom-c-see-input.3ded88c285":"a photo to objects and text","loom-c-see-input.3dc1296ea7":"a table to fields and anomalies","loom-c-see-input.1065fb57c9":"a recording to speech and events","loom-c-see-input.5c3dab1bf3":"a document to layout and dates","loom-c-see-input.cf3fd2b9a9":"the answer stays connected to what was seen","loom-c-see-input.833d8ed78a":"more than typed text","loom-c-see-input.108185cc5e":"observations you can trace","loom-c-see-input.eeb35d331b":"photo","loom-c-see-input.c3ee137d4f":"table","loom-c-see-input.5d7c6eae5b":"recording","loom-c-see-input.4f8278c89a":"document","loom-c-see-input.7c02c7e8a0":"observation","loom-c-see-input.921f840c55":"from pixels","loom-c-see-input.32367c8fe9":"from cells","loom-c-see-input.dd2f41dd31":"from audio","loom-c-see-input.736c69d187":"from layout","loom-c-see-input.f0f7f39ec7":"objects","loom-c-see-input.372ea08cab":"text","loom-c-see-input.70f132fe13":"fields","loom-c-see-input.bf162a8546":"anomalies","loom-c-see-input.f3690b9c34":"speech","loom-c-see-input.82d50d9042":"events","loom-c-see-input.207b877e13":"layout","loom-c-see-input.c09a2565eb":"dates","loom-c-memory.e076dbb815":"Helpful memory","loom-c-memory.6543f14e1a":"scoped","loom-c-memory.a05429d4f0":"only what each part needs","loom-c-memory.1ff9c17179":"current task","loom-c-memory.1c810735bb":"earlier cases","loom-c-memory.bb362935c7":"your facts","loom-c-memory.808d508a7b":"procedures","loom-c-memory.3ea9a63c30":"an unrelated personal conversation stays closed","loom-c-memory.a6bc69af79":"the right memory","loom-c-memory.d01dda6010":"to the right helper","loom-c-memory.0377815e0c":"today's question","loom-c-memory.4ec2fb9633":"current step","loom-c-memory.c2a453f821":"a similar case","loom-c-memory.baa81c4b54":"last outcome","loom-c-memory.df0e42641f":"your address","loom-c-memory.f880b20c04":"your preferences","loom-c-memory.4fc479b498":"how to book","loom-c-memory.e0a51ffa78":"how to file","loom-c-memory.35c3c425e1":"personal chat","loom-c-memory.8b06e41a56":"who borrows what","loom-c-memory.b99dc242d0":"planner takes procedures","loom-c-memory.4234e337b8":"finder takes your facts","loom-c-memory.cf9bda74d6":"a task helper takes the current task","loom-c-memory.8b0e3b330c":"a case helper takes earlier cases","loom-c-search-helpers.15ca04efa6":"Search helpers","loom-c-search-helpers.e3eaa6788a":"narrowed","loom-c-search-helpers.026071878a":"several small helpers","loom-c-search-helpers.ded8dae578":"find","loom-c-search-helpers.b64ca250f4":"compare","loom-c-search-helpers.cce55e4309":"order","loom-c-search-helpers.d56d985300":"check","loom-c-search-helpers.956ac95c79":"many source cards narrow to three, then one evidence set","loom-c-search-helpers.f8038fc2b4":"not the first similar paragraph","loom-c-search-helpers.959ad5b37a":"checked support","loom-c-search-helpers.3d1ff9e810":"many sources","loom-c-search-helpers.07fa1285d5":"three candidates","loom-c-search-helpers.4a2dd82f9a":"one evidence set","loom-c-search-helpers.fa577401eb":"matched passage","loom-c-search-helpers.101d133dea":"source named","loom-c-search-helpers.cf85fb3c5e":"small helpers","loom-c-exact-solver.70d9fd35db":"Schedules and rules","loom-c-exact-solver.a0ac741932":"solved","loom-c-exact-solver.b0f16e4e35":"every condition checked","loom-c-exact-solver.89636b631c":"mornings only","loom-c-exact-solver.c1f8afee72":"no overlap","loom-c-exact-solver.75bdc7f79e":"include travel","loom-c-exact-solver.38ffba45b8":"hard deadline","loom-c-exact-solver.203ea02c2f":"a valid schedule exits with every condition ticked","loom-c-exact-solver.59a6d99d5e":"an impossible case exits as no valid schedule, with the conflict named","loom-c-exact-solver.408ac6d7aa":"a chatbot suggests","loom-c-exact-solver.87f270cad3":"a checker works it out","loom-c-exact-solver.88e546d80c":"appointments","loom-c-exact-solver.caa155adf8":"rules","loom-c-exact-solver.b46ca1aac9":"dentist, Tue","loom-c-exact-solver.e6fb82af1f":"delivery, Tue","loom-c-exact-solver.ea38beb9e6":"school run, Wed","loom-c-exact-solver.840d11d7a8":"exact checker","loom-c-exact-solver.0dae46b38d":"Tue 09:00 dentist","loom-c-exact-solver.d43f20ae7f":"Tue 11:30 delivery","loom-c-exact-solver.9f6c261175":"Wed 08:15 school run","loom-c-exact-solver.8856bae4cb":"mornings only clashes with the deadline","loom-c-exact-solver.a95df46f3f":"valid schedule","loom-c-exact-solver.56da865aeb":"no valid schedule","loom-c-exact-solver.5f046455f9":"workable case","loom-c-exact-solver.9fd7ae0c87":"impossible case","loom-c-reasoning-forms.da2b9a6817":"Ways to reason","loom-c-reasoning-forms.bd902042fa":"combined","loom-c-reasoning-forms.6f2bf1100c":"a checked conclusion","loom-c-reasoning-forms.caa155adf8":"rules","loom-c-reasoning-forms.6ea5fcbb80":"causes","loom-c-reasoning-forms.304ea0fbae":"what if","loom-c-reasoning-forms.7462f61439":"the lanes rejoin at a checked conclusion","loom-c-reasoning-forms.2d2f309676":"more than one way to think","loom-c-reasoning-forms.3bd4e815e9":"not one fixed style","loom-c-reasoning-forms.c87cbf65d4":"why did the bill go up","loom-c-reasoning-forms.3f5b2d58fb":"the tariff changed","loom-c-reasoning-forms.d5d687b7b3":"usage rose in winter","loom-c-reasoning-forms.56592d856b":"a fixed plan would cap it","loom-c-reasoning-forms.2b74071571":"higher tariff plus winter usage, a capped plan helps","loom-c-reasoning-forms.75e4aedce4":"checked","loom-c-experts-speak.20f535ce1e":"Specialists at work","loom-c-experts-speak.eb4c0d1553":"selective","loom-c-experts-speak.15ec7fac52":"528 available","loom-c-experts-speak.f89739323b":"a large quiet room of specialist lights","loom-c-experts-speak.0481c41194":"only a few walk into the active weave","loom-c-experts-speak.2f4e9320b9":"the answer shows who joined","loom-c-experts-speak.66c161f9b1":"the relevant few speak","loom-c-experts-speak.f999496c1d":"the rest stay quiet","loom-c-experts-speak.ae6c512c31":"528 domain specialists, quiet","loom-c-experts-speak.0f298ac6b4":"active weave","loom-c-experts-speak.135ac9d514":"who joined this answer","loom-c-experts-speak.1a1dc191fd":"tax rules specialist","loom-c-experts-speak.5486757214":"calendar specialist","loom-c-experts-speak.19c894eada":"local law specialist","loom-c-experts-speak.3368b4e8ef":"checked the tax rules","loom-c-experts-speak.2bf52014bc":"lined up the dates","loom-c-experts-speak.b443d645f4":"read the local rules","loom-c-check-before.5e9df5afcf":"Answer check","loom-c-check-before.75e4aedce4":"checked","loom-c-check-before.ceffeeb374":"before it speaks","loom-c-check-before.bbb2b7b14f":"a certificate","loom-c-check-before.550fcf3e60":"an evidence trace","loom-c-check-before.190b56e37b":"a warning","loom-c-check-before.e97cb97b00":"a refusal","loom-c-check-before.f5e8f44476":"each card passes the check gate","loom-c-check-before.2ff9f5104d":"fluent is not supported","loom-c-check-before.1764ccc41c":"the check decides","loom-c-check-before.038d6815f3":"check gate","loom-c-check-before.6882c5d243":"result in","loom-c-check-before.6ef94c9c55":"verdict out","loom-c-check-before.02ab92a8a1":"a solved plan","loom-c-check-before.fea6c78b90":"a sourced answer","loom-c-check-before.642edecbc7":"a shaky guess","loom-c-check-before.a13bb82701":"no support found","loom-c-weave-receipt.ce1fe4fab8":"Your answer record","loom-c-weave-receipt.1d28e511b1":"readable","loom-c-weave-receipt.fcdcedcf76":"a plain-language record","loom-c-weave-receipt.29596a43a2":"saw","loom-c-weave-receipt.0aaea540f0":"remembered","loom-c-weave-receipt.200a4eeceb":"searched","loom-c-weave-receipt.1ce4362ede":"reasoned","loom-c-weave-receipt.a0ac741932":"solved","loom-c-weave-receipt.75e4aedce4":"checked","loom-c-weave-receipt.c5d55c88a5":"no private thoughts, only the useful record","loom-c-weave-receipt.dc6d5d22f8":"what helped","loom-c-weave-receipt.61b3a69d7e":"in plain language","loom-c-weave-receipt.86a8c3decd":"your answer","loom-c-weave-receipt.f0b09a8a13":"weave 0x4C7A","loom-c-weave-receipt.3bf7579a4c":"read your photo and letter","loom-c-weave-receipt.164e5ec9cf":"used your saved address","loom-c-weave-receipt.9713dd16a0":"found two matching sources","loom-c-weave-receipt.57a2224039":"weighed rules and causes","loom-c-weave-receipt.3d9f558fe4":"checked the schedule fits","loom-c-weave-receipt.71363f7c7b":"verified before answering","loom-c-weave-receipt.71ab933860":"step by step","loom-c-weave-receipt.e4d6ad3ba7":"what you got back","loom-c-weave-receipt.28468c53b7":"It came back in plain words you can read, next to the record of how it was reached.","loom-c-weave-receipt.dfcb43ff4e":"outcome","loom-c-weave-receipt.76e806bac8":"answered","loom-c-weave-receipt.883681c640":"the schedule fits","loom-c-weave-receipt.a503bc1da2":"refused","loom-c-weave-receipt.8063d28eba":"if a check does not pass, it says so instead of guessing","loom-c-can-stop.783f8fb9bd":"Honest refusal","loom-c-can-stop.5acf411124":"stopped","loom-c-can-stop.3ae9fa0492":"honest about limits","loom-c-can-stop.891d6b1b65":"the missing evidence, named","loom-c-can-stop.27830ad76a":"a suggested next step","loom-c-can-stop.67bb115124":"no confident guess","loom-c-can-stop.fa1263acc2":"an honest stop","loom-c-can-stop.7de0507232":"the weave runs, then stops","loom-c-can-stop.8a4a12bdb1":"saw the question","loom-c-can-stop.ab90a9aa77":"searched sources","loom-c-can-stop.63b7a6193e":"found nothing solid","loom-c-can-stop.1b480158e1":"stop","loom-c-can-stop.7a66117546":"honest refusal","loom-c-can-stop.547e1539c8":"I cannot confirm this yet","loom-c-can-stop.952164feb0":"no reliable source for this claim","loom-c-can-stop.8fd95edfee":"share the original document","loom-c-can-stop.5a013c4950":"missing","loom-c-can-stop.f6b4702b64":"next step","loom-c-runs-close.9832bc1913":"Where it runs","loom-c-runs-close.bc109fb076":"portable","loom-c-runs-close.fdee070296":"scale by task","loom-c-runs-close.8710ce9bad":"a laptop","loom-c-runs-close.eb2cc3e318":"a local server","loom-c-runs-close.76bce656e1":"an edge device","loom-c-runs-close.58eab4647c":"an EU data centre","loom-c-runs-close.665c387b88":"the scale changes with the task, not the idea","loom-c-runs-close.dea6d371cf":"close where supported","loom-c-runs-close.596b81715f":"not only distant clusters","loom-c-runs-close.5818b4d70d":"same weave","loom-c-runs-close.c5e6b977eb":"small tasks","loom-c-runs-close.2d0ac3688a":"team tasks","loom-c-runs-close.44c3bd4cbd":"on-site tasks","loom-c-runs-close.4355252c5d":"heavy tasks","loom-c-runs-close.60020d1b19":"scale","loom-c-runs-close.ada9aca4e9":"The same weave, wherever it runs","loom-c-runs-close.31a1d58418":"only the size of the job changes","loom-c-runs-close.93f90654c5":"one build, four scales","loom-c-runs-close.b5e6f35adc":"the same weave file, only the scale differs","loom-c-behind-fabric.dfd8cec69b":"Fabric beneath","loom-c-behind-fabric.489074d281":"revealed","loom-c-behind-fabric.b1b6958f56":"one surface, many weaves","loom-c-behind-fabric.0b3feec29e":"the Fabric conversation","loom-c-behind-fabric.65956ad98b":"a cutaway reveals the active weave","loom-c-behind-fabric.deeb40ea08":"earlier questions show different archived weaves","loom-c-behind-fabric.67dc9d80b2":"Fabric is the surface","loom-c-behind-fabric.c3bd9cac47":"Loom is the fabric underneath","loom-c-behind-fabric.a010de5c61":"Fabric","loom-c-behind-fabric.0fed3b96cc":"Loom","loom-c-behind-fabric.a98346e192":"why did my bill rise?","loom-c-behind-fabric.e45e196fea":"here is the answer, with evidence","loom-c-behind-fabric.2fe9a344e0":"the active weave underneath","loom-c-behind-fabric.1e1dfb3d85":"earlier questions","loom-c-behind-fabric.319a2b2ab0":"the letter deadline","loom-c-behind-fabric.3f0284fe10":"the schedule","loom-c-behind-fabric.c7a1707fa3":"the photo","loom-c-behind-fabric.29596a43a2":"saw","loom-c-behind-fabric.200a4eeceb":"searched","loom-c-behind-fabric.1ce4362ede":"reasoned","loom-c-behind-fabric.a0ac741932":"solved","loom-c-behind-fabric.75e4aedce4":"checked","loom-c-behind-fabric.34d0044a39":"answered with a date","loom-c-behind-fabric.878598d6dc":"answered with a plan","loom-c-behind-fabric.3e4a63175f":"answered with what it showed","loom-c-behind-fabric.140f86aae5":"input","loom-c-behind-fabric.bce16dd026":"sources","loom-c-behind-fabric.caa155adf8":"rules","loom-c-behind-fabric.25dc282b5a":"answer","loom-c-behind-fabric.86a4261d84":"verify","loom-c-behind-fabric.100a7b7a72":"answered with evidence","loom-c-close.9ae8d97241":"Answer grounds","loom-c-close.100ec4462d":"evidence","loom-c-close.f123a39d1f":"the consumer band","loom-c-close.527dff6c22":"exact spans","loom-c-close.394f818754":"purposeful memory","loom-c-close.c20f42b9a9":"staged search","loom-c-close.99606cf53b":"real checkers","loom-c-close.6177bcec89":"multiple reasoning forms","loom-c-close.2cdbe552c4":"only the relevant specialists","loom-c-close.6fa69ef6a9":"visible checks","loom-c-close.8b3139fd33":"the question stays simple","loom-c-close.9a232ae15e":"the weave fits the task","loom-c-close.3e19c3f01f":"ask once","loom-c-close.7eef4b2dbb":"highlighted in the source","loom-c-close.ea12564070":"only what is needed","loom-c-close.4120aaa0b8":"many sources to one","loom-c-close.9a33f31ac5":"checked, not guessed","loom-c-close.c3cf7b9b48":"rules, causes, what-if","loom-c-close.2c6d54039b":"the relevant few","loom-c-close.ceffeeb374":"before it speaks","loom-c-close.4db385caca":"every answer stands on","loom-c-sides.four_kl":"Four questions","loom-c-sides.four_ks":"four patterns","loom-c-sides.four_1n":"label this","loom-c-sides.four_1b":"a quick sorter wakes","loom-c-sides.four_2n":"read this letter","loom-c-sides.four_2b":"a reader finds the exact line","loom-c-sides.four_3n":"plan these times","loom-c-sides.four_3b":"a real planner checks the rules","loom-c-sides.four_4n":"why did this happen","loom-c-sides.four_4b":"a cause tracer gathers evidence","loom-c-sides.four_fl":"you ask once","loom-c-sides.four_fr":"the pattern forms","loom-c-sides.jobs_kl":"Six jobs","loom-c-sides.jobs_ks":"one method each","loom-c-sides.jobs_1n":"date finder","loom-c-sides.jobs_1s":"the exact line","loom-c-sides.jobs_2n":"image reader","loom-c-sides.jobs_2s":"what is in it","loom-c-sides.jobs_3n":"source search","loom-c-sides.jobs_3s":"find and weigh","loom-c-sides.jobs_4n":"schedule planner","loom-c-sides.jobs_4s":"rules that hold","loom-c-sides.jobs_5n":"cause tracer","loom-c-sides.jobs_5s":"why it happened","loom-c-sides.jobs_6n":"the writer","loom-c-sides.jobs_6s":"says it plainly","loom-c-sides.jobs_fl":"one bubble fits none","loom-c-sides.jobs_fr":"each job, its method","loom-c-sides.small_bt":"wake everything","loom-c-sides.small_bn":"the usual way","loom-c-sides.small_bc1":"everything on","loom-c-sides.small_bc2":"slow","loom-c-sides.small_bc3":"costly","loom-c-sides.small_bc4":"hard to check","loom-c-sides.small_at":"wake the small part","loom-c-sides.small_an":"with Loom","loom-c-sides.small_ac1":"read the letter","loom-c-sides.small_ac2":"find the date","loom-c-sides.small_ac3":"name the source","loom-c-sides.small_ac4":"done","loom-c-sides.small_fl":"a bounded job","loom-c-sides.small_fr":"bounded help","loom-c-sides.seen_kl":"Four inputs","loom-c-sides.seen_ks":"one method","loom-c-sides.seen_1l":"a photo","loom-c-sides.seen_1c":"objects and text","loom-c-sides.seen_1v":"seen","loom-c-sides.seen_2l":"a table","loom-c-sides.seen_2c":"fields and odd values","loom-c-sides.seen_2v":"read","loom-c-sides.seen_3l":"a recording","loom-c-sides.seen_3c":"speech and moments","loom-c-sides.seen_3v":"heard","loom-c-sides.seen_4l":"a document","loom-c-sides.seen_4c":"layout and dates","loom-c-sides.seen_4v":"kept","loom-c-sides.seen_fl":"typed text too","loom-c-sides.seen_fr":"linked to the source","loom-c-sides.mem_kl":"Four drawers","loom-c-sides.mem_ks":"opened by need","loom-c-sides.mem_1n":"the task now","loom-c-sides.mem_1t":"what you are doing today","loom-c-sides.mem_2n":"earlier cases","loom-c-sides.mem_2t":"how it went before","loom-c-sides.mem_3n":"your facts","loom-c-sides.mem_3t":"from your own notes","loom-c-sides.mem_4n":"your routines","loom-c-sides.mem_4t":"steps you repeat","loom-c-sides.mem_fl":"not one long log","loom-c-sides.mem_fr":"only what each part needs","loom-c-sides.search_kl":"Four small helpers","loom-c-sides.search_ks":"one clear set","loom-c-sides.search_1n":"find","loom-c-sides.search_1b":"gather the candidates","loom-c-sides.search_2n":"compare","loom-c-sides.search_2b":"weigh which really fit","loom-c-sides.search_3n":"order","loom-c-sides.search_3b":"put the best first","loom-c-sides.search_4n":"check","loom-c-sides.search_4b":"confirm they support the answer","loom-c-sides.search_fl":"not the first match","loom-c-sides.search_fr":"checked support","loom-c-sides.solve_kl":"Guess or check","loom-c-sides.solve_ks":"when rules matter","loom-c-sides.solve_bt":"a chatbot","loom-c-sides.solve_bs":"suggests","loom-c-sides.solve_b1":"writes a likely plan","loom-c-sides.solve_b2":"may break a rule","loom-c-sides.solve_b3":"sounds sure either way","loom-c-sides.solve_at":"a real checker","loom-c-sides.solve_as":"checks","loom-c-sides.solve_a1":"tests every rule","loom-c-sides.solve_a2":"finds a plan that fits","loom-c-sides.solve_a3":"or says none exists","loom-c-sides.solve_fl":"a plan that fits","loom-c-sides.solve_fr":"explained plainly","loom-c-sides.reason_kl":"Several ways","loom-c-sides.reason_ks":"joined for one answer","loom-c-sides.reason_1n":"rules","loom-c-sides.reason_1b":"follow what must be true","loom-c-sides.reason_1t":"logic","loom-c-sides.reason_2n":"causes","loom-c-sides.reason_2b":"trace why it happened","loom-c-sides.reason_2t":"why","loom-c-sides.reason_3n":"comparison","loom-c-sides.reason_3b":"match an earlier case","loom-c-sides.reason_3t":"like","loom-c-sides.reason_4n":"what-if","loom-c-sides.reason_4b":"test another condition","loom-c-sides.reason_4t":"test","loom-c-sides.reason_fl":"not one fixed style","loom-c-sides.reason_fr":"checked at the end","loom-c-sides.exp_kl":"A quiet room","loom-c-sides.exp_ks":"few speak","loom-c-sides.exp_1v":"many","loom-c-sides.exp_1l":"in the catalogue","loom-c-sides.exp_1c":"ready if needed","loom-c-sides.exp_2v":"a few","loom-c-sides.exp_2l":"join the answer","loom-c-sides.exp_2c":"only the relevant","loom-c-sides.exp_3v":"rest","loom-c-sides.exp_3l":"stay quiet","loom-c-sides.exp_3c":"no extra work","loom-c-sides.exp_4v":"shown","loom-c-sides.exp_4l":"who helped","loom-c-sides.exp_4c":"on the record","loom-c-sides.exp_fl":"the relevant few speak","loom-c-sides.exp_fr":"the rest stay quiet","loom-c-sides.check_kl":"One gate","loom-c-sides.check_ks":"every result","loom-c-sides.check_1n":"a certificate","loom-c-sides.check_1b":"proof is attached","loom-c-sides.check_1m":"pass","loom-c-sides.check_2n":"an evidence trail","loom-c-sides.check_2b":"the sources line up","loom-c-sides.check_2m":"pass","loom-c-sides.check_3n":"a warning","loom-c-sides.check_3b":"support is thin","loom-c-sides.check_3m":"flag","loom-c-sides.check_4n":"a refusal","loom-c-sides.check_4b":"nothing solid found","loom-c-sides.check_4m":"stop","loom-c-sides.check_fl":"fluent is not supported","loom-c-sides.check_fr":"the check decides","loom-c-sides.rec_kl":"What helped","loom-c-sides.rec_ks":"in plain words","loom-c-sides.rec_1n":"saw","loom-c-sides.rec_1b":"your photo and letter","loom-c-sides.rec_2n":"remembered","loom-c-sides.rec_2b":"your saved address","loom-c-sides.rec_3n":"searched","loom-c-sides.rec_3b":"two matching sources","loom-c-sides.rec_4n":"reasoned","loom-c-sides.rec_4b":"rules and causes","loom-c-sides.rec_5n":"solved","loom-c-sides.rec_5b":"the schedule fits","loom-c-sides.rec_6n":"checked","loom-c-sides.rec_6b":"before answering","loom-c-sides.rec_fl":"no hidden thoughts","loom-c-sides.rec_fr":"only the useful record","loom-c-sides.stop_kl":"An honest stop","loom-c-sides.stop_ks":"not a guess","loom-c-sides.stop_1n":"read","loom-c-sides.stop_1b":"took in the question","loom-c-sides.stop_2n":"searched","loom-c-sides.stop_2b":"looked for solid sources","loom-c-sides.stop_3n":"found nothing","loom-c-sides.stop_3b":"no reliable source","loom-c-sides.stop_4n":"stop","loom-c-sides.stop_4b":"says so, names the gap","loom-c-sides.stop_fl":"no confident guess","loom-c-sides.stop_fr":"a next step suggested","loom-c-sides.place_kl":"Same helper","loom-c-sides.place_ks":"four places","loom-c-sides.place_1n":"your own machine","loom-c-sides.place_1b":"small tasks, close by","loom-c-sides.place_1t":"home","loom-c-sides.place_2n":"a local server","loom-c-sides.place_2b":"team tasks, on site","loom-c-sides.place_2t":"site","loom-c-sides.place_3n":"a small device","loom-c-sides.place_3b":"right where the work is","loom-c-sides.place_3t":"edge","loom-c-sides.place_4n":"a European centre","loom-c-sides.place_4b":"heavy tasks, in Europe","loom-c-sides.place_4t":"EU","loom-c-sides.place_fl":"close where it can","loom-c-sides.place_fr":"not only far clusters","loom-c-sides.fab_kl":"One surface","loom-c-sides.fab_ks":"many weaves","loom-c-sides.fab_1n":"Fabric","loom-c-sides.fab_1b":"the workspace you open","loom-c-sides.fab_2n":"your question","loom-c-sides.fab_2b":"stays simple for you","loom-c-sides.fab_3n":"Loom","loom-c-sides.fab_3b":"the careful brain beneath","loom-c-sides.fab_4n":"the weave","loom-c-sides.fab_4b":"only the parts you need","loom-c-sides.fab_5n":"the record","loom-c-sides.fab_5b":"how the answer was built","loom-c-sides.fab_fl":"Fabric is the surface","loom-c-sides.fab_fr":"Loom is underneath","page-breadcrumb.c766e66518":"Breadcrumb","page-toc.7e439c353e":"On this page","product-loom.opening.technical.pill.graph":"TYPED GRAPH","product-loom.opening.technical.pill.trace":"WEAVE TRACE","product-loom.opening.consumer.pill.patterns":"FOUR PATTERNS","product-loom.opening.consumer.pill.checked":"ONE PLACE","product-loom.3d00bf2434":"Exact extraction","product-loom.b64527146f":"Solver output","product-loom.7ea014de7b":"Evidence","product-loom.97b695b9b9":"Expression","product-loom.a001cafb57":"perception","product-loom.3a93bc3e5a":"retrieval","product-loom.9662a505ed":"reasoning","product-loom.65279e1ecf":"verification","product-loom.1f827d01b7":"Routing model","product-loom.039726387c":"Open the system card","product-loom.0760b8560e":"Constraint learning first. A sparse domain-specialist layer narrows each task. Small bipolar language components render the result graph; they do not decide it. PAP traces, bit-identical replay under a fixed seed, and integer and bitwise inference keep the path inspectable and CPU-first.","product-loom.68c2a8d63c":"Read the trace. Replay the seed.","product-loom.90e40d5043":"Get started","product-loom.dfdf5b54e7":"Verify offline, on your tooling","product-loom.8476bf5919":"Proof certificates","product-loom.df84637182":"AION","product-loom.fdb4c0e664":"456 domain specialists compressed","product-loom.4b0f65f8a9":"Catalogue on disk","product-loom.ca2a7d0c42":"~150 GB","product-loom.d40532d3fd":"On fixed seed, any environment","product-loom.1109d62f23":"Replay class","product-loom.39e75dc1ee":"Bit-identical","product-loom.c435da0947":"On every answer, by default","product-loom.f68cece41b":"Trace attached","product-loom.7cbf7cdb77":"Than a generative stack","product-loom.aa50825f45":"Less energy","product-loom.c3ee84e69b":"around 4 to 8 on average","product-loom.e2b250ddf6":"Loom specialists","product-loom.ed2be7d725":"The substrate numbers the systems claim rests on.","product-loom.3822779c1e":"Loom, engineering","product-loom.4dc7cf5a29":"Not promised in prose","product-loom.cc11603ceb":"Asserted in test suite","product-loom.e57eb071f5":"Bit-identical on fixed seed","product-loom.f575b6907e":"Replay fidelity","product-loom.136a5821e5":"Correctly rounded by construction","product-loom.e9ba134322":"Arithmetic error","product-loom.51ab88da6e":"0 ULP","product-loom.8e03b3d796":"Integer and bitwise only","product-loom.87aab57e39":"Float operations on hot path","product-loom.1e76d09e8c":"Four guarantees, by construction","product-loom.2ec10d2ee7":"DETERMINISM PROPERTIES","product-loom.43c4336536":"SORTED ITERATION","product-loom.97c24eb9fb":"SEEDED RNG","product-loom.3e1251eb77":"INTEGER ARITHMETIC","product-loom.a6d248e4dc":"no wall-clock.","product-loom.b05915ec08":"Integer, bitwise,","product-loom.2956ef9ad6":"Determinism","product-loom.7056666c90":"Each cluster, independent","product-loom.78f0740785":"Ten clusters","product-loom.e26e5f8548":"Domain rules live with the specialist that enforces them, not in a shared global set.","product-loom.dbbfda0fb7":"Cluster-scoped constraints","product-loom.ca213d28e3":"New specialists join without retraining the full catalogue. The router scores the new specialist against existing signatures.","product-loom.6b1a966359":"Additive growth","product-loom.57ed73f714":"Each cluster trains on its own corpus. A defect in the finance cluster does not affect the science cluster.","product-loom.f40fe809b2":"Independent training","product-loom.9124208dcf":"Three architectural properties","product-loom.1f80b594b3":"WHY CLUSTERS MATTER","product-loom.dc0507035e":"VERIFICATION","product-loom.83456a5aee":"DOMAIN","product-loom.16de25af88":"CODE","product-loom.b61263161a":"SCIENCE","product-loom.40fee0a0a5":"MATHEMATICS","product-loom.8ffa1146c7":"REASONING","product-loom.301de0f349":"528 independent specialists.","product-loom.4c82a7110d":"Ten domain clusters.","product-loom.5622684f16":"Specialist taxonomy","product-loom.35471b951f":"PAP, HNSW, gate subsets","product-loom.9e650b229f":"Routing cost","product-loom.28e3933547":"528 top-level domain specialists","product-loom.80832cb05a":"Domain clusters","product-loom.0c37dcfa15":"Per specialist, in the catalogue","product-loom.f0ed31f837":"Four fields","product-loom.8930e92c25":"Content-addressed by weights, constraint set, and training corpus snapshot. Loom can expand a constraint set when its own calculations produce a candidate and Spindle researches and validates it; each addition is versioned. Upgrade is a deliberate, pinnable event.","product-loom.2750d32235":"Versioning","product-loom.fb822ce3ff":"Approximately 3 billion domain-specific constraints per specialist. Conflicts surface as typed refusals, not soft warnings.","product-loom.8c601c27ad":"Constraint set","product-loom.9fae5222a3":"Typed graph of Spindle nodes. Answers cite a content-addressed reference.","product-loom.dec1fccaba":"Knowledge","product-loom.e7266e42e6":"Domain corpus, domain evaluation suite. Failed specialists are tagged and not activated.","product-loom.b6fe7f5e79":"Training","product-loom.26a5fcfabb":"Four fields, per specialist","product-loom.e08369dcc2":"WHAT EACH SPECIALIST CARRIES","product-loom.d309b09e26":"PINNABLE","product-loom.79ae623ce9":"CONTENT ADDRESSED","product-loom.f75e48fee2":"CONSTRAINT SET","product-loom.5629594e2c":"DOMAIN TRAINED","product-loom.55661e35d2":"Not a slice of a giant model.","product-loom.5f22785735":"Each specialist owns a domain responsibility.","product-loom.5c223557a3":"Specialist anatomy","product-loom.1b7da91551":"No trust required","product-loom.01c501a76c":"Four steps","product-loom.19fe5533cc":"Fix the seed. Bit-identical results, compared to the last bit, in any environment.","product-loom.b2015e9fc2":"Replay an inference","product-loom.97f3c9f184":"An AION proof certificate attaches to every emit. Verify with your own tooling, not ours.","product-loom.d46b5f550a":"Check the certificate","product-loom.406098aa56":"The suites ship with the repos. Use your own hardware. Candidate pool size is a tunable parameter.","product-loom.2857409009":"Run the benchmarks","product-loom.2edf9f731c":"PAP, BitWeave, and the router are open foundations. Public repos with their own benchmarks.","product-loom.2846d2aa79":"Clone the repos","product-loom.64685798e2":"Four steps, no trust required","product-loom.4279f84451":"HOW TO VERIFY","product-loom.8932f4c7e1":"GATE SUBSETS","product-loom.936ee7e6a1":"HNSW","product-loom.1ac9452818":"PAP","product-loom.9061e12070":"HYPERVECTOR","product-loom.f88021a661":"Sparse activation.","product-loom.241c4c8795":"Sublinear routing.","product-loom.f954691208":"Routing internals","product-loom.17d120d8bb":"Replayable","product-loom.ac178d53c8":"Attached by default","product-loom.fe1d3729fc":"A proof certificate is attached on emit. Check it offline with your own tooling.","product-loom.49ad479e0b":"AION certificate","product-loom.586fc5f937":"None. The path from input to answer is enumerable from the trace.","product-loom.19adfc34cc":"Black-box decisions","product-loom.8f668f163c":"Every decision enumerates the constraints that fired. No implicit dispatch.","product-loom.20690c0420":"Explainable","product-loom.41a1300087":"On every answer","product-loom.a97d00bdb4":"WHAT THE TRACE GUARANTEES","product-loom.4e250b4fde":"REPLAY","product-loom.af5d1b291a":"VERIFY","product-loom.ddf8a55c68":"TRACE","product-loom.b549f98213":"Every answer ships with this. The trace enumerates the constraints that fired, names the active specialists, and carries an AION proof certificate you verify on your own tooling.","product-loom.ea43290899":"see why each specialist fired.","product-loom.39fa44fd6c":"Inspect the constraint trace,","product-loom.bcc9984a00":"Run it","product-loom.a20d5b20bb":"The inference path uses integer and bitwise operations end to end. No floating-point non-associativity, no reduction-order surprises, no vendor-specific math kernels. Ordering of iterable state is sorted, random draws are seeded, time is not consulted. Replay is byte-identical on supported targets.","product-loom.f6e8a58e8b":"The 528 domain specialists span 10 clusters: foundational (core reasoning, mathematics, science and research, code and systems, language) and specialised (multimodal, domain-specific industries, meta-cognitive, verification, transfer and adaptation). A 456-domain-specialist production base sits alongside 72 expansion modules in training through the 57-layer evolutionary pipeline. Clusters are independent. Adding specialists to one does not require retraining the others.","product-loom.06bb18fe3b":"A Loom specialist is trained on its domain, carrying its own knowledge graph, its own constraint set, and its own evaluation suite. Each specialist carries approximately 3 billion constraints in its domain catalogue. Two specialists working on the same token produce different outputs because they are different specialists, not because they were given different slices of the same neural network.","product-loom.b5ab2991b1":"Routing is staged. Coarse filtering via hypervector signatures and approximate neighbour search narrows the catalogue. Per-specialist gate subsets decide pass or fail before heavy constraint evaluation runs. Permuted Agreement Popcount adds structural similarity on top of Hamming. The active frontier never scales linearly with specialist count.","product-loom.8db48970f0":"Bit-identical on seed","product-loom.cee6f7529f":"AION proof certificates","product-loom.03128bed90":"Verification","product-loom.2129d1f5e2":"Hamming and Permuted Agreement","product-loom.3bf1eca830":"Similarity","product-loom.ca51f1d96a":"Integer and bitwise, no float","product-loom.f331153fcb":"Inference path","product-loom.4bf3bf335b":"around 4 to 8 on an average request","product-loom.1e2d4557a8":"Active set","product-loom.6e2a79951c":"Sublinear, O(log N)","product-loom.7d15dd1bec":"Routing","product-loom.ed92fdd68a":"Constraint-learning specialist layer","product-loom.b040b4179b":"Architecture","product-loom.4143d04801":"Engineering","product-loom.bc968f6d59":"Signal surface","product-loom.038a161d16":"Read the routing model","product-loom.038edb91b3":"Loom is a sparse cognitive graph built on binary constraint learning rather than a single weight tensor. A layered router selects a small set of domain specialists, solvers and other typed nodes for each query. Small language components render the completed graph. Inference is integer and bitwise, XNOR, AND, OR, and popcount. Outputs carry a full constraint trace, AION proof certificates, and are bit-identical under a fixed seed.","product-loom.383ac5859c":"Constraint-based inference.","product-loom.68f3503345":"Sublinear specialist selection.","product-loom.76db07eb20":"Product, Loom","product-loom.52a8d5d7d8":"See pricing","product-loom.038e05aff8":"See Charter","product-loom.79b5cfa58c":"Request a proof of value","product-loom.866992e0c3":"Specialists instead of one generalist. A constraint trace on every answer. Bit-identical replay under seed. EU residency, on-prem, or air-gapped. Scope a pilot, read the system card, or open the pricing page.","product-loom.f5b2669c57":"A model your reviewer can accept.","product-loom.fbf655df3a":"Loom assembles this trace as the inference happens","product-loom.61a0572c48":"Accepted","product-loom.5247a17dc0":"REPLAY READY","product-loom.cfd2fa11d7":"RULE APPLIED","product-loom.9991612749":"CONSTRAINT FIRED","product-loom.170e6e9b75":"SPECIALIST NAMED","product-loom.e29a79fe0c":"Review","product-loom.78ab6c7bce":"The trace is created as the inference runs, not assembled after the fact. Every output carries the source it came from, the specialist responsible, and a replay path. That is the record your reviewer accepts.","product-loom.4c2bcfc477":"Trace, provenance, and replay are product properties, not compliance decorations added after the questions get harder. That is what makes the review packet hold.","product-loom.d6c390d424":"Think about how an auditor accepts a number.","product-loom.ec7d9463e6":"How approval works","product-loom.329256ebae":"Loom, business","product-loom.b220a05ea0":"By construction","product-loom.df5a270073":"Article 25","product-loom.21c55d89bf":"Decisions carry provenance and replay bit-identical, attributable to a pinned version of the model.","product-loom.8ba59219bd":"Reproducible on demand","product-loom.c6e32a2948":"The trace records only what was actually consulted. Retention is configurable to your policy.","product-loom.d1fccb12d7":"Configurable trace retention","product-loom.e0535719e3":"Operated from our base in Arnhem, the Netherlands. Jurisdiction-selectable deployment.","product-loom.7c86201c80":"European data residency","product-loom.06ecbe729a":"GDPR Article 25, built in","product-loom.cbcf23d618":"PRIVACY BY DESIGN","product-loom.67b752639a":"MANAGED MESH","product-loom.d829b9c34f":"Keep control.","product-loom.560486613b":"Start with managed Fabric.","product-loom.1d197d39c5":"Where it runs","product-loom.b55e35ada1":"Data sovereignty preserved","product-loom.370fbd9d9e":"Lower operational cost","product-loom.5902e88c6f":"96% less","product-loom.6376a93576":"Binary constraint inference uses 96% less energy than a generative stack at equivalent task complexity.","product-loom.6409664656":"Energy overhead","product-loom.914d7a6f52":"Inference on your own hardware means no latency penalty for proximity and no dependency on an external API staying available.","product-loom.339c10a152":"Cloud round-trip","product-loom.8c71277670":"fixed cost","product-loom.98087ae369":"Fixed infrastructure cost replaces per-query billing. Teams stop rationing questions and start using the model as a daily tool.","product-loom.f576bd02f6":"Token billing","product-loom.e0c030b096":"When the model runs on your infrastructure","product-loom.a747d39a22":"THREE COSTS THAT DISAPPEAR","product-loom.7bc75fd0a9":"SCALE WITHOUT LIMITS","product-loom.f4646a7586":"96% LESS ENERGY","product-loom.da03acba45":"NO TOKEN COST","product-loom.e80adba481":"Start scaling by use.","product-loom.ad1f3843a1":"Stop paying by the token.","product-loom.1af4858f06":"What changes when it runs locally","product-loom.df45e07101":"Loom closes each one","product-loom.ec95d99629":"Four gaps","product-loom.15fe3af411":"no record","product-loom.64185d6339":"Nothing you could hand to procurement, audit, or a regulator as evidence.","product-loom.153cf7b201":"No record","product-loom.6fc982beee":"A transcript will not reproduce. Ask again tomorrow and the paragraph changes.","product-loom.b90a38ae7f":"No replay","product-loom.2a7596e1e7":"black box","product-loom.cacfcf6a75":"A single black box answered. You cannot name the specialist that was responsible.","product-loom.1d4c6b7587":"No attribution","product-loom.3aff926a8a":"A chat answer cites no rule and no version. There is nothing an auditor can trace back.","product-loom.857136f15b":"No policy version","product-loom.93d8537e9b":"Four gaps an auditor notices","product-loom.04ac85aedf":"WHAT A CHAT TAB LEAVES OUT","product-loom.2da7864289":"THE REPLAY","product-loom.256df1f17e":"THE SPECIALIST","product-loom.7f5464a063":"THE RULE","product-loom.090824d8fc":"THE ANSWER","product-loom.61cb671835":"Loom gives a record.","product-loom.7c9562c498":"A chat box gives a paragraph.","product-loom.b9011520d5":"Why not another AI tab","product-loom.2790ea68d5":"One trace per query","product-loom.ea0b388326":"Fix the seed. The same answer comes back bit-identical, with the full trace.","product-loom.040667c342":"Replay for the reviewer","product-loom.d377c8409a":"Source, rule, specialist, and decision state are recorded as the work moves.","product-loom.92d0758386":"Read the trace","product-loom.e369734d25":"Attach the documents. Loom routes to the specialists whose domain fits.","product-loom.c2c6c840d6":"Run with sources","product-loom.3510b292b5":"Choose the procurement review, compliance check, or audit query you need to answer.","product-loom.72fef1e3ae":"Pick the workflow","product-loom.400903293e":"Four steps, one record","product-loom.932f069ae7":"HOW A PROOF STARTS","product-loom.a97f1f9097":"EMIT","product-loom.eb44b37443":"CONSTRAINT","product-loom.0fdceaaeba":"ACTIVATE","product-loom.013d73e530":"ROUTE","product-loom.b45625f888":"A trace is a structured record: which specialists were active, which constraints satisfied, which rules triggered, and how the answer was assembled. A reviewer can point to the specific specialist that was wrong and correct exactly that, without touching the rest.","product-loom.a9abdfea02":"It is a record you can cite.","product-loom.39d42e999e":"A trace is not a transcript.","product-loom.c6c92a1c28":"The trace your reviewer can accept","product-loom.e163485bcf":"Four things","product-loom.e8a5a5a2c6":"Rules are explicit. Updating a policy updates the constraint set. The trace shows the change took effect the next time it fired.","product-loom.c11bf4bd0b":"Policy meets inference","product-loom.eb5f9cfcc9":"Catalogue versions are explicit. A specialist update does not rewrite past answers. Upgrade windows are deliberate events, not background noise.","product-loom.394b40e250":"No silent drift","product-loom.092082cff3":"Around 4 to 8 top-level domain specialists are selected on an average request. The actual weave varies with the question, state, and available resources.","product-loom.d2ab367337":"Bounded cost per query","product-loom.1f1dc7b31f":"If an answer is wrong, the trace tells you which specialist was responsible and which rule applied. Corrections land in a specific place.","product-loom.c4ec3d30ad":"A place to push back","product-loom.09563dfa85":"Four things a plausible answer cannot","product-loom.eafede3528":"WHAT THE TRACE GIVES YOU","product-loom.fe0ce1b835":"GOVERNANCE","product-loom.75fde9afb2":"COST","product-loom.e1a476f66b":"APPEAL","product-loom.5c762e0265":"AUDIT","product-loom.7e11321fa0":"A general-purpose model gives you a plausible answer and no way to defend it. Loom has 528 top-level domain specialists and selects around 4 to 8 on an average request. The actual weave follows the task, and every output records the specialists and other cognitive threads that were active.","product-loom.84d1af806a":"With traceability.","product-loom.d0dcf5154f":"Specialised knowledge.","product-loom.2badabc881":"Why specialists beat a generalist","product-loom.30d53b4762":"Managed customers use Fabric on the public Mesh and do not receive direct Loom access or operation rights. A licence permits direct Loom operation on customer infrastructure, including a fully isolated estate. The agreement states which managed capabilities are exposed through Fabric.","product-loom.4f98a122df":"You have already tried pasting the question into a chat tab. It gave you a confident paragraph with no policy version, no list of what fired, and nothing you could hand to an auditor. A chat box answers the question. Loom finishes it with the record attached.","product-loom.709aa1f60e":"When inference runs on your own hardware and costs almost nothing per query, the economics of AI inside an organisation change. You stop paying by the token. You stop waiting for a cloud round-trip. You stop rationing access to the model. Teams that previously had a quota on AI queries get unlimited experimentation at a fixed infrastructure cost.","product-loom.bfe06577c0":"CPU-first, no GPU required","product-loom.4b4eb7ad2c":"Hardware floor","product-loom.cc641dd6fb":"Bit-identical replay on seed","product-loom.e41aa1fbbb":"Reproducibility","product-loom.4a0c327eec":"Answer and constraint trace","product-loom.c03f08a8f2":"Output format","product-loom.265ca8c0e7":"managed outcome, licensed direct, or air-gapped","product-loom.c10e83c453":"Data residency","product-loom.b48513e4d1":"Commercial","product-loom.fef8e889dc":"Licence","product-loom.a3c686e711":"Category","product-loom.de4c77b587":"Dweve Loom","product-loom.dd3b86d1ef":"Product","product-loom.78abb73923":"Loom, model","product-loom.cba0624eb5":"Procurement sheet","product-loom.5b071dd462":"Read the system card","product-loom.c50dccd4b5":"See the trace","product-loom.bbb0713e72":"A general-purpose model gives you a plausible answer and no way to defend it. Loom produces a constraint trace on every output: which of the 528 domain specialists were consulted, which rules fired, and how the verdict was reached. The trace is the management story. Built for procurement, audit, and regulated workflows inside Europe.","product-loom.adc8c4f82b":"An answer you can","product-loom.challengeAccent":"challenge.","product-loom.c9a1fcf79b":"Open Fabric","product-loom.451c0155b5":"Ask Loom a question. See which specialists answered, which constraints fired, and why. Try it in Fabric, or run it where you need it.","product-loom.b9c3a5e1fc":"A team of specialists, not one guess.","product-loom.fe943f0139":"Loom, consumer","product-loom.a268e26994":"96% less energy","product-loom.2ad2743b67":"Local and fast","product-loom.08b73d4c99":"No special GPU. No upgrade project. Loom runs on the hardware you already have, using 96% less energy than a generative stack.","product-loom.d547b6667b":"Runs on your hardware","product-loom.beebb03206":"Rapid experiments cost almost nothing. Follow up on the answers that surprise you. Push boundaries without worrying about the bill.","product-loom.8268d9d68e":"Test wild ideas","product-loom.49a38ab868":"When there is no per-query cost and no latency penalty, you ask the next thought as it arrives.","product-loom.e0119b251b":"No more saving questions","product-loom.45c494d849":"When inference is cheap and local","product-loom.e8188264d4":"THREE THINGS THAT SHIFT","product-loom.491ed3c541":"EXPERIMENT FREELY","product-loom.536a08d31b":"FAST","product-loom.9be34046ca":"LOCAL","product-loom.33d2702e3c":"Start experimenting.","product-loom.f1309052cc":"Stop rationing questions.","product-loom.3e95780722":"What changes when it is fast and local","product-loom.3c18c89751":"528 top-level domain specialists. Around 4 to 8 selected on average, with the actual weave determined by the task, state, and available resources. The trace records what was really active.","product-loom.a8445a1566":"Ask the same question with the same settings and you get the same answer, word for word.","product-loom.8d32c4b095":"Every answer keeps three promises. No exceptions, no good days and bad days.","product-loom.837b2df885":"No exceptions","product-loom.e27b6b8b68":"Every constraint that fired is listed","product-loom.f138fd3cf8":"The ones who actually know","product-loom.1393aacad1":"Active per question","product-loom.1a356d705e":"One per domain, trained separately","product-loom.006ae60054":"Specialists on the team","product-loom.893953ffd5":"Three things, no exceptions","product-loom.d752d899f6":"WHAT EVERY ANSWER CARRIES","product-loom.ea4efcae4c":"TRANSPARENT","product-loom.fe453a5f12":"HONEST","product-loom.6c57b963b7":"RELIABLE","product-loom.b1bed5c35e":"Get the same answer.","product-loom.33357539d1":"Ask the same question.","product-loom.7b5d13926e":"Answers that show their work","product-loom.2a8c6d1a5a":"Team model closes them","product-loom.14c604cd7e":"Three gaps","product-loom.603f553592":"Without a fixed structure, the same question gets a different paragraph tomorrow.","product-loom.0731e03738":"Keep the same answer twice","product-loom.16068d05f0":"When the answer is wrong you need to know which specialist to correct, not retrain the whole model.","product-loom.8ff949afd0":"Show which part to challenge","product-loom.3fd8845b3b":"A single model trades depth for breadth. A team trades breadth for depth per domain.","product-loom.5fd9eebaf3":"Know everything deeply","product-loom.82e42fe309":"Three things a generalist cannot do","product-loom.47a5e8d005":"WHY ONE BRAIN IS NOT ENOUGH","product-loom.e686d8744d":"REASONING SHOWN","product-loom.2549ae2f1f":"4 TO 8 ACTIVE","product-loom.d9f0c01ea8":"528 DOMAIN SPECIALISTS","product-loom.a9e344a636":"Only the right ones answer.","product-loom.72d7be1c04":"Each specialist has a job.","product-loom.1dd6a17cb4":"How it works","product-loom.e68ace24f1":"When an answer costs almost nothing to produce, the way you use AI changes. You stop saving questions for important moments. You test ideas as they arrive. You follow up without worrying about the bill. That shift from rationing to experimenting is what a local, fast, specialist model makes possible.","product-loom.885e746810":"You should never have to wonder whether Loom had a bad day. Ask the same question with the same settings and you get the same answer, word for word. The specialists show their work so you can check before you trust. And because inference is fast and runs on the hardware you already have, you stop rationing questions. Ask freely. Experiment. Loom keeps up.","product-loom.e9fa16c290":"Ask about a medicine and the medical specialists answer. Ask about a tax rule and the finance specialists step in. Nobody wastes your time with things they do not know. And every answer shows you which specialists were involved, so if something looks wrong you know exactly where to look.","product-loom.576347ec82":"Europe","product-loom.15edfb4da4":"Made in","product-loom.856d828bc8":"Your device or ours","product-loom.3c1278f314":"Runs where","product-loom.032ed180c8":"Same answer, always","product-loom.d7e70b6e08":"Same question","product-loom.8b550c77da":"Shows its reasoning","product-loom.8a5949b3f8":"Every answer","product-loom.bf20864a5b":"around 4 to 8 on average","product-loom.7d44e71f57":"528, one per domain","product-loom.e87135c5ab":"Specialists","product-loom.0c707f2781":"Consumer","product-loom.8ac7aa173f":"What you get","product-loom.93d76e0e22":"See how it explains itself","product-loom.abfd569491":"Try Loom in Fabric","product-loom.f0a01b6cc9":"Most AI is one big model that tries to answer everything. Loom is different. It is a catalogue of 528 domain specialists, each trained for a defined responsibility, and only the ones that can support your answer get to speak. You see who answered, what they knew, and why. No guessing. No made-up facts.","product-loom.1bb62acfaa":"Not one know-it-all.","product-loom.ed32d3f216":"A team of specialists.","product-loom.0fed3b96cc":"Loom","product-loom.fe0a091fdb":"Products","product-loom.6754d07813":"Dweve Loom, Woven Intelligence","product-loom.ec0b92a272":"Dweve Loom, Woven intelligence","product-loom.a1b286cfb4":"Stop asking one model","product-loom.079637a183":"to be the whole mind.","product-loom.8dac685d2e":"Dweve Loom is a platform for deterministic AI, built around constraint learning and a typed, neurosymbolic cognitive graph. The task selects which cognition runs, from perception and retrieval to solvers, verification and 528 domain specialists, and the result carries an observable record of how it was established.","product-loom.b9b88cee6e":"See what one endpoint hides","product-loom.90214883b6":"Talk to us","product-loom.2762248dfa":"The proof rail","product-loom.3fb7fa8c4d":"WOVEN","product-loom.21a31b70dd":"Learned path","product-loom.06fb2de45d":"native one-bit","product-loom.f7424af90f":"Solver fabric","product-loom.53d7aff37f":"formal, first-class","product-loom.9a4a05cdd5":"Specialist layer","product-loom.d03ce7c763":"528, sparsely active","product-loom.a1a9a32197":"trace attached","product-loom.11548e1a10":"Active per task","product-loom.973b253451":"only the weave that ran","product-loom.ede6a56296":"One endpoint hides many jobs","product-loom.b0a606d0e8":"Uniform output,","product-loom.f6e80123ef":"invisible responsibility","product-loom.1f393e675a":"The modern AI endpoint accepts text and returns text. That interface makes many distinct responsibilities disappear. The output looks uniform because the system no longer exposes which kind of work was performed. That makes evaluation vague, correction expensive, and accountability difficult.","product-loom.c0d58b510b":"A class decision is hidden in prose. An extracted clause is hidden in prose. A relevance judgement is hidden in prose. A calculation is hidden in prose. A constraint check is hidden in prose. A proof obligation is hidden in prose. A refusal is hidden in prose. Loom restores the boundaries. Each cognitive thread has a defined responsibility, input, output, version, and evaluation surface. The answer can still be simple. The system behind it no longer has to be vague.","product-loom.210248ba79":"BOUNDARIES RESTORED","product-loom.44f4b24966":"TYPED OUTPUTS","product-loom.58568f36f2":"OWNED RESPONSIBILITY","product-loom.85479c1c69":"What the prose hides","product-loom.dbfac38c89":"four buried judgements","product-loom.a56d943ce2":"The class, buried","product-loom.74dc47bbd3":"A decision dressed up as a paragraph.","product-loom.328f952e24":"The span, paraphrased","product-loom.9d92bf1eec":"The exact clause replaced by a summary.","product-loom.b3e9b3d88e":"The solver status, guessed","product-loom.9d5a269559":"Satisfiable becomes probably fine.","product-loom.7388ab4997":"The correction target, unknown","product-loom.4e71e5afc8":"Nobody knows which job actually failed.","product-loom.f1f13c2436":"the answer can stay simple","product-loom.5f354b5dc1":"the system stops being vague","product-loom.b1f3147dd4":"Loom restores the boundary, then lets the task choose the pattern.","product-loom.5377346055":"The task chooses the weave","product-loom.49dd30b114":"Different work,","product-loom.3eb3c0bc2a":"different cognition","product-loom.3109c81d8d":"Different tasks need different cognitive shapes. Loom does not force a ticket label, a contract date, a policy search, and a constrained decision through one fixed pipeline. The task determines which perception, retrieval, memory, solver, reasoning, specialist, and verification threads become active, so the weave matches the work instead of carrying the cost of every capability.","product-loom.f86c9d5e17":"A request to label an incoming ticket may need one classifier. A request to find the cancellation date in a contract may need document perception and exact span extraction. A request to find the most relevant policy may need an embedder, retrieval, relevance scoring, ranking, and reranking. A request to determine whether several conditions can all be satisfied may need a constraint solver. A request to explain why an outcome occurred may need causal reasoning, memory, evidence retrieval, and verification. A complex cross-domain request may need several specialists and several solver or reasoning paths before a language model expresses the result.","product-loom.7188055556":"TASK-SHAPED","product-loom.ee0c8d034c":"NO FIXED PIPELINE","product-loom.a2982b7e7c":"ACTIVE ONLY","product-loom.4efe128ecd":"Four tasks, four weaves","product-loom.0bf7650187":"the pattern follows the request","product-loom.65fe0d8d63":"Label a ticket","product-loom.92e1c1653b":"One classifier. Nothing else wakes.","product-loom.d4927dda05":"tiny","product-loom.b4ffbae56a":"Find the date","product-loom.0ba13585df":"Document perception, exact span extraction.","product-loom.89f6229a11":"small","product-loom.c018adce42":"Find the policy","product-loom.2fbcec179c":"Embed, retrieve, score, rank, rerank.","product-loom.414768af9d":"wide","product-loom.d2a414a056":"Explain the outcome","product-loom.bb4416404c":"Memory, causal reasoning, evidence, verification.","product-loom.3dde59ff3d":"deep","product-loom.83055097ee":"no repeated topology","product-loom.710619412e":"the task is the pattern","product-loom.a6b92c3a48":"The weave starts with seeing the input correctly.","product-loom.fd47f80b24":"Perception is first-class","product-loom.e92dc5e332":"What was seen","product-loom.0957bcb5e8":"remains addressable","product-loom.f4f38abfef":"Correct reasoning starts with a trustworthy observation. Loom treats perception as a typed, addressable stage for text, documents, images, audio, video, and structured data. It records spans, regions, entities, fields, events, and signals so later reasoning and verification can point to what was actually seen, rather than hiding an early parsing mistake inside a fluent answer.","product-loom.1de54e4f6d":"The perception systems can detect structure, identify entities and relations, inspect tables, measure patterns, align modalities, profile datasets, find anomalies, and turn raw input into typed cognitive state. A later reasoning or verification thread can point back to the span, region, event, entity, field, or signal it actually used.","product-loom.3871c2cac3":"SIX MODALITIES","product-loom.9fe4a73eef":"TYPED OBSERVATIONS","product-loom.e3702e2209":"ADDRESSABLE LATER","product-loom.4dd87531e4":"From input to observation","product-loom.302f944ea0":"what perception hands the weave","product-loom.1b8f117fd7":"Documents and text","product-loom.9034c1f2d0":"Layout, spans, entities, tables, fields.","product-loom.cf0628c054":"Images and video","product-loom.06418a2162":"Regions, events, aligned modalities.","product-loom.acdac205f5":"Audio","product-loom.53e2cae062":"Signals and events, typed and timed.","product-loom.3ac3d79b11":"Structured data","product-loom.e3e1ec4674":"Profiles, patterns, anomalies.","product-loom.585b842772":"not a prompt-prep step","product-loom.43eeb199ae":"typed state the weave can cite","product-loom.7acc30f511":"Seeing creates state. Memory decides what remains relevant.","product-loom.a7330f54d2":"Memory is not a longer context window","product-loom.df77956bfc":"Memory has jobs,","product-loom.f4e7a62a3f":"not just history","product-loom.2bbef1f1f5":"Memory is useful only when its purpose and scope are clear. Loom separates working, episodic, semantic, and procedural memory, then limits recall to the active task and policy. It can consolidate, decay, forget, replay, and inspect state without turning every prior conversation into context. Solvers and domain specialists receive only the facts and cases relevant to their responsibility.","product-loom.248b0696df":"Working memory holds the active task and intermediate state. Episodic memory preserves prior cases and outcomes. Semantic memory holds facts, relationships, and durable knowledge. Procedural memory stores reusable skills and ways of working. A solver receives the facts and constraints it needs. A domain specialist receives relevant cases and knowledge. A classifier is not burdened with unrelated history.","product-loom.26c7aec06e":"FOUR TIERS","product-loom.e1a729b6a0":"SCOPED RECALL","product-loom.0c4f8cc9d8":"BOUNDED BY POLICY","product-loom.587cbd0e73":"Four memory responsibilities","product-loom.29e31a8f5e":"and who gets access to what","product-loom.3b4dfc9713":"Working","product-loom.0206a32810":"The active task and its intermediate state.","product-loom.2de3ef9c04":"Episodic","product-loom.b6631e89ad":"Prior cases and their outcomes.","product-loom.2b3a5e307c":"Semantic","product-loom.0b08ba9697":"Facts, relationships, durable knowledge.","product-loom.4fce7cb895":"Procedural","product-loom.80a0a14a1f":"Reusable skills and ways of working.","product-loom.034cbf387d":"recall is scoped to the thread","product-loom.8fb6459d22":"never an unbounded transcript","product-loom.7b0659f959":"Memory can supply context. Search still has to find evidence.","product-loom.d15944bfe3":"Search is a weave","product-loom.5d8e8aabb9":"Represent, retrieve, judge,","product-loom.478264f2fd":"order, verify","product-loom.0ee65bf339":"Search contains several separate judgements. The query must be represented. Candidates must be found. Relevance must be estimated. Results must be ranked. The strongest candidates may need reranking. Policy, recency, provenance, or domain constraints may need to modify the final order. Loom gives each stage its own thread.","product-loom.a9708b3116":"Native binary embeddings produce compact search representations. Retrieval finds the candidate set. Relevance and ranking models narrow it. Reranking applies the deeper comparison only where it is worth the cost. Verification can check that the selected evidence actually supports the eventual answer. The expensive cognition is reserved for the difficult tail.","product-loom.233fa59588":"BINARY EMBEDDINGS","product-loom.bf84ef9ee1":"STAGED JUDGEMENT","product-loom.ab30551dea":"VERIFIED SUPPORT","product-loom.5358966352":"The narrowing funnel","product-loom.d51a1e2bba":"expensive cognition at the frontier only","product-loom.ddfe163345":"01","product-loom.35a1df921c":"Represent and retrieve","product-loom.aa3ceffe87":"Compact binary embeddings find candidates.","product-loom.bcac9d1d8e":"02","product-loom.54c192fcdd":"Score and rank","product-loom.130010ccfc":"Relevance models narrow the set.","product-loom.f235f22d8e":"mid","product-loom.3ea6c91e24":"03","product-loom.f78de65864":"Rerank the frontier","product-loom.0d7b9124bd":"Deep comparison only where it pays.","product-loom.8def4372cd":"few","product-loom.798f861ee7":"04","product-loom.5866f77781":"Verify support","product-loom.8ffd475256":"Evidence stays linked to its sources.","product-loom.d594c2cc0a":"final","product-loom.70b016d7dd":"candidates narrow, provenance survives","product-loom.e331820c7e":"each stage is a thread","product-loom.19b151bdee":"the tail gets the depth","product-loom.d489390d9e":"Sometimes evidence is not enough. The problem itself must be solved.","product-loom.3ed007ac9c":"Some questions should be solved, not predicted","product-loom.46b3449dc3":"Solve, do not guess","product-loom.717ff18f7a":"Use the native method","product-loom.4b004adfb7":"A transformer can write a plausible answer to a scheduling problem. That does not mean the schedule satisfies every constraint. A language model can guess whether a formula is satisfiable. That is not a SAT result. A model can describe an optimisation trade-off. That is not the optimum. Loom treats solvers as first-class cognitive threads.","product-loom.50e107ab4a":"The solver fabric spans logical, constraint, optimisation, temporal, spatial, symbolic, graph, automata, equilibrium, probabilistic, and synthesis problems. When the problem has a solver-shaped answer, Loom can route to a solver. The result can be satisfiable, unsatisfiable, optimal, infeasible, proved, disproved, unknown, or bounded by the solver's actual contract. It does not need to become persuasive prose first.","product-loom.a40817c546":"SOLVER-SHAPED ANSWERS","product-loom.078d1f9dac":"REAL STATUSES","product-loom.af1b8c7518":"NO PROSE CONFIDENCE","product-loom.8e60f6ad19":"Problem families, solver outcomes","product-loom.c544686009":"the statuses stay exact","product-loom.84ab5d1430":"Logic and constraints","product-loom.bfe6fddf3d":"SAT, UNSAT, or honestly unknown.","product-loom.6987c279b2":"Optimisation and scheduling","product-loom.2e3d310bff":"Optimal, infeasible, or bounded.","product-loom.ad70922af5":"Temporal, spatial, graph","product-loom.0f83c4f74e":"Consistency and structure, checked as such.","product-loom.6e91a51e0a":"Symbolic and synthesis","product-loom.4626bb173a":"Proved, disproved, or a produced artefact.","product-loom.c831b9645c":"the status is the answer","product-loom.93b1c01a07":"never rewritten as confidence","product-loom.58f8935088":"Solving is one form of reasoning, not the only one.","product-loom.3b4736a10a":"You have read this answer","product-loom.27ea9bf501":"The model produced a confident answer again.","product-loom.banner.business.headlineLead":"The model produced","product-loom.banner.business.headlineAccent":"a confident answer again.","product-loom.2656912f8a":"The date was paraphrased instead of extracted, and a scheduling constraint quietly went missing.","product-loom.6fe143e9a7":"A satisfiability question received a probability. And nobody can tell whether the answer came from memory, retrieval, or invention.","product-loom.63c996352e":"Around one confident answer","product-loom.163a90db4a":"The date, paraphrased","product-loom.306d304d05":"The constraint, missed","product-loom.423757e867":"The SAT question, given a probability","product-loom.f6f8b09df0":"The source, unknowable","product-loom.b1eef76078":"You did not need a larger model. You needed different forms of cognition.","product-loom.5a82b3e27a":"Loom, Woven Intelligence","product-loom.c327f965dd":"Reasoning has more than one shape","product-loom.b8b69c3f15":"Rules, causes,","product-loom.3ca36e3dfa":"hypotheses, alternatives","product-loom.64859e04c4":"Reasoning is not one operation with different labels. Deduction, induction, abduction, analogy, causality, time, space, probability, decisions, and counterfactuals preserve different structures and failure modes. Loom dispatches by problem shape, lets compatible modes combine, and keeps solver results and uncertainty typed instead of smoothing them into one confident paragraph.","product-loom.8607869aee":"Deductive reasoning derives conclusions from rules. Inductive reasoning learns patterns from examples. Abductive reasoning ranks explanations. Analogical reasoning transfers relational structure. Causal reasoning examines interventions and dependency. Temporal and spatial reasoning preserve their own relation systems. Probabilistic reasoning uses fixed-point factor graphs where uncertainty is part of the problem. Decision reasoning evaluates actions and utility. Metacognitive reasoning monitors which reasoning mode is working. Counterfactual reasoning asks what changes under a different condition. Fuzzy, default, modal and deontic reasoning preserve degrees, exceptions, possible worlds and obligations.","product-loom.94a574fb4a":"15 FORMS","product-loom.0f6b2d16bf":"DISPATCHED BY STRUCTURE","product-loom.8a3884e7df":"COMBINABLE","product-loom.f01d9c1fba":"Four of the shapes","product-loom.70d4eff396":"different questions, different artefacts","product-loom.8d34fa0520":"Deductive","product-loom.fe3eac46c9":"From rules to a derivation.","product-loom.0446305442":"derive","product-loom.f9509e19e6":"Abductive","product-loom.5a8ed5b924":"From evidence to ranked explanations.","product-loom.f7c1380a0d":"explain","product-loom.4ca169f381":"Causal","product-loom.9299242c81":"From interventions to dependency paths.","product-loom.10cff4034c":"why","product-loom.3194e21de4":"Counterfactual","product-loom.3e0c3d23ad":"From a changed condition to a changed outcome.","product-loom.304ea0fbae":"what if","product-loom.b94c55e3d9":"the problem picks the mode","product-loom.2c010531f4":"a metacognitive eye watches","product-loom.fbbb2ac04c":"Domain depth joins only where the general weave is not enough.","product-loom.82dbc79d00":"Domain depth arrives as specialists","product-loom.170c792b5a":"Specialists join the weave,","product-loom.5839e7092e":"they do not replace it","product-loom.08f13d1695":"Some work requires more than a general cognitive path. Loom includes a catalogue of 528 domain specialists across foundational reasoning, mathematics, science, code and systems, language, domain industries, multimodal work, verification, transfer, and metacognition. A query does not activate all of them.","product-loom.630e61bc57":"Routing narrows the catalogue and selects a bounded relevant set. The selected specialists contribute inside the wider weave alongside memory, retrieval, solvers, constraints, and verification. The specialist catalogue adds depth. It does not replace the rest of the cognitive system.","product-loom.cacbb2e2ec":"528 AVAILABLE","product-loom.fe15e79518":"BOUNDED ACTIVE SET","product-loom.37f8882108":"INSIDE THE WEAVE","product-loom.8a7c1bf199":"The specialist layer, in numbers","product-loom.68af197af0":"scoped to the layer, not the weave","product-loom.1d7deb7af3":"528","product-loom.b3c8ef87da":"specialist catalogue","product-loom.5670986eca":"Reasoning to industries, verification to metacognition","product-loom.ea20726887":"typically active","product-loom.884215f3a7":"Routing selects a bounded relevant set","product-loom.91032ad7bb":"20","product-loom.f7dabf6b11":"for complex queries","product-loom.a58899288f":"The upper range, not the default","product-loom.dd3f5dac83":"traced","product-loom.5033628d7f":"every activation","product-loom.271b055f02":"Selected specialists appear in the trace","product-loom.ec1414ee5f":"the graph stays the hero","product-loom.c5ed2c29e8":"the catalogue stays behind it","product-loom.7ec58e9974":"A transformer can also join, but it does not own the whole weave.","product-loom.5c63b72480":"A transformer is one thread too","product-loom.e0fad84770":"Language is one thread","product-loom.e7d989e605":"language is the work","product-loom.a1a5cd7ca9":"Language components are valuable. They are not the only shape intelligence can take. Loom includes fully bipolar transformer paths for tasks that genuinely require contextual language modelling, composition, or generation, while constraint learning and typed solvers govern the result.","product-loom.791de1cf69":"The transformer can sit after exact extraction, retrieval, solver output, or verification rather than being forced to imitate those capabilities inside its own weights. This changes its role. The transformer becomes the right language thread inside a larger cognitive path. It no longer has to pretend to be the entire path.","product-loom.03008eb029":"AFTER THE FACTS","product-loom.497174a7a4":"CONSTRAINED EXPRESSION","product-loom.7b211021e7":"ONE THREAD OF MANY","product-loom.c9378d192d":"Two ways to use a transformer","product-loom.b7407c07e6":"the difference is what feeds it","product-loom.2adbc86009":"Monolith","product-loom.76784b70b6":"everything in the prompt","product-loom.828a108309":"facts summarised into text","product-loom.775b0586a7":"constraints described, not enforced","product-loom.b08136393e":"solver results paraphrased","product-loom.a413586f70":"one output, one blame target","product-loom.5e554820b1":"the model imitates the whole path","product-loom.67e51bf50e":"typed inputs to expression","product-loom.4b7ff415eb":"exact spans enter as spans","product-loom.445cee33fd":"solver statuses enter as statuses","product-loom.6283eb3254":"evidence enters with provenance","product-loom.2f33a4f9eb":"expression is the last thread","product-loom.26f0f4e501":"the model does the language work","product-loom.194329187b":"hard sockets, not prompt summaries","product-loom.2cef0ab09a":"expression comes last","product-loom.34f1336e6c":"Every learned thread shares one compact computational contract.","product-loom.8e28620254":"One-bit is the common learned thread","product-loom.672c5952d6":"Trained for the representation","product-loom.5d4c8b64a9":"it runs","product-loom.a0f1fb1e6e":"Loom's learned models share a native bipolar one-bit foundation, while the governed knowledge state is built through constraint learning. Classifiers, span models, embeddings, retrieval models, rankers, transformers, routers, and domain specialists are trained for the representation they run. There is no full-precision canonical model waiting behind an inference-only quantiser.","product-loom.c712dc4e5e":"Training, adaptation, routing, and control remain binary and integer. This common substrate makes the learned threads compact, deterministic on supported paths, and economical to activate. It is what allows Loom to carry many specialised capabilities without requiring all of them to run on every request. Symbolic and solver components retain their own native structures: graphs, intervals, proofs, and constraints are not forced into a neural representation.","product-loom.3eb77048dd":"NATIVE ONE-BIT","product-loom.c9f2ed8c4e":"BINARY + INTEGER TRAINING","product-loom.9191dc0d4d":"SYMBOLIC STAYS SYMBOLIC","product-loom.62d47e9c2d":"Two materials, one weave","product-loom.cb6a31cc3c":"each thread in its native form","product-loom.c510bd86c3":"1-bit","product-loom.1001d48ae7":"Learned threads","product-loom.885542d97f":"Models, embeddings, routers, specialists","product-loom.45a9360afa":"bipolar","product-loom.1178cafbd6":"integer","product-loom.abf5a32a92":"Training and control","product-loom.bfeda85da1":"Binary and integer end to end","product-loom.b55e22fe78":"exact","product-loom.6f5b01580c":"native","product-loom.33fc3f3bff":"Solver structures","product-loom.5e1d8194fa":"Graphs, intervals, constraints","product-loom.06f9b0facd":"symbolic","product-loom.ef64363d23":"proof","product-loom.3004cccd56":"Verification objects","product-loom.1ae072e376":"Proofs where supported, as proofs","product-loom.6e69a85ca4":"formal","product-loom.79502a6978":"no canonical float model","product-loom.fde956c96f":"no marketing symmetry either","product-loom.4728971b70":"Because activation is sparse, capability can grow without every query growing with it.","product-loom.3cd98070b8":"Cost follows active cognition","product-loom.96a31917cc":"Pay for the weave that ran","product-loom.cf5450c03e":"query cost follows the weave","product-loom.e5aaf099ca":"A catalogue can grow without every query becoming more expensive. Loom activates only the threads required by the weave. The organisation pays for the cognitive path that ran, not for every capability that was available.","product-loom.7f93a10a08":"One classifier does not need to wake a transformer. One exact extraction does not need 528 domain specialists. One retrieval task does not need a theorem prover. One formal problem does not need a generative loop. This creates a better scaling law than one monolith growing more expensive as it absorbs every new capability.","product-loom.31cdaa1425":"SPARSE ACTIVATION","product-loom.88690c4f0a":"LIT NODES ONLY","product-loom.7f9cd8dfda":"ADDITIVE CATALOGUE","product-loom.6465e47ef5":"Three queries, three scopes","product-loom.56bffc164b":"illustrative relative activation","product-loom.c3d4d5c216":"tiny weave","product-loom.d74ddce0b5":"One classifier lights up","product-loom.817c73aa6d":"Find and rank policies","product-loom.dd789acfd7":"medium weave","product-loom.5ff2842abd":"The search stages light up","product-loom.8e9f6123d4":"Cross-domain explanation","product-loom.29cd95aa72":"deep weave","product-loom.51c7673918":"Reasoning, solver, specialists join","product-loom.fbfc00a425":"the meter follows lit nodes","product-loom.c404d7b892":"never the catalogue","product-loom.a20e7e2685":"The catalogue can therefore expand additively.","product-loom.3c82c3e92c":"Capability grows additively","product-loom.2e2739ca40":"Add a thread","product-loom.8ee508f04b":"without retraining the mind","product-loom.e62fdedc83":"OWN TESTS","product-loom.d356d13ad2":"OWN VERSION","product-loom.182cc51aa2":"EXISTING THREADS STABLE","product-loom.0300106e93":"How a thread joins","product-loom.845ad27f0d":"four steps, no global retrain","product-loom.61cc55aa04":"Add","product-loom.0689ed8927":"The new thread arrives with its contract.","product-loom.d7132581b3":"Evaluate","product-loom.f6174587f9":"Its own tests, on its own responsibility.","product-loom.d672995a14":"Register","product-loom.870a1257db":"Versioned into the catalogue.","product-loom.4999528efe":"Route","product-loom.259c179d6b":"Tasks that need it start using it.","product-loom.9f6f017a2d":"richer weave","product-loom.6d17839534":"never one indivisible model","product-loom.6fbcfd0a83":"Additive growth also gives every correction a location.","product-loom.077d728635":"Corrections have an address","product-loom.9c556a3f78":"Fix the responsibility","product-loom.e867c507a3":"that failed","product-loom.6a1ec19a89":"When a monolith is wrong, the correction target is often unclear. Loom records the active weave. A defect can be traced to the thread that owned the responsibility. The correction can land there rather than becoming another global retraining project.","product-loom.e5a48117c5":"Was the document parsed incorrectly? Was the wrong memory retrieved? Was the ranking weak? Was the reasoning mode inappropriate? Did the solver receive the wrong constraints? Did a specialist apply an outdated rule? Did the language thread distort a correct result while explaining it? Each of those questions now has an owner.","product-loom.5048c0897c":"LOCAL REPAIR","product-loom.c096276c8a":"PINNED NEIGHBOURS","product-loom.698a5b0b83":"NO GLOBAL RETRAIN","product-loom.bd6b2bca53":"Symptom to owner","product-loom.868ce1d9df":"where each defect lands","product-loom.52c2eb5de9":"Wrong field found","product-loom.a7e6451543":"Perception or extraction owns it.","product-loom.a18e1b3934":"Irrelevant evidence","product-loom.2ec4b5dd45":"Retrieval or ranking owns it.","product-loom.c308236eb8":"Impossible schedule accepted","product-loom.5c12822005":"Constraints or solver input owns it.","product-loom.be4031d1ee":"Right result, wrong story","product-loom.cd4db8ead0":"The expression thread owns it.","product-loom.3af706c2a3":"one red node","product-loom.6e8929d6de":"one local fix, replay green","product-loom.f843696180":"That is possible because the trace maps the weave.","product-loom.19c518f5f1":"The trace maps the weave","product-loom.8ff3387588":"Observable cognition,","product-loom.2ecbbf790f":"not private thoughts","product-loom.5b5d1a05dd":"Every answer can carry the observable cognitive path behind it. This is not private chain-of-thought. It is the operational map of the cognitive system that actually ran.","product-loom.455e6fed56":"The trace can name the perception outputs used, the memory items recalled, the models activated, the retrieval and ranking stages, the reasoning modes selected, the solver invoked and its result, the constraints applied, the specialists consulted, the verifier or proof checker used, and the final expression path.","product-loom.d870450254":"EVERY ACTIVE NODE","product-loom.e7d3e63262":"EVERY HANDOFF","product-loom.b2ccbf24f3":"NO THOUGHT TRANSCRIPT","product-loom.b39e3b29b5":"What one trace names","product-loom.aaf049c8d4":"in the order it happened","product-loom.3e4c3c40e9":"Seen and recalled","product-loom.fe52941ec0":"Perception outputs and memory items.","product-loom.d841fd3394":"Found and ranked","product-loom.6b483a50ca":"Retrieval stages and their evidence.","product-loom.6506c64fdd":"Reasoned and solved","product-loom.2e3ed32163":"Modes, solver calls, constraints, specialists.","product-loom.5631cfb564":"Checked and expressed","product-loom.3490a95885":"Verification verdicts and the final path.","product-loom.aa15479c67":"a readable map","product-loom.df3b59e19b":"of the cognition that ran","product-loom.25ceb44bb0":"Sometimes the trace ends in refusal, and that is a feature.","product-loom.33c3db17fc":"Refusal is a valid weave outcome","product-loom.64a40c2c4b":"No defensible path,","product-loom.7386808e8d":"no fabricated answer","product-loom.d7261893b0":"TYPED REFUSAL","product-loom.eed896c904":"CAUSE NAMED","product-loom.a98d9a92f0":"NEXT STEP CLEAR","product-loom.b0208741aa":"Refusal, with a reason","product-loom.8fb60d21db":"and what to do about it","product-loom.911fa7a0a1":"Insufficient evidence","product-loom.579ad5d50a":"Retrieval came back weak. Add sources.","product-loom.100ec4462d":"evidence","product-loom.2b643cc1c2":"Conflicting constraints","product-loom.16928e7840":"The requirements cannot all hold. Review them.","product-loom.ae214513a0":"conflict","product-loom.ab2fa71287":"Solver unknown","product-loom.fba88391e4":"The problem outran the budget. Tighten it.","product-loom.50d8b4a941":"unknown","product-loom.e10d7e514c":"Verification failed","product-loom.79782a3d09":"The draft did not survive the check. Inspect it.","product-loom.1f087a5954":"rejected","product-loom.a75b8366a5":"four paths emit results","product-loom.9b2d27f142":"the fifth stops honestly","product-loom.b949f41022":"The same cognitive contract can run in several deployment postures.","product-loom.cc077db610":"One cognitive system, several deployment postures","product-loom.62a219b26f":"Where the weave runs","product-loom.490bb81079":"is an operating choice","product-loom.8b89358dba":"Managed customers use Fabric through Dweve's service boundary; this does not grant direct Loom access. Direct Loom operation belongs to a licensed deployment on customer-controlled, edge, or isolated infrastructure. The licensed scope and managed capabilities are stated separately in the agreement.","product-loom.26852e9043":"The deployment determines where the weave runs. It does not redefine what the weave means. Available integrations and hardware change with the posture. The cognitive contract does not.","product-loom.b31d88128b":"ON-PREMISES + EDGE","product-loom.fa760e3c7d":"AIR-GAPPED","product-loom.585ee0f798":"Four postures, one contract","product-loom.5a294dd121":"what changes, what never does","product-loom.9b78aaba24":"Managed Mesh","product-loom.9645dfa1c7":"Managed Fabric service; no direct Loom access.","product-loom.c238039dee":"On-premises","product-loom.eeb7ec537c":"Your infrastructure, your control.","product-loom.8440b9eb7d":"Edge","product-loom.cccc80a8d2":"Close to where the work happens.","product-loom.5df2ebe7c0":"Air-gapped","product-loom.8074235dbf":"No outbound sockets at all.","product-loom.12ef701ae6":"location and integrations change","product-loom.77660df195":"the cognitive contract does not","product-loom.80c40f6bb3":"One final map shows where Loom sits in Dweve.","product-loom.4272e414ab":"Loom in the Dweve stack","product-loom.1434cd4348":"Operations run below","product-loom.4f6ec89f5c":"cognition is woven here","product-loom.dec6dc90a1":"Core supplies the complete AI machine that executes Loom's learned and algorithmic operations. Spindle supplies governed, lossless knowledge and provenance. Loom weaves cognition from perception, memory, models, retrieval, reasoning, solvers, verification, and specialists.","product-loom.a25a92e5c4":"Nexus places cognitive capabilities inside executable organisations with agents, tools, workflows, authority, and people. Mesh supplies distributed compute. Fabric is where people ask, inspect, and use the result.","product-loom.b5efa00a30":"ONE STACK","product-loom.335e52b2ac":"CLEAR RESPONSIBILITIES","product-loom.f4a1774adc":"NOTHING REBUILT","product-loom.fe3660a5fe":"Who does what","product-loom.eacb20ad2b":"around the woven middle","product-loom.68836c550e":"Core","product-loom.ec8e6c4f8d":"The machine that executes the operations.","product-loom.4a86080d67":"Spindle","product-loom.a6e6e9a043":"Governed, lossless knowledge and provenance.","product-loom.485a04fef4":"Nexus","product-loom.e253bdc81a":"Cognition placed inside executable organisations.","product-loom.6a454bb788":"Mesh and Fabric","product-loom.48efa3ffa3":"Distributed compute, and the human surface.","product-loom.43e2511125":"Loom weaves the cognition","product-loom.2ee19976e4":"the stack carries the rest","product-loom.d573b4f1e5":"Bring the task that exposes","product-loom.494b6c47e5":"A task that needs exact extraction and explanation. A task that mixes retrieval with formal constraints. A task that needs memory, causal reasoning, and verification. A task where a solver should replace a guess. A task that crosses domains without waking an entire monolith. See the weave Loom builds around it.","product-loom.c2921360a2":"Bring one real task","product-loom.9e3c1d591d":"What the claims stand on","product-loom.c1dfd96eea":"6","product-loom.a8d7e63e6a":"perception modalities","product-loom.1634988224":"Text, documents, images, audio, video, structured data","product-loom.1b64538924":"4","product-loom.1f45f8a7a7":"memory tiers","product-loom.ea597fedc8":"Working, episodic, semantic, procedural, scoped by policy","product-loom.7859321229":"Sparsely active: around 4 to 8 top-level domain specialists on average, without a hard cap","product-loom.a3ca9913ae":"native learned path","product-loom.883328ff87":"Trained binary and integer, observable trace attached","product-loom.182386ab5d":"Dweve Loom, Woven Intelligence runtime","product-loom.0d35ddce7d":"A request becomes","product-loom.068adc9bb9":"a cognitive graph","product-loom.a6e52838d4":"Loom is Dweve's deterministic AI platform for solver-backed language models: a woven cognitive model built around constraint learning. The typed cognitive graph is compiled from the task, policy, evidence requirements and execution budget. Only the required nodes activate, and the result leaves with a weave trace.","product-loom.1336c4004c":"See a graph compile","product-loom.161480a2d3":"Read the architecture","product-loom.1fdf175152":"RUNTIME","product-loom.91a4801bd6":"Graph nodes","product-loom.d5a420b405":"typed contracts","product-loom.239c093a9b":"Solver dispatch","product-loom.437db53efc":"native and integrated","product-loom.46d73ac1d5":"Specialist routing","product-loom.8866c2bd7e":"sparse, deterministic","product-loom.cf1caaae2a":"Every run","product-loom.a2b913dfc3":"observable trace","product-loom.6bea5119a1":"Threads have typed contracts","product-loom.9c12be0dcf":"Every node","product-loom.78839b524a":"signs a contract","product-loom.cdf56b1847":"A cognitive thread is not an untyped callback. It has an input shape, output shape, capability signature, version, cost model, determinism contract, failure modes, and evidence surface. Typed outputs keep the cognitive graph from dissolving into prose between every stage.","product-loom.83f00d001f":"A span extractor emits offsets and labels. An embedder emits a binary representation. A retrieval stage emits candidates and distances. A ranker emits order and scores. A SAT solver emits SAT, UNSAT, or unknown, plus a model or certificate where supported. A causal reasoner emits paths, interventions, or independence results. A verifier emits an accepted result, a rejected result, or a named reason it cannot certify the path.","product-loom.7100d12f3e":"TYPED IN AND OUT","product-loom.06d55361b1":"VERSIONED","product-loom.107f2cdb74":"EVIDENCE SURFACE","product-loom.16482ca0e6":"Callback versus contract","product-loom.c5f1a39195":"why the planner needs types","product-loom.cc59dc5c15":"Untyped callback","product-loom.ecb492bd10":"a tool that returns text","product-loom.f2d473e69c":"shape unknown until it runs","product-loom.776d877666":"no version to pin","product-loom.852a17f45f":"failure looks like prose","product-loom.406551baa8":"no evidence surface","product-loom.187df05ecc":"the planner cannot reason about it","product-loom.7d264a7b0d":"Typed thread","product-loom.5202e46c02":"a signed cognitive contract","product-loom.700309dab2":"input and output shapes declared","product-loom.ed06225704":"capability signature and version","product-loom.7db8fb921b":"named failure modes","product-loom.9ba4e49a41":"an evidence surface to cite","product-loom.9ae4dc2620":"the planner can compose it","product-loom.f2d1eef122":"not a callback","product-loom.482168508b":"a contract the planner reads","product-loom.22cf1f7b08":"Contracts let the planner build a graph rather than a prompt chain.","product-loom.cd6803f484":"The graph is planned","product-loom.807f4ad0b4":"Task, modality, domain,","product-loom.59f8c79e7f":"evidence, budget","product-loom.f5c5862198":"Loom does not run one fixed pipeline for all tasks. The task is classified by structure, modality, domain, required evidence, and possible solver family. The planner can construct a shallow graph for a bounded task or a deeper graph for a complex one. The graph is explicit enough to inspect and dynamic enough to fit the task.","product-loom.569056e679":"Parallel perception or retrieval branches can converge before reasoning. Several reasoning modes can operate over the same state. A solver can replace or constrain a predictive stage. A verifier can reject the graph before expression.","product-loom.b223103d5d":"SHALLOW OR DEEP","product-loom.ea4d93b3a5":"INSPECTABLE","product-loom.ccaef55933":"The compile stages","product-loom.f282929bfd":"from task to emitted result","product-loom.28b6dc3117":"Analyse","product-loom.f0b45b09bf":"Classify the task by structure, modality, and domain.","product-loom.ae2f98a099":"Plan","product-loom.3e4ad5c2c4":"Resolve a shallow or deep graph for the work.","product-loom.ac7f958cc0":"Resolve","product-loom.188f62c2c0":"Pin versions for every node in the graph.","product-loom.6ea36ce8d4":"Execute","product-loom.23ccb28ff2":"Run parallel branches where the graph allows.","product-loom.dda6ac27b9":"Verify","product-loom.2617cc1a3b":"Gate the result before it can be expressed.","product-loom.08afa6be84":"Emit","product-loom.3a08abad6c":"Return one result plus the weave trace.","product-loom.f6f7031af0":"the topology fits the task","product-loom.987cc9aab3":"never the reverse","product-loom.d647306ef4":"Learned nodes share one compact representation.","product-loom.8570c5a454":"One binary representation connects learned components","product-loom.2515956c6d":"XNOR, popcount,","product-loom.c9105c8e51":"XOR bind, majority bundle","product-loom.a102062846":"Loom's learned components use compact bipolar representations. Binary hypervectors provide a common space for similarity, routing, memory association, retrieval, and specialist activation. The shared representation reduces conversion between learned threads and makes exact routing state reproducible.","product-loom.b0e7797c6e":"XNOR and population count replace floating-point dot products on the hot path. XOR binding provides self-inverse association. Majority bundling provides compositional representations. The symbolic and solver layers keep their own native structures rather than being coerced into vectors where that would lose meaning.","product-loom.92555fa877":"SHARED BIPOLAR SPACE","product-loom.5433465a9c":"NO FLOAT HOT PATH","product-loom.334f61cb93":"SYMBOLIC STAYS NATIVE","product-loom.4f50c7f8c5":"Two planes, one weave","product-loom.834ac46778":"learned binary, symbolic native","product-loom.bac7081cb6":"Learned plane","product-loom.aebc1afa47":"XNOR","product-loom.b8810e6986":"Population count replaces dot products","product-loom.3313261f4e":"hot path","product-loom.c7807b9951":"Symbolic plane","product-loom.cdb7e65462":"CNF, graphs, intervals, proofs","product-loom.e345c5bf73":"own structure","product-loom.4d46077c78":"bound","product-loom.3f44e41832":"Conversion marker","product-loom.d6b2d6b28a":"Cross only where meaning survives","product-loom.a45c2264b8":"explicit","product-loom.31c50967be":"learned threads share the plane","product-loom.606ad2ec39":"symbolic keeps its own","product-loom.52d39ef543":"The first major thread family is perception.","product-loom.a40c6bcdac":"Perception fabric","product-loom.725992262a":"Typed observations,","product-loom.0bece009bd":"not generic prompt context","product-loom.2e0d7c4b15":"The perception fabric covers text, documents, images, audio, video, and structured data. All outputs remain addressable to downstream memory, reasoning, and verification nodes.","product-loom.15a87bb79a":"Text paths include tokenisation, normalisation, entity and relation extraction, classification, intent, discourse, sentiment, argument structure, and summarisation hints. Image paths include thresholding, edges, connected components, descriptors, texture, segmentation, similarity, and classification. Audio paths include energy, voice activity, pitch, spectral analysis, rhythm, fingerprinting, and speaker features. Video paths include motion, tracking, scene change, and activity recognition. Document and structured paths include layout, metadata, tables, schema, statistics, quality, anomalies, drift, patterns, joins, and temporal analysis.","product-loom.44404d4a15":"ADDRESSABLE FIELDS","product-loom.78fcc99b5a":"Modality by capability","product-loom.8d6d81eafb":"representative operator families","product-loom.c3328c39b0":"Text","product-loom.fca7863dd8":"Tokenise, entities, relations, classification, discourse.","product-loom.687c82861c":"Documents","product-loom.da32051668":"Layout, metadata, tables, schema, quality.","product-loom.09e871c98f":"Images","product-loom.c00d35d344":"Edges, components, descriptors, segmentation.","product-loom.13fa1c32b7":"Energy, voice activity, pitch, spectral, speaker.","product-loom.bc17c1f017":"Video","product-loom.7b64b85499":"Motion, tracking, scene change, activity.","product-loom.14dde95874":"Structured","product-loom.dc83ae429b":"Statistics, anomalies, drift, joins, temporal.","product-loom.e882bbc47c":"exact fields and spans","product-loom.b276fc8e20":"consumed downstream","product-loom.21ce91f7e1":"Observations become useful when memory can retain and recall them.","product-loom.6049a0a038":"Memory fabric","product-loom.4d14879d27":"Working, episodic,","product-loom.4a0fc03465":"semantic, procedural","product-loom.dbac689942":"Loom's memory system separates working, episodic, semantic, and procedural state. Working memory is bounded and attention-aware. Episodic memory supports content-addressed recall and temporal context. Semantic memory stores typed triples and knowledge structures. Procedural memory stores decision trees, production rules, and reusable skills.","product-loom.09d881bc32":"Lifecycle components handle consolidation, decay, compaction, replay, scheduling, and forgetting. Retrieval can use direct similarity, LSH, graph association, temporal indexing, or knowledge queries. Monotonic counters keep internal memory evolution independent of wall-clock drift.","product-loom.81d43619e5":"BOUNDED LIFECYCLE","product-loom.77ecc5595d":"CONTENT-ADDRESSED","product-loom.ede832931e":"Tier, structure, lifecycle","product-loom.e73b42a68d":"four responsibilities","product-loom.02e4ef5045":"Bounded, attention-aware active state.","product-loom.b41659d7a9":"Content-addressed recall with temporal context.","product-loom.190293cf7c":"Typed triples and knowledge structures.","product-loom.991f0667a3":"Decision trees, production rules, reusable skills.","product-loom.141f40846d":"recall is scoped","product-loom.1577782403":"evolution is bounded","product-loom.13d6ad8941":"Bounded learned decisions become micro-model threads.","product-loom.1ad40e4b55":"Micro-model fabric","product-loom.a120e09058":"Bounded responsibility,","product-loom.bb57f7d4e4":"bounded evaluation","product-loom.7733a18822":"Classification, exact span extraction, relevance, ranking, reranking, retrieval, and embeddings are examples, not the boundary of the model family. Loom can host purpose-built learned components wherever a bounded statistical decision is the correct abstraction. Each model owns one responsibility and one evaluation surface.","product-loom.df0d7ce120":"A micro model can sit before a solver, after perception, inside retrieval, beside a reasoning mode, or as a gate that decides whether a deeper branch is worth running. The model is small because the responsibility is narrow, not because the capability is secondary.","product-loom.64344f8715":"ONE RESPONSIBILITY","product-loom.8b9919c285":"ONE METRIC","product-loom.a76c6a121f":"ARCHITECTURAL PRIMITIVE","product-loom.0cd58bfe0a":"Responsibility to output","product-loom.1911a15371":"narrow decisions, typed results","product-loom.a16c4ed9fa":"Classify","product-loom.0f28e0528e":"A bounded label decision.","product-loom.8d767bf5b7":"class","product-loom.6d84ceaf60":"Extract","product-loom.2dfbdf2312":"Exact spans with offsets.","product-loom.6325f8c630":"span","product-loom.dd48a11495":"Rank","product-loom.367996582a":"Order and scores over candidates.","product-loom.cce55e4309":"order","product-loom.5701b5f6fe":"Gate","product-loom.459ff7ab95":"Whether a deeper branch runs.","product-loom.3c5b492776":"gate","product-loom.44a8e3bd89":"small because narrow","product-loom.cf9f0e08ba":"not because secondary","product-loom.64bd63599c":"Search is where several of these nodes form one native graph.","product-loom.35df9fec63":"Native search graph","product-loom.05d47eeb65":"Encode, retrieve, score,","product-loom.95e2dd5bd2":"rank, rerank, verify","product-loom.a2b11670a3":"The search weave is a graph of independent nodes: encode the query, retrieve candidates, score relevance, rank, rerank the selected candidates, apply policy and provenance filters, then verify support. The embedder and retrieval index operate natively in binary space.","product-loom.8121d57d92":"Hamming and XNOR plus population count provide compact similarity paths. Reranking remains separate from candidate generation, so expensive comparison is applied only to the frontier. The final evidence set remains typed and source-addressable.","product-loom.9e43eafeb7":"BINARY INDEX","product-loom.185203bdf5":"STAGED NODES","product-loom.1cfb5f74d7":"SOURCE-ADDRESSED","product-loom.3150c77a57":"Candidates by stage","product-loom.fa95f6de5a":"the frontier narrows","product-loom.782e817cf0":"Encode","product-loom.182067a4f3":"Query to binary representation.","product-loom.eb55f47b49":"Retrieve","product-loom.62c573c77c":"Candidate set from the index.","product-loom.e6ee45c97e":"Relevance narrows the set.","product-loom.6117abc77a":"Rerank and verify","product-loom.5c9596dcf0":"Deep compare, then support check.","product-loom.346e3652ea":"compact generation","product-loom.8180e6d82f":"depth at the frontier","product-loom.ddcec71404":"Language composition can join without owning the earlier stages.","product-loom.6be6198617":"Fully bipolar transformers","product-loom.27f3c3e38d":"Native one-bit,","product-loom.409fa7ddf3":"constrained by typed state","product-loom.0d9873d31b":"Loom's transformer paths are trained and executed in the native bipolar representation. They are language-rendering threads inside a constraint-learning graph, not a traditional transformer-only model. There is no hidden float model serving as the canonical source of truth for inference. Transformers can perform contextual language work, sequence modelling, composition, and generation inside the cognitive graph.","product-loom.ee86b8640c":"Their output can be constrained by exact extraction, solver results, knowledge state, specialist contributions, or verification nodes. They are powerful threads. They are not privileged above every other thread.","product-loom.c25c5c82cf":"TYPED PORTS","product-loom.bfa0313543":"NOT PRIVILEGED","product-loom.66e129b91c":"The transformer, in numbers","product-loom.08514f3dd5":"representation and state","product-loom.52b6dd2efb":"learned representation","product-loom.330a6b6340":"Weights and activations bipolar","product-loom.b612df5e84":"accumulator class","product-loom.a78849de39":"No float master model underneath","product-loom.33469d3faa":"pinned","product-loom.6c8fdb437c":"replay state","product-loom.1448385c79":"Versions and complete state recorded","product-loom.bc743d5e22":"facts arrive through ports","product-loom.86608a7aa2":"not a prompt summary","product-loom.a7c9f3920f":"Deeper cognition dispatches across explicit reasoning modes.","product-loom.c4f9d7b60a":"Reasoning mode dispatch","product-loom.2291e34bf2":"Different reasoning,","product-loom.08b6cbc15e":"different engines","product-loom.6b22a2bca4":"The unified reasoning layer exposes multiple modes behind typed requests and responses. The mode is part of the trace, not hidden inside a generic generation call.","product-loom.582bb19085":"Deductive and backward chaining operate over facts and rules. Induction retains the evidence behind a generalisation. Abduction ranks hypotheses, while analogy transfers relational structure. Causal, temporal and spatial reasoning use their own graphs and relation systems. Probabilistic and decision reasoning retain distributions, utilities and policies. Metacognitive, counterfactual, fuzzy, default, modal and deontic paths preserve their own outcomes and limits.","product-loom.b6580e0220":"15 FORMS","product-loom.ab0056be22":"TYPED DISPATCH","product-loom.3d27f2f51a":"MODE IN THE TRACE","product-loom.e021c1e06f":"Mode, structure, output","product-loom.3c799fbd25":"four of the fifteen","product-loom.9ac115e7b5":"Facts and rules, backward chaining.","product-loom.94b134aabd":"Graphs and d-separation.","product-loom.0efe5fd2a3":"Temporal","product-loom.495f28c63c":"Allen relations, consistency propagation.","product-loom.30603fa9e0":"when","product-loom.c2a4d654c5":"Interventions, alternative worlds.","product-loom.27a377f226":"the mode is explicit","product-loom.99171cb7e8":"not one confidence score","product-loom.c1d1faf7d5":"Some requests are not reasoning-mode problems. They are solver problems.","product-loom.73f6b70e85":"Logic, maths, optimisation","product-loom.4da0f917a3":"graph, automata, synthesis","product-loom.6abe9d295c":"Loom can dispatch across native and integrated solver families according to problem features and required guarantees. The purpose is not to call every solver. The purpose is to recognise when the problem belongs to one.","product-loom.c033e9b084":"The surface includes SAT, QBF, DQBF, MaxSAT and XOR-SAT; SMT with bit-vectors, arrays, strings and linear and nonlinear arithmetic; CSP with global constraints and temporal networks; ASP, Datalog, constrained Horn clauses, first-order proving and finite-model finding; LP, QP, ILP, MIP, MINLP and scheduling; graph algorithms, automata, model checking and Petri nets; symbolic algebra, intervals, differential systems and matrix equations; equilibrium, probabilistic inference and planning; and CEGIS, SyGuS and reactive synthesis. Native and integrated methods remain identified in the weave trace.","product-loom.17d9a7fd00":"24 REGIONS","product-loom.d41e241ea6":"NATIVE AND INTEGRATED","product-loom.ee5e98b952":"EXACT VOCABULARY","product-loom.514505c185":"Family, native, integrated","product-loom.69b7510ae9":"backends where enabled","product-loom.122f1700f9":"Logical","product-loom.75eca81f8b":"SAT, MaxSAT, XOR-SAT. Kissat and CaDiCaL where enabled.","product-loom.ba42eddff9":"SMT","product-loom.dafd435d71":"Bit-vectors, arrays, strings, arithmetic. cvc5 and Z3 integrated.","product-loom.c794704381":"Optimisation","product-loom.475a76af93":"LP, QP, ILP, MIP. HiGHS and SCIP integrated.","product-loom.52e68a873a":"Constraints","product-loom.7a72356fe9":"CSP, global constraints, temporal and spatial networks.","product-loom.028cfac607":"Graph and automata","product-loom.17b0cec71d":"Flow, matching, model checking, Petri nets.","product-loom.41dcffd22d":"Algebra, e-graphs, CAD, CEGIS, SyGuS.","product-loom.79011627d5":"recognise the shape","product-loom.8aa469b52a":"never call every engine","product-loom.d191b4fb0e":"The right solver is selected through an explicit portfolio contract.","product-loom.5d7132606f":"You have written this pipeline","product-loom.c10c26f945":"You just asked a transformer to solve SAT again.","product-loom.banner.technical.headlineLead":"You just asked a transformer","product-loom.banner.technical.headlineAccent":"to solve SAT again.","product-loom.4ed04d8479":"Generation stood in for classification, and embeddings stood in for an exact match.","product-loom.1a77bea1d6":"RAG answered an arithmetic question, a prompt planned a schedule, and an LLM judge ruled on a theorem the solver could decide.","product-loom.b306f81c70":"One prompt, five wrong tools","product-loom.7756c6bff1":"Generation for classification","product-loom.a7d191ed95":"Embeddings for exact match","product-loom.52ee927d42":"RAG for arithmetic","product-loom.fd1c534d70":"A prompt for scheduling","product-loom.607a8a303b":"A prompt is not a cognitive architecture.","product-loom.ef612a1386":"Solver dispatch and portfolio","product-loom.d829ede4ca":"Problem features, theory,","product-loom.06ad8bc5f0":"proof need, budget","product-loom.0e68a66fd9":"Problem features can route the query into the solver registry or a portfolio. The dispatcher can consider the theory, variable types, constraint density, expected proof needs, warm-start state, available backend, and execution budget. A portfolio can select or sequence solvers rather than locking the system to one engine.","product-loom.69445ca602":"Native solvers and FFI-backed solvers remain distinguishable in the trace. WIP capabilities remain versioned and excluded from production routing until their contract is ready. The output keeps the solver's real status. SAT is not confidence. UNSAT is not a refusal written by a model. Unknown is not silently converted into a plausible answer.","product-loom.8380ba31d1":"REGISTRY AND PORTFOLIO","product-loom.dbcdb08d5c":"NATIVE VERSUS FFI","product-loom.4e1194f79c":"REAL STATUS OUT","product-loom.9a8023b14d":"Outcome to next action","product-loom.ec4ace7629":"what each status permits","product-loom.615e0eb11d":"SAT plus model","product-loom.8cd7818d3f":"The assignment can feed the next node.","product-loom.6ef16c9410":"UNSAT plus core","product-loom.5fb54cd8f2":"The conflict names the failing constraints.","product-loom.5193a46114":"Optimal","product-loom.0fb19babf1":"The bound is exact, not a suggestion.","product-loom.bc7819b34f":"Unknown","product-loom.d67c4fceae":"Timeout stays unknown, never a guess.","product-loom.9b89d78978":"the portfolio scores candidates","product-loom.d9e4b9bd2c":"the status stays exact","product-loom.0853724625":"Solver output can feed knowledge and verification without becoming prose.","product-loom.2bae309548":"Knowledge, contradiction, belief revision, proof","product-loom.290eda8663":"Facts, conflicts,","product-loom.540463f942":"revisions, certificates","product-loom.2d36d03349":"Knowledge bases and graphs store facts, rules, entities, relations, analogies, abstractions, and commonsense structures. Contradiction detection finds direct, implicit, or temporal conflicts. Belief revision can expand, contract, or revise the current belief base.","product-loom.4c06030dd5":"Proof checking validates proof objects where the selected method provides them. AION can attach independently checkable certificates to supported paths. Not every cognitive path is formally certifiable. The trace names what was checked, by which mechanism, and what remained an inference rather than a proof.","product-loom.9227352649":"AGM-STYLE REVISION","product-loom.73aadabd91":"PROOF CHECKED","product-loom.a0e3cf5280":"CERTIFICATES WHERE SUPPORTED","product-loom.608810eca6":"The verification steps","product-loom.15dbe359b9":"each artefact stays distinct","product-loom.80628bd746":"Detect","product-loom.16a9f6b2dd":"Find direct, implicit, or temporal conflicts.","product-loom.b450347518":"Revise","product-loom.849e392d0b":"Expand, contract, or revise the belief base.","product-loom.68cb47d254":"Prove","product-loom.577d5637aa":"Check proof objects where the method provides them.","product-loom.55cbfd1b6c":"Explain","product-loom.a9e1ecf824":"Name what was proved and what stayed inference.","product-loom.a5314cb102":"proof is not explanation","product-loom.c39f20104a":"inference is named as inference","product-loom.82bd055d4f":"Domain depth adds another sparse layer.","product-loom.4a84504d31":"Sparse specialist layer","product-loom.54b9ee0851":"Domain depth","product-loom.10fd82d1fa":"inside the wider graph","product-loom.81ad4d5351":"Each specialist is separately versioned, evaluated, and addressable. Routing begins from compact signatures and narrows the catalogue through approximate neighbours, Permuted Agreement Popcount, and per-specialist gate subsets. A deterministic top-k selects the active set.","product-loom.eca832ae80":"The specialists run concurrently where the graph allows it. Their outputs can feed constraints, solvers, verification, or final synthesis. The specialist layer supplies domain depth. It does not replace perception, memory, retrieval, reasoning, or solvers.","product-loom.a215525516":"528 CATALOGUE","product-loom.42dee5f028":"DETERMINISTIC TOP-K","product-loom.544499f269":"ONE SPARSE LAYER","product-loom.48b41e2711":"Catalogue to active","product-loom.ce3ea9ea2e":"the routing funnel","product-loom.5a24102340":"Catalogue","product-loom.4b83a910fe":"528 domain specialists, versioned and addressable.","product-loom.52e6d8ab88":"full","product-loom.b5bf8067cf":"Candidates","product-loom.fb50d154bf":"Approximate neighbours narrow the field.","product-loom.c63ff60ad3":"Gates","product-loom.0fb038eaf8":"Per-specialist gate subsets score the pool.","product-loom.a733b809d2":"Active","product-loom.1fcf44857f":"Deterministic top-k selects the set.","product-loom.170a9cd325":"around 4 to 8 top-level domain specialists on average; complex requests can select more or fewer as needed","product-loom.62a3bbf21d":"domain depth added","product-loom.f2bd456148":"Loom routes at more than the specialist level.","product-loom.4ee4556cf7":"Routing at several levels","product-loom.3d54b1bdfe":"Graph, modality, memory,","product-loom.71e8a91bf4":"search, solver, specialist","product-loom.3012721934":"Loom routes more than specialists. It can route from task to cognitive graph, from modality to perception path, from memory query to tier and index, from search query to retrieval and ranking path, from problem structure to reasoning mode, from theory and constraint shape to solver, from domain signature to specialist subset, and from evidence state to verifier or refusal.","product-loom.7f16a43919":"Each routing decision can be recorded with the candidate set, scores, thresholds, version, and selected path.","product-loom.7c2aa40a90":"8 LEVELS","product-loom.105d5703f4":"CANDIDATES SCORED","product-loom.4772357d5c":"DECISION RECORDED","product-loom.f1869ba465":"Route type to target","product-loom.4ee7c6e280":"a decision at each level","product-loom.e55379cc0a":"Task to graph","product-loom.5acdfca805":"Shallow or deep topology selected.","product-loom.f42822147c":"Problem to solver","product-loom.ca132ba146":"Theory and features pick the backend.","product-loom.07037798bc":"Domain to specialists","product-loom.2bccb52a38":"Signatures select a bounded subset.","product-loom.95f43a4b0d":"Evidence to verifier","product-loom.13b8fc4e0b":"Or to a typed refusal edge.","product-loom.2744998d5a":"candidate set and scores","product-loom.156906de3b":"threshold, version, path","product-loom.4cd755a91f":"Different outputs require explicit merge semantics.","product-loom.81379de5ef":"Typed merge semantics","product-loom.874d275102":"Union, rank, quorum,","product-loom.93eacf2e03":"arbitration, dominance","product-loom.53064a6285":"Different threads require different merge operations. Candidate lists may be unioned, intersected, or reranked. Constraint sets may be combined or found inconsistent. Specialist outputs may require quorum, arbitration, or a verifier. Reasoning modes may contribute competing hypotheses rather than one averaged answer.","product-loom.297b00b3d0":"Solver results may dominate predictive suggestions where exactness is required. The merge is part of the graph contract. It is not an implicit final prompt that asks a language model to make everything sound coherent.","product-loom.e157d2d6ae":"MERGE IS TYPED","product-loom.a802708c28":"EXACT DOMINATES","product-loom.acbdeaab79":"NO COHERENCE PROMPT","product-loom.0dfbee03b0":"Input family to rule","product-loom.d9bb4c8013":"five different operators","product-loom.d61d6c69cb":"Candidate lists","product-loom.909b724dd7":"Union, intersect, or rerank.","product-loom.da2ea5bee8":"lists","product-loom.b172dbcf67":"Combine, or find inconsistent.","product-loom.d6c822010f":"sets","product-loom.551cfb69a4":"Specialists","product-loom.0d7b2e7cea":"Quorum, arbitration, or a verifier.","product-loom.cfbede3cef":"specialists","product-loom.924a9a91f2":"Solver results","product-loom.81d67ebec0":"Dominate predictive suggestions.","product-loom.25f33363da":"merge is in the contract","product-loom.a3b299abc3":"not a final prompt","product-loom.e933a7f2d2":"Learned threads are trained for the representation they run.","product-loom.702e25da1f":"Binary and integer training","product-loom.3168a01ebc":"No float master model","product-loom.a32a0cb735":"hiding underneath","product-loom.08a79c79d2":"TARGET IN TRAINING","product-loom.0fc9c5e530":"NO POST-HOC QUANTISER","product-loom.51dfac7b2d":"INDEPENDENT FAMILIES","product-loom.33f786c5af":"The training stages","product-loom.ff1552bf5f":"to a sealed one-bit artefact","product-loom.1504d94575":"Initialise","product-loom.75c36392de":"Start from the binary data representation.","product-loom.3b278ea9e7":"Train","product-loom.3038a36837":"Integer and fixed-point updates, no float master.","product-loom.6752f198b5":"Validate","product-loom.f9ff094102":"Evaluate on the component's own surface.","product-loom.ff8fc1c58a":"Seal","product-loom.ed5d054fa1":"Freeze the one-bit artefact.","product-loom.519b98b024":"Version it into the catalogue.","product-loom.2ea226c93b":"the target trains","product-loom.8d8fc7002d":"families train alone","product-loom.299c7e48d8":"Independent training makes versioning and additive growth possible.","product-loom.d4b4a1951c":"Version every boundary","product-loom.83d86421ff":"Models, memories, solvers,","product-loom.5ca718e1a9":"knowledge, rules, specialists","product-loom.9a7b505877":"Every cognitive thread has an identity and version. Memory snapshots, model artefacts, solver builds, constraint sets, specialist definitions, policy bundles, and knowledge graph states can be pinned. A weave can therefore be reconstructed from the versions that actually ran.","product-loom.08374dcb29":"An upgrade is an explicit event. It does not silently rewrite the meaning of earlier answers.","product-loom.4d3ecd41ea":"EVERY BOUNDARY PINNED","product-loom.c3f3496746":"EXPLICIT UPGRADES","product-loom.3fb69833ea":"RECONSTRUCTABLE","product-loom.b99a24fabc":"The weave manifest","product-loom.a1c6904b35":"a swapped ranker, isolated","product-loom.5b9622665c":"Original weave","product-loom.81f412a15c":"ranker v3, pinned","product-loom.20cbd629e8":"Changed node","product-loom.f99efc36ce":"ranker v4, swapped in","product-loom.844e371901":"Isolated effect","product-loom.7cfb66da77":"only the ranking node differs","product-loom.eda94fe216":"hashes per node","product-loom.d591c89c44":"the diff stays isolated","product-loom.5b5895cf33":"Replay includes the complete graph and state.","product-loom.1199212a1b":"Deterministic cognitive replay","product-loom.aa0a06e882":"Graph, routes, memory","product-loom.ff2a491fec":"models, solvers, versions","product-loom.5847a36dfc":"Loom replay is pure deterministic replay. The runtime records the complete graph, canonical execution order, routing decisions, model and specialist artefacts, memory state, indices, knowledge, constraints, solver state, limits, dependency outputs and external evidence used by the run. Learned, integer, fixed-point, bitwise, symbolic and exact paths all execute without random seeds.","product-loom.5133b955da":"External sources remain external. Captured replay reuses the recorded evidence and dependency outputs. Live replay can call them again and identify where the world changed.","product-loom.c06a9f8f92":"CANONICAL ORDERING","product-loom.5c6bb90ca9":"NO REDUCTION DRIFT","product-loom.eefe8e9b72":"CONTROLLED SURFACE","product-loom.8f1b183710":"Captured versus live","product-loom.a33eeb9ee2":"two replay contracts","product-loom.5be168c1e2":"Captured replay","product-loom.fac744678a":"Reuse recorded evidence and outputs.","product-loom.544130cf2c":"deterministic","product-loom.e8df00d54c":"Live replay","product-loom.af7fb06388":"Call sources again, find what changed.","product-loom.53215ac87b":"diverges","product-loom.4151fc1ab0":"the controlled core is exact","product-loom.ffff1a9f48":"external evidence can change","product-loom.9d5a00a6c2":"The record is observable execution, not hidden reasoning prose.","product-loom.aa235be851":"Trace, not chain-of-thought","product-loom.d94e16e27b":"Nodes and artefacts,","product-loom.61c9ac5ece":"never private thoughts","product-loom.7cfef38375":"OBSERVABLE EXECUTION","product-loom.f293b5bdf7":"ARTEFACT IDS","product-loom.3623d58dd5":"NO PRIVATE CHANNEL","product-loom.e96fa04331":"Event sequence","product-loom.c14234f657":"with artefact IDs","product-loom.5cce1ad944":"01 perceive","product-loom.8726ddf71c":"observation IDs","product-loom.e3345cd5f8":"02 recall","product-loom.0694606010":"memory IDs","product-loom.c585282415":"03 retrieve","product-loom.dde83cfde3":"evidence IDs","product-loom.098e43cda9":"04 solve","product-loom.adf28a84d9":"status and model","product-loom.90c990b7c7":"05 express","product-loom.4a35d9d27f":"final path ID","product-loom.7d293abfaa":"a readable sequence","product-loom.a317dfd306":"no thought transcript","product-loom.b7e8f86be3":"Many underlying capabilities overlap, but the product semantics change.","product-loom.3132432738":"Loom and Nexus share foundations, not semantics","product-loom.89a6aab2b4":"Same foundations,","product-loom.eb5f8c3549":"different product","product-loom.91be22eff4":"Both products can use perception, memory, reasoning, solvers, knowledge, and verification. In Loom, those components are nodes inside one cognitive graph producing one intelligence result. In Nexus, they can be capabilities inside agents with identity, lifecycle, authority, tasks, communication, and organisational responsibility.","product-loom.78b66810bd":"Loom can serve one Nexus agent, several Nexus agents or Fabric directly. The two products can share cognitive faculties without sharing the same responsibility: Loom composes a cognitive graph, while Nexus organises agents, authority and work.","product-loom.1f73770ed2":"COGNITION","product-loom.1786fd01f9":"SHARED FACULTIES","product-loom.ad91819695":"CLEAR BOUNDARY","product-loom.ef996ee0d6":"Loom thread versus Nexus agent","product-loom.df14e1adc8":"the same foundation, two roles","product-loom.ae662fe493":"Loom thread","product-loom.06d1564e3c":"a node in one cognitive graph","product-loom.d535ec68ec":"one intelligence result","product-loom.c5e3924847":"typed input and output","product-loom.e67bc93ce2":"no identity of its own","product-loom.02a270ae91":"lives inside the weave","product-loom.49e1a9c411":"cognition composed for a task","product-loom.0e1bf2726b":"Nexus agent","product-loom.d590f1b2a0":"a capability with authority","product-loom.c4b074c2a0":"identity and lifecycle","product-loom.fd760295f9":"tasks and communication","product-loom.12b549df6c":"organisational responsibility","product-loom.df0def04b4":"can call Loom","product-loom.c3240e8c44":"work organised across agents","product-loom.87cda63596":"a thread in a graph","product-loom.65dff89c96":"a capability in an agent","product-loom.0550318632":"Both ultimately execute on Core.","product-loom.290e5c1d05":"Core executes the weave","product-loom.01ce726b2a":"This layer chooses the cognition","product-loom.ee02620069":"the machine below runs it","product-loom.5fce5fe98e":"Core provides the operation universe, numeric representations, training engine, kernels, dispatch, and hardware backends that execute Loom's learned and algorithmic paths. Loom determines which cognitive graph should run. Core determines how its operations run on the available machine.","product-loom.3425ccb07f":"Selected strategic deployments can place Core on Kera for the deeper graph-native execution foundation.","product-loom.c5f7330673":"OPERATION UNIVERSE","product-loom.0443e4c73d":"KERNELS AND DISPATCH","product-loom.830303b35d":"CORE ON KERA","product-loom.2fd4504582":"The execution map","product-loom.7d2f4f837d":"from graph to backend","product-loom.be5cedcef0":"Chooses which cognitive graph runs.","product-loom.b14e0fee60":"Core Native","product-loom.128af5682f":"Executes operations on the machine.","product-loom.5bd7bdd9b6":"Core on Kera","product-loom.f455dfa73b":"Deeper graph-native execution foundation.","product-loom.b58bbcc943":"Loom chooses cognition","product-loom.48c5535667":"Core executes operations","product-loom.e8fe3c9285":"One graph contract deploys into every posture.","product-loom.e967f41da3":"Deploy the cognitive graph","product-loom.7b4cd15c39":"Same graph contract,","product-loom.058380e648":"five deployment postures","product-loom.f37804f36b":"ONE CONTRACT","product-loom.f883a1d197":"SOCKETS BY POLICY","product-loom.dd157c7e31":"MEANING UNCHANGED","product-loom.a4700fc7a8":"Posture to dependencies","product-loom.f0c1c5d176":"what each posture permits","product-loom.310b994a27":"Your infrastructure, controlled sockets.","product-loom.b4b834e976":"Close to the work, local-first.","product-loom.bb3d98dfee":"Distributed","product-loom.1856094531":"Across Mesh compute.","product-loom.35538c11e6":"the contract travels","product-loom.330f8d474d":"the meaning does not change","product-loom.17f7d9b9ad":"Then build only the graph the task deserves.","product-loom.4a51f37db6":"Build the cognition","product-loom.32baa047a4":"Do not route every request to one model. Represent perception, memory, models, reasoning, solvers, verification, and specialists as the distinct computational responsibilities they are. Then let Loom weave only the graph the task needs.","product-loom.c871b775ac":"Compile a real task","product-loom.81419fcdde":"What the runtime stands on","product-loom.6215a83a54":"Working, episodic, semantic, procedural, bounded by lifecycle","product-loom.17ba079149":"15","product-loom.1c792f01cf":"reasoning modes","product-loom.4177720b25":"Deductive through counterfactual, dispatched by structure","product-loom.20b62f582e":"specialist sparse layer","product-loom.bec9caa341":"Deterministic top-k, native one-bit learned path, observable trace","product-loom.bfaf7bdc6e":"Dweve Loom, the cognitive fabric behind Fabric","product-loom.c93cdbdbcc":"Different questions wake","product-loom.eb6331ceb7":"different kinds of intelligence.","product-loom.7a57211ed2":"Loom is Dweve's deterministic AI platform behind many Fabric answers: a cognitive model that brings together language, perception, memory, retrieval, reasoning, solvers and verification. It does not force every job through language generation. You ask once, and Loom weaves the right intelligence for the question.","product-loom.faq.eyebrow":"Questions teams ask","product-loom.faq.title":"What to know before you scope a Loom task","product-loom.faq.intro":"These answers set out Loom's category, its relationship to language models, and the limits that matter when you evaluate a result.","product-loom.faq.what":"What is Dweve Loom?","product-loom.faq.whatAnswer":"Loom is Dweve's platform for deterministic AI: a typed cognitive graph that combines language components, perception, memory, retrieval, reasoning, solvers, verification, and domain specialists. It compiles the path a task needs instead of routing every request through one model.","product-loom.faq.llm":"How is Loom different from a large language model?","product-loom.faq.llmAnswer":"A large language model is one part of Loom's cognitive graph. Loom can place language work alongside exact extraction, retrieval, memory, reasoning, solver, and verification threads, so a task receives the components its structure and evidence requirements call for.","product-loom.faq.unsupported":"How does Loom handle an answer it cannot support?","product-loom.faq.unsupportedAnswer":"When evidence is weak, constraints conflict, or a solver returns unknown, Loom can refuse instead of turning the gap into a confident answer. The weave trace records observable execution, while deterministic replay applies only on supported paths with a fully pinned state.","product-loom.faq.specialists":"How many specialists does Loom use?","product-loom.faq.specialistsAnswer":"The catalogue contains 528 top-level domain specialists. Around 4 to 8 are selected on an average request, but the actual weave varies with the question, state, and available resources.","product-loom.b40aa88880":"See how it decides","product-loom.f2b83c0462":"Try in Fabric","product-loom.d9f8d71d1c":"What wakes up","product-loom.ad07af46d6":"A simple label","product-loom.1a86d96f90":"one tiny model","product-loom.6d50dd0357":"A document","product-loom.5cc944acc2":"a reader and an exact extractor","product-loom.12db7acdfb":"A schedule","product-loom.f09d25fb50":"a real planner","product-loom.bf24c55e36":"A hard question","product-loom.83f8ca0954":"memory, reasoning, and checks","product-loom.2a4a05bcb4":"One assistant, every job the same way","product-loom.33744f7715":"Finding is not explaining","product-loom.5f2f0c6cfd":"Solving is not guessing","product-loom.ef56da7016":"Finding a date is not the same as explaining a contract. Recognising what is in a picture is not the same as planning a schedule. Searching for evidence is not the same as deciding whether the evidence is enough. Loom keeps those jobs separate behind the scenes.","product-loom.895a4c2845":"Checking whether a set of rules can all be satisfied is not the same as writing a paragraph about them. The answer still arrives through one conversation. The work underneath uses the method that fits the question.","product-loom.0479a1cbeb":"DIFFERENT JOBS","product-loom.b9a7c86262":"DIFFERENT METHODS","product-loom.f996815bcf":"ONE CONVERSATION","product-loom.40f3676867":"One size fits none","product-loom.bf6b774dff":"the wrong tool for each job","product-loom.d63529ef6f":"Find a date","product-loom.f20db997e0":"A paraphrase is not the exact date.","product-loom.e3533a1f08":"Plan a schedule","product-loom.fb6444d80d":"A guess is not a checked plan.","product-loom.ff6edb2546":"Weigh the evidence","product-loom.71950d52f6":"Fluent is not the same as enough.","product-loom.05fcaffda4":"Explain a contract","product-loom.473bd856fe":"A summary can drop the clause.","product-loom.968f75aa40":"different jobs","product-loom.fc34019c70":"different thinking","product-loom.b0410e8db1":"Small jobs should remain small.","product-loom.855d7ee3f7":"Small questions stay small","product-loom.b7455b3b6a":"Exact answer,","product-loom.05a28267ef":"exact source","product-loom.61da6ace45":"Suppose you ask what the payment deadline is in a letter. Loom does not need to wake a giant reasoning system. It can read the document, find the exact span, and return the date with its source location. A bounded job uses bounded intelligence.","product-loom.ebf798979c":"That makes simple work faster, cheaper, and easier to check. Everything else in the system stays dim while the small job runs.","product-loom.9a640f5f3e":"BOUNDED JOB","product-loom.d7c423178e":"BOUNDED INTELLIGENCE","product-loom.440b396948":"EASY TO CHECK","product-loom.795d0fa3fb":"Three small steps","product-loom.fe23c3cc6b":"read, find, return","product-loom.852b438f91":"Read","product-loom.60b0d77d6c":"Open the document as it is.","product-loom.df251b06ee":"Find","product-loom.c830604596":"Locate the exact span.","product-loom.38f1b6f3b6":"Give the date and its source.","product-loom.247ced472f":"nothing else wakes","product-loom.90be500a51":"the small job stays small","product-loom.21d6251d3e":"Some questions begin with another kind of seeing.","product-loom.bd3556dbe8":"It can see what you give it","product-loom.7e1550c65c":"More than text","product-loom.0f6458b39b":"documents, images and sound","product-loom.bbfc08346e":"Loom can work with more than typed text. It can inspect documents, tables, images, audio, video, and structured information where the product supports them. It can notice layout, find names and dates, recognise patterns, inspect fields, and turn the input into something the next part of the system can use.","product-loom.e4c172f01e":"The later answer can remain connected to what was actually seen.","product-loom.498b9c72fa":"MANY INPUTS","product-loom.99ef554b2f":"TURNED INTO OBSERVATIONS","product-loom.e1b64c02b7":"STILL CONNECTED","product-loom.238a1125fe":"Input to observation","product-loom.a26b677e40":"what the system notices","product-loom.e0bf1b93d4":"A photo","product-loom.6ff6a74a0f":"becomes objects and text.","product-loom.0e76292794":"image","product-loom.a938a02e4d":"A table","product-loom.61c6930f35":"becomes fields and anomalies.","product-loom.a17c9aaa61":"data","product-loom.ffbb0fe1a9":"A recording","product-loom.1ade5faff3":"becomes speech segments and events.","product-loom.a06a492959":"audio","product-loom.deac5e4013":"becomes layout, names, and dates.","product-loom.4f8278c89a":"document","product-loom.833d8ed78a":"more than typed text","product-loom.9eeca359cf":"connected to the source","product-loom.6261c44d57":"Useful observations need the right memory.","product-loom.245950d7d3":"It remembers the right things","product-loom.27f9ac318d":"Memory with a purpose,","product-loom.e6b14403f4":"not one endless log","product-loom.468fc414b5":"Memory is not one endless chat log. Loom can keep track of the current task, remember earlier cases, use facts from your knowledge base, and recall procedures you use repeatedly. Each part receives only the memory relevant to its job.","product-loom.fbe4ff9092":"The helper checking a deadline does not need every unrelated conversation. The helper working on a recurring form can remember how you handled it last time.","product-loom.f37968b30d":"FOUR PURPOSES","product-loom.74fbd8b46e":"ONLY WHAT MATTERS","product-loom.5111d2ca04":"NOT A TRANSCRIPT","product-loom.bde89fff82":"Memory with a job","product-loom.61b3a69d7e":"in plain language","product-loom.a70ee5167a":"Current task","product-loom.8e816fb7f8":"What you are working on right now.","product-loom.e7a5f95b36":"Earlier cases","product-loom.e8f1cd326b":"How similar work turned out before.","product-loom.dc601cb7cc":"Your facts","product-loom.16a0790366":"Knowledge you have given the system.","product-loom.d682fc9dab":"Procedures","product-loom.e808aa45ab":"Steps you repeat and reuse.","product-loom.a6bc69af79":"the right memory","product-loom.d01dda6010":"to the right helper","product-loom.6b35181300":"Memory helps, but it still has to search for new evidence.","product-loom.477d6f16c8":"It knows how to search","product-loom.4da73ec83b":"Find, compare,","product-loom.41817b303c":"order, check","product-loom.830b073f37":"A good search does more than return the first similar paragraph. Loom can represent the question, retrieve candidates, judge relevance, rank the results, compare the strongest ones again, and check whether they really support the answer.","product-loom.5ac08dfeda":"The system can use several small helpers rather than asking one large helper to remember everything.","product-loom.0dd4fcfed8":"SEVERAL HELPERS","product-loom.cc4dc22272":"FIND AND COMPARE","product-loom.c101fb7a91":"CHECKED SUPPORT","product-loom.00a75363cf":"Many to few","product-loom.cbd1dec2e3":"the search narrows","product-loom.ca2f02a1c9":"Many possible sources.","product-loom.f25470201a":"many","product-loom.90f3c4a018":"Judge","product-loom.4c9b1e198f":"Which ones are relevant.","product-loom.eb87581285":"some","product-loom.945f6e753b":"The strongest first.","product-loom.4b5e84be0e":"Check","product-loom.35bd338f35":"Do they support the answer.","product-loom.fe05bcdcdc":"one","product-loom.f18627a428":"many sources narrow to one evidence set","product-loom.f8038fc2b4":"not the first similar paragraph","product-loom.959ad5b37a":"checked support","product-loom.39c9c29c46":"Some questions need more than evidence. They need a real solution.","product-loom.cad14c808a":"Some problems get an exact answer","product-loom.7c33d8b9cc":"Schedules, rules,","product-loom.44603528f1":"maths, constraints","product-loom.3ca7d13512":"Imagine you ask Fabric to arrange several appointments under a set of rules. One person is only available in the morning. Two meetings cannot overlap. Travel time must be included. One appointment has a hard deadline. A chatbot can suggest a schedule. A careful checker works out whether the schedule meets every condition.","product-loom.1eb49bab2a":"Loom can bring in that exact checker when needed, then explain the result in plain language.","product-loom.746ba09eec":"A REAL CHECKER","product-loom.bd5f7b2d77":"EVERY CONDITION CHECKED","product-loom.6151b146ef":"EXPLAINED PLAINLY","product-loom.fbfff95bfa":"Rule and status","product-loom.419d5b122c":"each condition checked","product-loom.84e24f1fc4":"Mornings only","product-loom.3ae0ab357e":"Checked against every slot.","product-loom.19273cca45":"No overlap","product-loom.611d8d19aa":"Two meetings cannot collide.","product-loom.036fe1f0e6":"Include travel","product-loom.76ac6a7e6a":"Time between places counts.","product-loom.f35215d743":"Hard deadline","product-loom.6ecfedcc68":"One slot cannot move.","product-loom.408ac6d7aa":"a chatbot suggests","product-loom.87f270cad3":"a solver checks","product-loom.1a415f78ed":"Other hard questions need several ways of reasoning.","product-loom.59f8efcea4":"You have met this assistant","product-loom.ffcc228ecf":"The same assistant is doing your search, your maths, your planning, and your fact check.","product-loom.banner.consumer.headlineLead":"One assistant searches.","product-loom.banner.consumer.headlineAccent":"It calculates, plans, fact-checks.","product-loom.04e8bf7557":"It paraphrased the date, and then it guessed the schedule.","product-loom.ea24070690":"It ranked the wrong source first, and it sounded certain when the rules conflicted.","product-loom.e1c2b5f06c":"One assistant, four jobs","product-loom.fe3a90567e":"The schedule, guessed","product-loom.a1222113c2":"The wrong source, ranked first","product-loom.d51744beaf":"Certain while the rules conflicted","product-loom.ae344fbd7b":"Different jobs deserve different ways of thinking.","product-loom.4955e6d5ae":"More than one way to reason","product-loom.126de609bd":"comparisons, what-ifs","product-loom.9f53a5357d":"Some questions need rules. Some need causes. Some need comparison with an earlier case. Some need probabilities. Some need a what-if test. Loom can combine several ways of reasoning rather than forcing every problem through the same style of answer.","product-loom.afd66d7f27":"The right combination depends on the question, not on one fixed style.","product-loom.beac15c8fd":"SEVERAL WAYS","product-loom.24ed7b0eed":"COMBINED","product-loom.d61d417896":"NOT ONE STYLE","product-loom.ec33b6ed08":"Ways to think","product-loom.bb11a8e3f8":"Rules","product-loom.dde39e78fc":"Work step by step from what must hold.","product-loom.caa155adf8":"rules","product-loom.5aea12fd28":"Causes","product-loom.586cacfe7c":"Ask what actually led to the outcome.","product-loom.6ea5fcbb80":"causes","product-loom.2dfcf898a3":"Comparison","product-loom.d0e6cf2dbd":"Match against an earlier case.","product-loom.b64ca250f4":"compare","product-loom.9b0576a6d8":"What if","product-loom.4f9121cc96":"Test how a change would play out.","product-loom.c118362b10":"more than one way","product-loom.3bd4e815e9":"not one fixed style","product-loom.34fd35b8c3":"Subject specialists join only when deeper knowledge is needed.","product-loom.823c71fd62":"The right specialists speak","product-loom.8f14b2a9b7":"528 available,","product-loom.50e5a33afd":"the relevant few active","product-loom.d4a9924c68":"Loom has a catalogue of 528 domain specialists, but they do not all speak at once. The relevant few can join when a question needs deeper knowledge or harder reasoning. The rest stay quiet.","product-loom.9e989d0cfd":"The specialists are only one part of the wider system. They can work with memory, evidence, exact checkers, and checks rather than trying to do everything alone.","product-loom.ab863d27d5":"THE RELEVANT FEW","product-loom.c4e9461d83":"THE REST STAY QUIET","product-loom.3e62353c88":"Available and active","product-loom.fde9906f4b":"only the relevant few speak","product-loom.e20af2fe03":"specialists available","product-loom.a11915a24d":"Across many domains and reasoning forms","product-loom.90c3ee1eec":"The relevant few for a question","product-loom.7c9d1ae358":"about 4-8 on average","product-loom.66c161f9b1":"the relevant few speak","product-loom.f999496c1d":"the rest stay quiet","product-loom.c9af4e6abe":"Every contribution still has to survive checking.","product-loom.00e900b255":"It checks before it speaks","product-loom.8090556966":"A fluent answer","product-loom.9b3fcba01f":"is not a supported answer","product-loom.3f3a031d54":"COMPARED TO EVIDENCE","product-loom.4d3b321af0":"CHECKED FIRST","product-loom.2782ffdb30":"HONEST ABOUT PROOF","product-loom.c9bdba3b80":"Result type and meaning","product-loom.a3083c74c2":"what each card carries","product-loom.10de46b3b3":"Certificate","product-loom.e44fbddd9a":"A formal check the result passed.","product-loom.6a55f82754":"proved","product-loom.5bb922d772":"Sources that support the answer.","product-loom.2a5e9b1b17":"supported","product-loom.e9c4556335":"Warning","product-loom.b74bd093bd":"Something did not fully line up.","product-loom.5f2222812f":"caution","product-loom.6d4685a6dc":"Refusal","product-loom.559565258d":"Not enough to answer safely.","product-loom.5acf411124":"stopped","product-loom.2ff9f5104d":"fluent is not supported","product-loom.1764ccc41c":"the check decides","product-loom.83df1d808a":"You can see the useful record without reading hidden thoughts.","product-loom.1f881c3bb1":"You can see what helped","product-loom.cd9c18d34b":"The map of the weave,","product-loom.32c25a28e4":"in plain language","product-loom.729fae1f03":"You do not need to read the system's private internal thoughts. You can see the useful record: what it read, what it remembered, what it searched, which method it used, which exact checker or specialists helped, which checks passed or failed, and why it answered or refused.","product-loom.273199188c":"That is the map of the weave, written as observable work rather than a claim about hidden reasoning.","product-loom.3ee46af548":"AN OBSERVABLE RECORD","product-loom.b838c172bb":"PLAIN LANGUAGE","product-loom.0fa3489509":"NO HIDDEN THOUGHTS","product-loom.90688469b9":"What helped","product-loom.61a578b1fb":"the useful record","product-loom.459f2c24a8":"Saw","product-loom.b0eeb800c7":"the document and its fields","product-loom.4d3ebfca62":"Remembered","product-loom.fe212e1005":"the relevant earlier case","product-loom.ba050ff7a3":"Searched","product-loom.a100c04b04":"and compared the sources","product-loom.0eddb54c75":"Reasoned","product-loom.8d817e279c":"in the way the question needed","product-loom.e1d3550543":"Solved","product-loom.ae1829ccb9":"the exact part with a solver","product-loom.b2bc0c005f":"Checked","product-loom.03ba5c1b09":"before it answered","product-loom.ec942a8914":"observable work","product-loom.0a154f056c":"not a claim about thoughts","product-loom.181dac595c":"Sometimes the right answer is not to answer.","product-loom.fd5e84b7dc":"It can stop instead of guessing","product-loom.b0d87016cb":"Weak evidence and","product-loom.b09178387f":"conflicts stay visible","product-loom.c4bdb3e131":"Sometimes the document is unclear. Sometimes the evidence is weak. Sometimes the rules conflict. Sometimes the checker cannot decide within its limits. Sometimes the specialists disagree. Loom can tell you that.","product-loom.56459fdb3b":"It does not have to turn uncertainty into a confident sentence. A refusal names the missing piece and suggests a next step.","product-loom.f1a05bae18":"HONEST LIMITS","product-loom.85d2fe557c":"NEXT STEP SUGGESTED","product-loom.f52e0f7dee":"Cause and next step","product-loom.55a1364140":"what you can provide","product-loom.5641f0daa9":"Unclear document","product-loom.0e368a652f":"Share a clearer copy.","product-loom.40171d62aa":"Weak evidence","product-loom.c870108e58":"Add a source or two.","product-loom.9aea13c8e4":"Conflicting rules","product-loom.db7d05ab79":"Decide which rule wins.","product-loom.45da49e04f":"Solver undecided","product-loom.4e6258e035":"Loosen a hard limit.","product-loom.67bb115124":"no confident guess","product-loom.fa1263acc2":"an honest stop","product-loom.db30f03af1":"Compact cognition can also run closer to the user.","product-loom.5ae70b056c":"Useful intelligence can run close to you","product-loom.921fa125b0":"It can run close","product-loom.b53450fbb9":"to where you are","product-loom.50953fa4c4":"Loom is built to stay light. Its helpers are small and efficient, so many jobs can run on ordinary computers, local servers, small devices, and controlled European systems rather than a distant data centre.","product-loom.8c5b392774":"Where each job runs depends on the work. Loom can be used through managed Fabric on the public Mesh, or run directly under a licence on infrastructure you control. It does not assume every question must travel to a distant data centre to be answered.","product-loom.92c3ca06c2":"BUILT LIGHT","product-loom.bf3b4c5a52":"RUNS CLOSE WHERE SUPPORTED","product-loom.167ebe2e8c":"EU INFRASTRUCTURE","product-loom.87fd83158d":"Device and task","product-loom.cce87ff43f":"scale changes by workload","product-loom.a3b9e5c17c":"Ordinary processors","product-loom.9fc96499de":"Everyday supported tasks.","product-loom.edb06e97da":"Local servers","product-loom.53f5e9437d":"Heavier team workloads.","product-loom.0b8cdda4e5":"Edge devices","product-loom.783a0a1b15":"Work close to where it happens.","product-loom.c69366a198":"EU data centre","product-loom.1f732e78e6":"Larger or shared workloads.","product-loom.dea6d371cf":"close where supported","product-loom.596b81715f":"not only distant clusters","product-loom.f2515fffcc":"All of this sits behind Fabric.","product-loom.1c78b7ab8d":"The cognitive fabric behind Fabric","product-loom.b4cb110f9a":"One conversation,","product-loom.b259bd8dbd":"many forms of intelligence","product-loom.fd5987ccea":"Fabric is the workspace you open. Loom is the cognitive fabric underneath many of its answers. It can read, remember, search, classify, extract, rank, reason, solve, verify, consult specialists, and explain, with only the required parts active for your task.","product-loom.ac87775c5a":"Different questions leave different weaves behind them, each one a record of how that answer was put together.","product-loom.824d6eee51":"FABRIC IS THE SURFACE","product-loom.28b9c98933":"LOOM IS UNDERNEATH","product-loom.24b09368d1":"ONLY WHAT YOU NEED","product-loom.6bf68da7a3":"Fabric to Loom to Core","product-loom.d90ca1b08d":"surface, fabric, machine","product-loom.a010de5c61":"Fabric","product-loom.645b283301":"The workspace you open.","product-loom.9c1bb21ffd":"The cognitive fabric underneath.","product-loom.98302811bb":"The machine that runs it.","product-loom.2d3c1b5fff":"one surface","product-loom.08627806eb":"many forms of intelligence","product-loom.eb45f0c9cc":"You ask once, and the right cognition takes shape.","product-loom.7990ca57cd":"Ask once","product-loom.379edbac84":"The question stays simple for you. The weave underneath can be as small or as deep as the task requires.","product-loom.9ae8d97241":"What Fabric answers stand on","product-loom.5755cb799e":"spans and sources","product-loom.bad876b199":"The answer stays linked to what was seen","product-loom.19922afd66":"memory purposes","product-loom.e4c1e2f10a":"Current task, cases, facts, procedures","product-loom.c9c64e071d":"real","product-loom.b9cf6caf03":"solvers","product-loom.0ec5f2aebd":"Schedules and rules checked, not guessed","product-loom.2238839604":"visible","product-loom.7354fb826f":"checks","product-loom.dd76ca8fb7":"Sparse specialists, staged search, honest refusal","product-loom.3a12124675":"A new capability should not require retraining the entire mind. Loom can add a classifier, a solver, a reasoning mode, a memory procedure, a verifier, or a domain specialist as a new thread with its own tests and version. Existing threads do not need to be rewritten merely because the catalogue expanded. A new solver can handle a new problem class. A new specialist can add domain depth. A new perception stage can make another modality available. A new micro model can replace a broad generative path for a narrow task.","product-loom.8d331b0009":"The weave becomes richer without becoming one indivisible model.","product-loom.f351257270":"A system that can weave many capabilities must also know when none of them can support a defensible result. Loom can surface that state explicitly. It does not need to force every cognitive path into a fluent answer. The perception may be incomplete. The memory may not contain the required fact. Retrieval may return weak evidence. The constraints may conflict. The solver may return unknown. The specialists may disagree.","product-loom.527b953724":"The verifier may reject the result. Each of those states can terminate in a typed refusal with the cause named.","product-loom.53f3f2b451":"Loom's learned model families are not float models compressed at release time. The target bipolar representation participates in training. Gradients, updates, routing, adaptation, and control use binary, integer, or fixed-point machinery according to the component. Transformers, micro models, embeddings, routers, and domain-specialist models share the same native training posture. Specialisation can be trained and evaluated independently.","product-loom.04506015fe":"New specialists or models join additively rather than requiring a full catalogue retrain.","product-loom.e0ebd789e9":"Loom does not need to expose hidden model chain-of-thought. The trace records observable execution, not private reasoning text. It names the nodes activated, typed inputs and outputs, routing decisions, memories recalled, evidence retrieved, constraints evaluated, reasoning modes used, solver calls and statuses, specialists consulted, verification results, refusal reasons, and the final expression path.","product-loom.7c617944c9":"This is enough to inspect the cognitive operation without treating generated internal prose as proof.","product-loom.b077aa8e02":"The graph contract can support managed Fabric on the public Mesh and licensed operation on customer-controlled, edge, distributed, or air-gapped infrastructure. Typed graph, trace, memory, solver, refusal, and version contracts remain recognisable; direct product rights, available threads, and integrations vary by agreement and operating model.","product-loom.b6ba549d25":"External sockets are enabled or sealed according to policy, but the meaning of the weave does not change with its location.","product-loom.1ce16fe946":"A fluent answer is not automatically a supported answer. Loom can compare the result with the evidence, constraints, rules, checker results, and the checks used by the task. Where a formal certificate is available, the system can attach it.","product-loom.85cedaed99":"Where it can only reason its way there rather than prove it, it says so instead.","product-loom.c0d11e40d9":"The compile","product-loom.d600a63137":"A question becomes","product-loom.e63cd08b40":"a working graph","product-loom.0927f4471b":"One request is classified, planned, and compiled into exactly the graph it needs: perception, memory, retrieval, solvers, specialists, verification. Only the required nodes wake up, and the emitted trace names every one of them with its version.","product-loom.a7e638b54c":"only what is needed wakes","product-loom.d3d78b3704":"every node named in the trace","product-loom.abfa32a842":"Four questions","product-loom.f172fb16e2":"Different questions,","product-loom.be516853ea":"different thinking","product-loom.3f37aed941":"Finding a date, planning a schedule, checking a rule, and explaining a decision are different kinds of thinking. Each one wakes a different pattern of helpers behind the scenes. You just ask; the right pattern forms on its own.","product-loom.ffd3cf0880":"you just ask","product-loom.3261c460cd":"the right pattern forms","product-loom.72a44afdf3":"Compile steps","product-loom.e0647adaff":"PIPELINE","product-loom.b6b29b995f":"Task read for structure, modality, and domain.","product-loom.2f1d8fc4e4":"Graph shape resolved for the request.","product-loom.83a9e49fc4":"Bind","product-loom.1487c16fdb":"Nodes bound to versions and an execution budget.","product-loom.2369d1f6ae":"Trace stamped before the graph executes.","product-loom.a35129918f":"typed at every edge","product-loom.5fa6703151":"nothing implicit","product-loom.banner.business.composition.eyebrow":"The operating model","product-loom.banner.business.composition.headlineLead":"One request can hide","product-loom.banner.business.composition.headlineAccent":"four different responsibilities.","product-loom.banner.business.composition.body":"A business question may require perception, memory, retrieval, calculation, policy checks, or verification. Loom composes only the required threads and records the path, so ownership does not disappear behind a single text endpoint.","product-loom.banner.business.composition.instrumentHeader":"From request to record","product-loom.banner.business.composition.instrumentTag":"VISIBLE","product-loom.banner.business.composition.items.input.label":"Business request","product-loom.banner.business.composition.items.input.status":"received","product-loom.banner.business.composition.items.shape.label":"Task shape","product-loom.banner.business.composition.items.shape.status":"classified","product-loom.banner.business.composition.items.threads.label":"Required threads","product-loom.banner.business.composition.items.threads.status":"selected","product-loom.banner.business.composition.items.record.label":"Execution record","product-loom.banner.business.composition.items.record.status":"attached","product-loom.banner.business.composition.verdictLabel":"Operating consequence","product-loom.banner.business.composition.verdictValue":"Distinct work stays visible","product-loom.banner.business.composition.footer":"The task selects the cognition; the trace preserves the path.","product-loom.banner.business.languageBoundary.eyebrow":"The decision boundary","product-loom.banner.business.languageBoundary.headlineLead":"Language has a role.","product-loom.banner.business.languageBoundary.headlineAccent":"It does not own every decision.","product-loom.banner.business.languageBoundary.body":"Loom uses bipolar language components for contextual language work and generation, then uses constraint learning, solvers or verifiers when the task requires an exact status or a support check. Fluent expression stays separate from the evidence or constraint that justifies it.","product-loom.banner.business.languageBoundary.instrumentHeader":"Work matched to its contract","product-loom.banner.business.languageBoundary.instrumentTag":"SEPARATE","product-loom.banner.business.languageBoundary.compare.left":"Language work","product-loom.banner.business.languageBoundary.compare.right":"Decision work","product-loom.banner.business.languageBoundary.items.language.label":"Context and composition","product-loom.banner.business.languageBoundary.items.language.status":"model","product-loom.banner.business.languageBoundary.items.generation.label":"Final expression","product-loom.banner.business.languageBoundary.items.generation.status":"generated","product-loom.banner.business.languageBoundary.items.solver.label":"Constraints and optimisation","product-loom.banner.business.languageBoundary.items.solver.status":"solved","product-loom.banner.business.languageBoundary.items.verifier.label":"Evidence and rule checks","product-loom.banner.business.languageBoundary.items.verifier.status":"verified","product-loom.banner.business.languageBoundary.verdictLabel":"Decision control","product-loom.banner.business.languageBoundary.verdictValue":"The right mechanism owns the claim","product-loom.banner.business.languageBoundary.footer":"Generation expresses the result; it does not replace the check.","product-loom.banner.business.growth.eyebrow":"The growth contract","product-loom.banner.business.growth.headlineLead":"The catalogue can grow.","product-loom.banner.business.growth.headlineAccent":"Each query stays bounded.","product-loom.banner.business.growth.body":"A new specialist, solver, verifier, or learned component joins as a separately tested and versioned thread. Existing capabilities do not need a full retrain, and a request still activates only the path it needs.","product-loom.banner.business.growth.instrumentHeader":"Add one capability","product-loom.banner.business.growth.instrumentTag":"ADDITIVE","product-loom.banner.business.growth.items.capability.label":"New capability","product-loom.banner.business.growth.items.capability.status":"defined","product-loom.banner.business.growth.items.tests.label":"Evaluation surface","product-loom.banner.business.growth.items.tests.status":"owned","product-loom.banner.business.growth.items.version.label":"Independent version","product-loom.banner.business.growth.items.version.status":"pinned","product-loom.banner.business.growth.items.catalogue.label":"Existing catalogue","product-loom.banner.business.growth.items.catalogue.status":"unchanged","product-loom.banner.business.growth.verdictLabel":"Scale effect","product-loom.banner.business.growth.verdictValue":"More capability without one larger monolith","product-loom.banner.business.growth.footer":"Catalogue size and active query cost remain separate questions.","product-loom.banner.business.accountability.eyebrow":"The review handover","product-loom.banner.business.accountability.headlineLead":"A defensible answer includes","product-loom.banner.business.accountability.headlineAccent":"the path and its limits.","product-loom.banner.business.accountability.body":"Loom records observable execution rather than private chain-of-thought. A reviewer can see which path ran, what evidence and constraints shaped it, which checks completed, and why the system stopped when it could not support a result.","product-loom.banner.business.accountability.instrumentHeader":"Review receipt","product-loom.banner.business.accountability.instrumentTag":"GOVERNED","product-loom.banner.business.accountability.items.path.label":"Active cognitive path","product-loom.banner.business.accountability.items.path.status":"recorded","product-loom.banner.business.accountability.items.evidence.label":"Evidence and memory","product-loom.banner.business.accountability.items.evidence.status":"referenced","product-loom.banner.business.accountability.items.checks.label":"Rules and checks","product-loom.banner.business.accountability.items.checks.status":"named","product-loom.banner.business.accountability.items.stop.label":"Refusal or limit","product-loom.banner.business.accountability.items.stop.status":"explicit","product-loom.banner.business.accountability.verdictLabel":"Review state","product-loom.banner.business.accountability.verdictValue":"Inspect, correct, or stop at the owning thread","product-loom.banner.business.accountability.footer":"The record supports review without presenting hidden reasoning text as proof.","product-loom.banner.business.recognition.instrumentTag":"MISMATCH","product-loom.banner.business.recognition.items.date.status":"paraphrased","product-loom.banner.business.recognition.items.constraint.status":"missing","product-loom.banner.business.recognition.items.sat.status":"not solved","product-loom.banner.business.recognition.items.source.status":"not attributable","product-loom.banner.business.recognition.verdictLabel":"Diagnosis","product-loom.banner.technical.contract.eyebrow":"The thread contract","product-loom.banner.technical.contract.headlineLead":"The graph starts with contracts.","product-loom.banner.technical.contract.headlineAccent":"Not callbacks and prompt text.","product-loom.banner.technical.contract.body":"Every cognitive thread declares its input and output shapes, capability signature, version, cost model, determinism contract, failure modes, and evidence surface. Typed edges keep intermediate state inspectable instead of collapsing it into prose.","product-loom.banner.technical.contract.instrumentHeader":"One thread at compile time","product-loom.banner.technical.contract.instrumentTag":"TYPED","product-loom.banner.technical.contract.items.input.label":"Input shape","product-loom.banner.technical.contract.items.input.status":"bound","product-loom.banner.technical.contract.items.capability.label":"Capability signature","product-loom.banner.technical.contract.items.capability.status":"declared","product-loom.banner.technical.contract.items.output.label":"Output shape","product-loom.banner.technical.contract.items.output.status":"typed","product-loom.banner.technical.contract.items.failure.label":"Failure surface","product-loom.banner.technical.contract.items.failure.status":"exposed","product-loom.banner.technical.contract.verdictLabel":"Graph invariant","product-loom.banner.technical.contract.verdictValue":"No untyped prose between nodes","product-loom.banner.technical.contract.footer":"The planner can only bind a thread whose contract fits the task.","product-loom.banner.technical.perception.eyebrow":"The perception boundary","product-loom.banner.technical.perception.headlineLead":"Perception stays addressable.","product-loom.banner.technical.perception.headlineAccent":"Typed facts reach the next node.","product-loom.banner.technical.perception.body":"Text, documents, images, audio, video, and structured data can enter through different perception paths. Their outputs remain addressable to memory, reasoning, and verification instead of becoming disposable prompt preparation.","product-loom.banner.technical.perception.instrumentHeader":"Perception outputs","product-loom.banner.technical.perception.instrumentTag":"ADDRESSABLE","product-loom.banner.technical.perception.items.text.label":"Text and documents","product-loom.banner.technical.perception.items.text.status":"typed","product-loom.banner.technical.perception.items.visual.label":"Images and video","product-loom.banner.technical.perception.items.visual.status":"addressable","product-loom.banner.technical.perception.items.audio.label":"Audio","product-loom.banner.technical.perception.items.audio.status":"addressable","product-loom.banner.technical.perception.items.structured.label":"Structured data","product-loom.banner.technical.perception.items.structured.status":"typed","product-loom.banner.technical.perception.verdictLabel":"Downstream contract","product-loom.banner.technical.perception.verdictValue":"The observed input remains identifiable","product-loom.banner.technical.perception.footer":"Perception is a graph node, not a hidden preprocessing step.","product-loom.banner.technical.search.eyebrow":"The retrieval path","product-loom.banner.technical.search.headlineLead":"Search is a graph.","product-loom.banner.technical.search.headlineAccent":"Each judgement stays separate.","product-loom.banner.technical.search.body":"Encoding, candidate retrieval, relevance scoring, ranking, reranking, policy filtering, provenance filtering, and support verification remain independent nodes. The trace can identify which judgement changed the final source set.","product-loom.banner.technical.search.instrumentHeader":"A bounded search weave","product-loom.banner.technical.search.instrumentTag":"STAGED","product-loom.banner.technical.search.items.encode.label":"Encode the query","product-loom.banner.technical.search.items.encode.status":"represented","product-loom.banner.technical.search.items.retrieve.label":"Retrieve candidates","product-loom.banner.technical.search.items.retrieve.status":"bounded","product-loom.banner.technical.search.items.rank.label":"Rank and rerank","product-loom.banner.technical.search.items.rank.status":"ordered","product-loom.banner.technical.search.items.verify.label":"Verify support","product-loom.banner.technical.search.items.verify.status":"checked","product-loom.banner.technical.search.verdictLabel":"Inspection point","product-loom.banner.technical.search.verdictValue":"A bad source has an owning stage","product-loom.banner.technical.search.footer":"One search result can be replayed as a sequence of typed decisions.","product-loom.banner.technical.knowledge.eyebrow":"The state boundary","product-loom.banner.technical.knowledge.headlineLead":"A solver status is not","product-loom.banner.technical.knowledge.headlineAccent":"a knowledge-base update.","product-loom.banner.technical.knowledge.body":"Loom keeps solver dispatch, returned status, stored facts, contradiction detection, and belief revision as separate operations. That makes it possible to inspect where a claim entered the graph and how the knowledge state changed.","product-loom.banner.technical.knowledge.instrumentHeader":"Status versus state","product-loom.banner.technical.knowledge.instrumentTag":"SEPARATE","product-loom.banner.technical.knowledge.compare.left":"Solver contract","product-loom.banner.technical.knowledge.compare.right":"Knowledge contract","product-loom.banner.technical.knowledge.items.solverInput.label":"Problem and guarantees","product-loom.banner.technical.knowledge.items.solverInput.status":"typed","product-loom.banner.technical.knowledge.items.solverStatus.label":"Returned solver status","product-loom.banner.technical.knowledge.items.solverStatus.status":"reported","product-loom.banner.technical.knowledge.items.fact.label":"Fact or relation","product-loom.banner.technical.knowledge.items.fact.status":"stored","product-loom.banner.technical.knowledge.items.revision.label":"Belief revision","product-loom.banner.technical.knowledge.items.revision.status":"versioned","product-loom.banner.technical.knowledge.verdictLabel":"State rule","product-loom.banner.technical.knowledge.verdictValue":"Execution results do not mutate knowledge implicitly","product-loom.banner.technical.knowledge.footer":"A trace can distinguish a solved problem from a revised belief base.","product-loom.banner.technical.routingMerge.eyebrow":"The composition rule","product-loom.banner.technical.routingMerge.headlineLead":"Routing selects contributors.","product-loom.banner.technical.routingMerge.headlineAccent":"Merge rules decide the result.","product-loom.banner.technical.routingMerge.body":"The router narrows candidate threads and specialists, but selection alone does not define composition. Lists may be unioned or reranked, constraints combined or rejected as inconsistent, and specialist outputs sent through quorum, arbitration, or verification.","product-loom.banner.technical.routingMerge.instrumentHeader":"Selection to composition","product-loom.banner.technical.routingMerge.instrumentTag":"EXPLICIT","product-loom.banner.technical.routingMerge.items.signature.label":"Task or domain signature","product-loom.banner.technical.routingMerge.items.signature.status":"matched","product-loom.banner.technical.routingMerge.items.frontier.label":"Active frontier","product-loom.banner.technical.routingMerge.items.frontier.status":"selected","product-loom.banner.technical.routingMerge.items.merge.label":"Merge operation","product-loom.banner.technical.routingMerge.items.merge.status":"declared","product-loom.banner.technical.routingMerge.items.verifier.label":"Verifier or arbiter","product-loom.banner.technical.routingMerge.items.verifier.status":"applied","product-loom.banner.technical.routingMerge.verdictLabel":"Composition invariant","product-loom.banner.technical.routingMerge.verdictValue":"Selection and merge remain independently inspectable","product-loom.banner.technical.routingMerge.footer":"No average silently stands in for a task-specific merge rule.","product-loom.banner.technical.replay.eyebrow":"The replay contract","product-loom.banner.technical.replay.headlineLead":"Replay requires the weave.","product-loom.banner.technical.replay.headlineAccent":"Versions alone are incomplete.","product-loom.banner.technical.replay.body":"Loom records the complete graph, canonical execution order, routing decisions, model and specialist artefacts, memory and solver state, constraints, limits, dependency outputs, and external evidence used by the run.","product-loom.banner.technical.replay.instrumentHeader":"Captured replay state","product-loom.banner.technical.replay.instrumentTag":"COMPLETE","product-loom.banner.technical.replay.items.graph.label":"Graph and canonical order","product-loom.banner.technical.replay.items.graph.status":"recorded","product-loom.banner.technical.replay.items.artifacts.label":"Models, specialists, and solvers","product-loom.banner.technical.replay.items.artifacts.status":"pinned","product-loom.banner.technical.replay.items.state.label":"Memory and runtime state","product-loom.banner.technical.replay.items.state.status":"captured","product-loom.banner.technical.replay.items.evidence.label":"Evidence and dependencies","product-loom.banner.technical.replay.items.evidence.status":"retained","product-loom.banner.technical.replay.verdictLabel":"Replay condition","product-loom.banner.technical.replay.verdictValue":"The run can be reconstructed from what actually executed","product-loom.banner.technical.replay.footer":"Live replay can call external sources again and expose where the world changed.","product-loom.banner.technical.boundaries.eyebrow":"The product boundary","product-loom.banner.technical.boundaries.headlineLead":"Loom composes cognition.","product-loom.banner.technical.boundaries.headlineAccent":"Nexus composes agents.","product-loom.banner.technical.boundaries.body":"Both products can use perception, memory, reasoning, solvers, knowledge, and verification. Loom represents them as nodes in one cognitive graph producing one intelligence result. Nexus places capabilities inside agents with identity, lifecycle, authority, tasks, and communication.","product-loom.banner.technical.boundaries.instrumentHeader":"Two composition layers","product-loom.banner.technical.boundaries.instrumentTag":"BOUNDARY","product-loom.banner.technical.boundaries.compare.left":"Loom","product-loom.banner.technical.boundaries.compare.right":"Nexus","product-loom.banner.technical.boundaries.items.loom.label":"Typed cognitive graph","product-loom.banner.technical.boundaries.items.loom.status":"cognition","product-loom.banner.technical.boundaries.items.result.label":"One intelligence result","product-loom.banner.technical.boundaries.items.result.status":"returned","product-loom.banner.technical.boundaries.items.nexus.label":"Agent lifecycle and tasks","product-loom.banner.technical.boundaries.items.nexus.status":"organisation","product-loom.banner.technical.boundaries.items.authority.label":"Identity and authority","product-loom.banner.technical.boundaries.items.authority.status":"governed","product-loom.banner.technical.boundaries.verdictLabel":"Boundary test","product-loom.banner.technical.boundaries.verdictValue":"Cognitive composition versus organisational action","product-loom.banner.technical.boundaries.footer":"Core executes the operations beneath both layers; Loom determines the cognitive graph.","product-loom.banner.technical.recognition.instrumentTag":"WRONG ABSTRACTION","product-loom.banner.technical.recognition.items.classification.status":"generative","product-loom.banner.technical.recognition.items.match.status":"approximate","product-loom.banner.technical.recognition.items.arithmetic.status":"unsupported","product-loom.banner.technical.recognition.items.schedule.status":"unchecked","product-loom.banner.technical.recognition.verdictLabel":"Architecture finding","product-loom.banner.consumer.jobs.eyebrow":"What changes behind the answer","product-loom.banner.consumer.jobs.headlineLead":"You ask once.","product-loom.banner.consumer.jobs.headlineAccent":"The job still changes.","product-loom.banner.consumer.jobs.body":"Finding a date, checking a schedule, searching for support, and explaining a rule are different jobs. Loom recognises the job, wakes the helpers it needs, and leaves the others quiet.","product-loom.banner.consumer.jobs.instrumentHeader":"One question, one fitting path","product-loom.banner.consumer.jobs.instrumentTag":"MATCHED","product-loom.banner.consumer.jobs.items.question.label":"Your question","product-loom.banner.consumer.jobs.items.question.status":"asked","product-loom.banner.consumer.jobs.items.job.label":"Kind of job","product-loom.banner.consumer.jobs.items.job.status":"recognised","product-loom.banner.consumer.jobs.items.helpers.label":"Relevant helpers","product-loom.banner.consumer.jobs.items.helpers.status":"selected","product-loom.banner.consumer.jobs.items.answer.label":"Answer and useful record","product-loom.banner.consumer.jobs.items.answer.status":"returned","product-loom.banner.consumer.jobs.verdictLabel":"Practical effect","product-loom.banner.consumer.jobs.verdictValue":"Small jobs stay small; hard jobs get more help","product-loom.banner.consumer.jobs.footer":"You do not have to choose the machinery before you ask.","product-loom.banner.consumer.memory.eyebrow":"What memory should do","product-loom.banner.consumer.memory.headlineLead":"Useful memory is selective.","product-loom.banner.consumer.memory.headlineAccent":"The whole chat does not follow.","product-loom.banner.consumer.memory.body":"Loom can keep the current task in view, recall a relevant earlier case, use facts from your knowledge base, or follow a stored procedure. Each helper receives the memory that belongs to its job.","product-loom.banner.consumer.memory.instrumentHeader":"Memory for this job","product-loom.banner.consumer.memory.instrumentTag":"SCOPED","product-loom.banner.consumer.memory.items.task.label":"Current task","product-loom.banner.consumer.memory.items.task.status":"in scope","product-loom.banner.consumer.memory.items.case.label":"Relevant earlier case","product-loom.banner.consumer.memory.items.case.status":"when useful","product-loom.banner.consumer.memory.items.facts.label":"Known facts","product-loom.banner.consumer.memory.items.facts.status":"available","product-loom.banner.consumer.memory.items.procedure.label":"Stored procedure","product-loom.banner.consumer.memory.items.procedure.status":"when needed","product-loom.banner.consumer.memory.verdictLabel":"Memory rule","product-loom.banner.consumer.memory.verdictValue":"Relevant context reaches the helper that needs it","product-loom.banner.consumer.memory.footer":"Memory is organised by purpose rather than kept as one endless transcript.","product-loom.banner.consumer.support.eyebrow":"The confidence test","product-loom.banner.consumer.support.headlineLead":"A smooth answer sounds right.","product-loom.banner.consumer.support.headlineAccent":"Support can be inspected.","product-loom.banner.consumer.support.body":"Loom can compare an answer with the evidence, rules, constraints, and checker results used for the task. When it has a formal certificate, it can attach one. When it cannot prove the result, it says what kind of support it does have.","product-loom.banner.consumer.support.instrumentHeader":"What confidence is based on","product-loom.banner.consumer.support.instrumentTag":"CHECKED","product-loom.banner.consumer.support.compare.left":"Sounding certain","product-loom.banner.consumer.support.compare.right":"Being supported","product-loom.banner.consumer.support.items.fluent.label":"Fluent wording","product-loom.banner.consumer.support.items.fluent.status":"unverified","product-loom.banner.consumer.support.items.certain.label":"Confident tone","product-loom.banner.consumer.support.items.certain.status":"unclear basis","product-loom.banner.consumer.support.items.evidence.label":"Evidence and rules","product-loom.banner.consumer.support.items.evidence.status":"referenced","product-loom.banner.consumer.support.items.check.label":"Relevant check","product-loom.banner.consumer.support.items.check.status":"passed or failed","product-loom.banner.consumer.support.verdictLabel":"Trust boundary","product-loom.banner.consumer.support.verdictValue":"Support is visible; tone is not evidence","product-loom.banner.consumer.support.footer":"A useful answer lets you see what was checked and what remains uncertain.","product-loom.banner.consumer.recordOrStop.eyebrow":"The honest outcome","product-loom.banner.consumer.recordOrStop.headlineLead":"The useful outcome is","product-loom.banner.consumer.recordOrStop.headlineAccent":"a record or a clear stop.","product-loom.banner.consumer.recordOrStop.body":"When Loom can support an answer, it can show what it read, which method it used, and which checks passed. When the input is unclear, the evidence is weak, the rules conflict, or the checker cannot decide, it can name that limit instead.","product-loom.banner.consumer.recordOrStop.instrumentHeader":"What comes back","product-loom.banner.consumer.recordOrStop.instrumentTag":"HONEST","product-loom.banner.consumer.recordOrStop.items.input.label":"Input and sources","product-loom.banner.consumer.recordOrStop.items.input.status":"referenced","product-loom.banner.consumer.recordOrStop.items.method.label":"Method or helper","product-loom.banner.consumer.recordOrStop.items.method.status":"named","product-loom.banner.consumer.recordOrStop.items.checks.label":"Checks","product-loom.banner.consumer.recordOrStop.items.checks.status":"pass or fail","product-loom.banner.consumer.recordOrStop.items.limit.label":"Limit or disagreement","product-loom.banner.consumer.recordOrStop.items.limit.status":"stated","product-loom.banner.consumer.recordOrStop.verdictLabel":"Result boundary","product-loom.banner.consumer.recordOrStop.verdictValue":"A supported answer, or a reason not to pretend","product-loom.banner.consumer.recordOrStop.footer":"Stopping clearly is part of doing the job well.","product-loom.banner.consumer.recognition.instrumentTag":"ONE ASSISTANT","product-loom.banner.consumer.recognition.items.date.status":"not extracted","product-loom.banner.consumer.recognition.items.schedule.status":"not checked","product-loom.banner.consumer.recognition.items.source.status":"misranked","product-loom.banner.consumer.recognition.items.rules.status":"conflict hidden","product-loom.banner.consumer.recognition.verdictLabel":"What went wrong","product-loom.ec8c2ac856":"the limits of one model","product-loom.2de9d21c51":"See the thread contract","product-loom.e73b199991":"The weave","product-loom.e9b39edcd9":"PER TASK","product-loom.bb12e8aaae":"Threads","product-loom.19ec974bcb":"Perception, memory, models, reasoning, solvers","product-loom.d1f8e5ea93":"Selection","product-loom.f96336627f":"Only the responsibilities the task needs","product-loom.1c54132b8e":"Record","product-loom.22975d460b":"Nodes and artefacts, never private thoughts","product-loom.e79ff8e145":"One question in, one accountable graph out.","product-loom.0db9a297e8":"the cognition it deserves","product-loom.10f54c814a":"Read the graph contract","product-loom.c82ea674f0":"The contract","product-loom.1b4306be38":"TYPED","product-loom.260f7a8cd4":"Node","product-loom.650387f7ed":"Task, modality, domain, evidence, budget","product-loom.40f65b4ff4":"Different reasoning, different engines","product-loom.eb485ae62d":"No defensible path, no fabricated answer","product-loom.f1a80e6485":"Every node signs the same contract.","product-loom.d7e169e12b":"and the right cognition takes shape","product-loom.a4fb970cd2":"See where it runs","product-loom.5cf8744d5c":"Underneath","product-loom.5929e55cd5":"The relevant few active, not all of them","product-loom.2eb56be3c2":"Sources","product-loom.d1b4902383":"An exact answer with an exact source","product-loom.0e1a829245":"Honesty","product-loom.59cb5de8e2":"Weak evidence and conflicts stay visible","product-loom.2bc04cfaa3":"You ask once. The weave does the rest.","product-loom.6d2466db6f":"Operations below","product-loom.1b3637d383":"cognition above","product-loom.6f06f33fbe":"Cognition above","product-loom.7bbd479514":"operations below","product-loom-secondary-business.hiddenJobs.left":"A fluent answer can conceal an extraction, a calculation, a relevance judgement or a constraint check. When those jobs share one opaque output, a team cannot assess the right thing or locate the right failure.","product-loom-secondary-business.hiddenJobs.right":"Loom gives each thread a responsibility, typed input and output, version, and evaluation surface. The answer may remain straightforward for its reader, while the operating record shows which work established it and where a correction belongs.","product-loom-secondary-business.taskWeave.left":"Ticket triage may activate one classifier, while locating a cancellation date calls for document perception and exact span extraction. Neither task benefits from carrying a fixed, heavyweight pipeline behind a simple request.","product-loom-secondary-business.taskWeave.right":"A policy search can combine embedding, retrieval, scoring, ranking and reranking. A cross-domain explanation may add memory, causal reasoning, evidence checks, specialists and solvers before a language model presents the result.","product-loom-secondary-business.perception.left":"Perception turns text, documents, images, audio, video and structured data into typed observations. It can retain a span, region, event, entity, field or signal instead of reducing the input to an untraceable prompt summary.","product-loom-secondary-business.perception.right":"Later threads work from those observations rather than from a recollection of them. A reasoning or verification step can point to the precise item it used, which keeps the source of a conclusion available for inspection.","product-loom-secondary-business.memory.left":"Working memory carries the active task, episodic memory preserves prior cases, semantic memory holds durable knowledge, and procedural memory keeps reusable methods. These are different responsibilities, not one ever-growing transcript.","product-loom-secondary-business.memory.right":"Access follows the work in progress. A solver receives the facts and constraints it needs, a specialist receives relevant cases and knowledge, and a classifier is not given unrelated history simply because it exists.","product-loom-secondary-business.search.left":"Binary embeddings and BitWeave retrieval keep the wide candidate path compact. Relevance scoring and ranking narrow the set, then reranking applies deeper comparison only to the frontier where that additional work is warranted.","product-loom-secondary-business.search.right":"Provenance, policy, recency and domain rules remain a declared eligibility stage, not an unexplained score adjustment. The final evidence stays tied to its source and location before another thread relies on it.","product-loom-secondary-business.solve.left":"A schedule, configuration or optimisation problem can be routed to the formal method that matches its structure. Loom keeps logic, constraints, temporal work, optimisation, verification and synthesis available as solver-backed cognitive paths.","product-loom-secondary-business.solve.right":"The result retains its actual status and artefact: an assignment, core, optimum, bound, schedule, proof, counterexample or unknown. A language model can make that result usable without first having to invent it.","product-loom-secondary-business.reasoning.left":"Different questions call for different forms of reasoning. Rules can yield a derivation, evidence can support ranked explanations, and interventions can expose causal dependencies without being compressed into one generic confidence score.","product-loom-secondary-business.reasoning.right":"A weave can combine temporal, spatial, probabilistic, decision and counterfactual reasoning where the work requires them. Each method remains named in the trace, so a reader can distinguish its contribution from the language that expresses it.","product-loom-secondary-business.experts.left":"The 528-domain-specialist catalogue gives Loom domain depth without treating every request as a committee meeting. Routing selects a bounded relevant set, typically four to eight specialists, and can reach twenty for complex cross-domain work.","product-loom-secondary-business.experts.right":"Selected specialists join the existing graph beside memory, retrieval, constraints, solvers and verification. Their activation is recorded, and the catalogue adds a specialised faculty rather than replacing the cognitive system around it.","product-loom-secondary-business.transformer.left":"Loom's native bipolar language components remain central to contextual understanding, composition and generation. They can receive exact spans, sourced evidence, scoped memory and formal artefacts through typed inputs instead of approximating every supporting faculty in their weights.","product-loom-secondary-business.transformer.right":"Solver outputs can arrive as assignments, cores, schedules, bounds or proofs, while specialists add domain findings and verification can reject a weak path. The language model reasons over that state and expresses the result without claiming to be the whole process.","product-loom-secondary-business.oneBit.left":"Classifiers, span models, embeddings, rankers, transformers, routers and specialists are trained for the native bipolar representation in which they run. Loom does not rely on a full-precision canonical model hidden behind an inference-only quantiser.","product-loom-secondary-business.oneBit.right":"Binary and integer training, adaptation, routing and control keep learned threads compact and deterministic on supported paths. Graphs, intervals, constraints and proofs remain in their own native symbolic forms rather than being forced into a neural representation.","product-loom-secondary-business.cost.left":"The full catalogue can remain deployed and ready without every request activating it. A ticket label may use one micro model, a policy search can use a focused retrieval path, and a constrained decision can call for a wider weave.","product-loom-secondary-business.cost.right":"Per-request work follows the nodes, models, evidence, solvers and specialists that actually run. Language-model computation is used where language work earns it, while exact and compact paths remain available for tasks that need less.","product-loom-secondary-business.additive.left":"A new model, perception operator, memory method, solver adapter, verifier or specialist enters with its own contract, routing signature, evaluation suite, version and release record. Unrelated learned models do not need retraining merely because the catalogue grows.","product-loom-secondary-business.additive.right":"Integration still has a defined scope: the capability is evaluated, registered, routed and exercised in dependent cognitive graphs. Existing artefacts stay pinned, while only paths that can select or consume the new faculty change.","product-loom-secondary-business.corrections.left":"The weave trace turns a wrong result into an accountable diagnosis. A parsing defect, weak ranking, inappropriate reasoning mode or incorrect solver input can be traced to the thread that held that responsibility.","product-loom-secondary-business.corrections.right":"A local repair can then address the responsible component while its neighbours remain pinned. If the language thread distorts a correct formal result, the correction belongs in expression, not in an indiscriminate system-wide retraining effort.","product-loom-secondary-business.trace.left":"The trace is an operational map of the cognitive path that actually ran. It records observable work and artefacts, not private model chain-of-thought, so its boundary is as important as the detail it includes.","product-loom-secondary-business.trace.right":"It names perception outputs, recalled memory, activated models, retrieval, reasoning modes, solver results, constraints, specialists, verification and the expression path. Versioned handoffs make that record suitable for inspection and replay.","product-loom-secondary-business.refusal.left":"A defensible result may not exist when perception is incomplete, evidence is weak, constraints conflict, a solver returns unknown or specialists disagree. Loom does not need to turn those conditions into a confident-sounding answer.","product-loom-secondary-business.refusal.right":"A typed refusal names the cause and the failed or missing input, making the stopping point visible to the operator. That gives the next action a basis: add evidence, revise the constraints, narrow the problem or inspect the rejected result.","product-loom-secondary-business.deploy.left":"Managed customers use Loom through Fabric within Dweve's European service boundary. Licensed deployments can run on-premises, at the edge, across Mesh compute or in an air-gapped environment, subject to the operating agreement.","product-loom-secondary-business.deploy.right":"Placement changes hardware, sockets, integrations and operational ownership, but not the cognitive contract. Typed observations, exact solver statuses, weave artefacts and deterministic replay retain the same meaning across the supported postures.","product-loom-secondary-business.stack.left":"Core executes Loom's learned and algorithmic operations, while Spindle provides governed, lossless knowledge and provenance. Loom composes perception, memory, models, retrieval, reasoning, solvers, verification and specialists into the cognitive path.","product-loom-secondary-business.stack.right":"Nexus places those capabilities in executable organisations with agents, tools, workflows, authority and people. Mesh supplies distributed compute, and Fabric is the surface where people ask, inspect and use the resulting work.","product-loom-secondary-technical.opening.first":"Compilation settles the work before execution begins. The planner identifies the task, modality, domain, evidence requirement and resource budget, then binds each required node to a versioned contract. Parallel branches are explicit, and a node stays absent when the request does not need it.","product-loom-secondary-technical.opening.second":"The emitted graph preserves both sides of the result: the useful artefact and the weave trace that names the activated nodes, routes, versions and typed outputs. A reviewer can inspect the graph that ran instead of inferring a hidden process from the fluency of its answer.","product-loom-secondary-technical.83f00d001f.first":"A contract preserves the native result of a cognitive operation. A span extractor returns offsets and labels; retrieval returns candidates, distances and provenance; a SAT solver returns a status and, where available, a model, core or certificate. Those artefacts remain structured when the next node receives them.","product-loom-secondary-technical.83f00d001f.second":"The planner can therefore distinguish a ranked list from a causal path, an accepted result from a failed verification, and a declared limit from a useful answer. A node also names its version, cost, deterministic behaviour, evidence surface and failure modes. Composition is based on those contracts, not on a later component guessing what a prose response meant.","product-loom-secondary-technical.569056e679.first":"The planner resolves the graph from the work in front of it. A bounded classification can stop at one model, while a document investigation can activate perception, memory, retrieval, reasoning and verification in parallel. Policy, required evidence and the execution budget constrain the shape before the graph runs.","product-loom-secondary-technical.569056e679.second":"Branches may join before a reasoning step, and several reasoning forms may operate over the same typed state. A solver can replace a predictive stage when the problem requires an exact method, while a verifier can stop the graph before expression. The resulting topology is inspectable because every fork, join and gate belongs to the compiled graph.","product-loom-secondary-technical.b0e7797c6e.first":"The common one-bit foundation does not turn every learned result into the same vector. Binary hypervectors support associative memory, retrieval, binding and parts of routing, while language models, classifiers and rankers retain the representations required by their own contracts. An adapter is explicit whenever one compatible learned artefact crosses into another family.","product-loom-secondary-technical.b0e7797c6e.second":"The numerical path stays deterministic through bipolar arithmetic, integer control and named conversions. Formal artefacts do not enter that path merely for uniformity: a formula, temporal network, interval, assignment or proof keeps its native structure. That preserves the information a solver or verifier needs to inspect rather than reducing it to an approximate learned representation.","product-loom-secondary-technical.15a87bb79a.first":"Each modality produces observations that downstream nodes can address directly. Text can retain spans, entities and relations; documents can retain layout, tables and metadata; images can retain regions and descriptors; audio and video can retain timestamps, signals and tracked events. Structured inputs keep fields, schemas, quality findings and temporal relationships rather than becoming an undifferentiated prompt.","product-loom-secondary-technical.15a87bb79a.second":"Those observations carry their source identity, coordinates, type and confidence through the graph. Memory can recall a particular observation, a reasoner can use the field it needs, and verification can point back to the perceived evidence. The perception fabric is broad because the inputs differ, but its output contract stays precise enough for later work to cite and challenge.","product-loom-secondary-technical.09d881bc32.first":"The four memory tiers have different jobs. Working memory holds bounded active state, episodic memory recalls prior cases with temporal context, semantic memory stores typed facts and relations, and procedural memory keeps reusable rules and decision structures. A query is routed to the tier and retrieval structure that fit its purpose instead of searching one long transcript.","product-loom-secondary-technical.09d881bc32.second":"Consolidation, decay, compaction, replay, scheduling and forgetting are part of the memory contract. Retrieval may use direct similarity, LSH, graph association, temporal indexes or knowledge queries, depending on the record being sought. Monotonic logical counters make internal evolution independent of wall-clock drift, so that evolution can also be replayed and inspected.","product-loom-secondary-technical.df0d7ce120.first":"A micro model owns a bounded statistical decision and the evaluation surface that judges it. It may classify a request, extract exact spans, score relevance, rank candidates, create an embedding or decide whether a deeper branch should run. Its narrow responsibility makes its input, output and failure cases easier to test without making the model an afterthought.","product-loom-secondary-technical.df0d7ce120.second":"Placement follows the graph rather than a fixed model catalogue. A micro model can run after perception, inside retrieval, beside a reasoning form, before a solver or as a gate for more expensive work. When a gate declines a branch, the trace can show that the branch was intentionally skipped and why, instead of leaving an unexplained absence in the result.","product-loom-secondary-technical.8121d57d92.first":"Search keeps candidate generation separate from deeper judgement. A binary embedder and BitWeave index create a compact, deterministic candidate path using Hamming distance and XNOR plus population count. Scoring and ranking then establish an ordered frontier before reranking spends more work on the candidates that remain.","product-loom-secondary-technical.8121d57d92.second":"Policy, recency, provenance and domain constraints are their own visible stage, so a highly ranked item can still be excluded for a declared reason. Verification receives a typed evidence set with source identity, representation version, distance, ranking state and support outcome. If the selected evidence cannot support the answer, the graph can verify that failure or return a typed refusal.","product-loom-secondary-technical.ee86b8640c.first":"A bipolar transformer is a learned thread with typed ports, not an authority that reconstructs every fact from a prompt. It can receive exact spans, evidence, knowledge state, solver assignments and specialist findings in their declared forms. Its own weights, activations, attention and outputs follow the native one-bit and integer contract.","product-loom-secondary-technical.ee86b8640c.second":"The transformer remains central where contextual language, composition or generation is the right method. But exact extraction, formal results and verification outcomes keep their own status when they enter the graph, and the final expression can still be constrained or rejected. A language model therefore works with the other faculties rather than silently overriding them.","product-loom-secondary-technical.582bb19085.first":"Reasoning dispatch names the form of inference instead of hiding every inference behind one confidence score. Deductive and abductive paths work over rules and explanations; causal, temporal and spatial paths use their own graphs and relations; probabilistic and decision paths retain distributions, utilities and policies. Counterfactual, fuzzy, default, modal and deontic forms also keep their distinct semantics.","product-loom-secondary-technical.582bb19085.second":"The selected mode determines what a later node can inspect. A causal path can expose interventions or independence results, a temporal path can expose a consistent network, and a deontic path can expose obligations and permissions. Several forms may contribute to one graph, but the trace records their versions, artefacts and limits separately so their roles cannot be mistaken for one another.","product-loom-secondary-technical.c033e9b084.first":"The solver registry spans formal logic, constraint programming, optimisation, mathematics, dynamics, verification, planning and synthesis. Its regions cover satisfiability and theories, schedules and temporal networks, graphs and automata, model checking, probabilistic inference, symbolic algebra, differential systems and program search. The cognitive task is to identify the structure that needs one of those methods, not to run the full registry by default.","product-loom-secondary-technical.c033e9b084.second":"Each selected method returns the artefact it actually establishes: a decision, assignment, model, optimum, bound, schedule, strategy, proof, core, counterexample, program or honest unknown. Native Rust methods and integrated backends remain identified in the weave trace. That distinction matters because the result type, available certificate and operational limit belong to the solver region that produced them.","product-loom-secondary-technical.69445ca602.first":"Dispatch considers the problem theory, variable domains, objective, constraint density, requested artefacts, warm-start state, available backends, memory ceiling and execution budget. It can choose one solver, sequence methods, compare independent results or use one method to validate another. Candidate selection and tie-breaking are deterministic, so the portfolio decision can be reconstructed from the pinned state.","product-loom-secondary-technical.69445ca602.second":"The outcome leaves the node unchanged. SAT, UNSAT, optimal, infeasible, unbounded, safe, disproved, dynamically controllable and unknown are different statuses with different next actions. Models, cores, bounds, counterexamples, strategies, schedules and certificates remain attached to the property established, while an unknown result changes method or produces a refusal rather than a plausible substitute.","product-loom-secondary-technical.4c06030dd5.first":"Knowledge operations keep facts, rules, entities, relations and assumptions explicit enough to revise. Contradiction detection can find direct, implicit and temporal conflicts, after which the graph can expand, contract or revise the belief base. The trace preserves the conflict and the revision instead of silently replacing an earlier claim.","product-loom-secondary-technical.4c06030dd5.second":"Proof checking is narrower than explanation. Where a selected method emits a proof object, it can be checked, and AION can attach an independently checkable certificate on supported paths. Other cognitive results remain inferences, with the mechanism and limit named in the trace; the system does not present every useful conclusion as a formal proof.","product-loom-secondary-technical.eca832ae80.first":"The specialist catalogue contains 528 separately versioned and evaluated specialists. Compact signatures, approximate neighbour retrieval, Permuted Agreement Popcount and per-specialist gate subsets reduce that catalogue before deterministic top-k routing selects the active set. A typical specialist path activates roughly four to eight top-level domain specialists, while the selected number can still follow the task.","product-loom-secondary-technical.eca832ae80.second":"Active specialists can run concurrently when the graph permits it, then contribute findings to constraints, solvers, verification or final synthesis. Their domain depth is additive to the weave. Perception, memory, retrieval, reasoning and formal methods remain independent faculties, so a specialist result is a typed contribution to the graph rather than a replacement for the rest of cognition.","product-loom-secondary-technical.7f16a43919.first":"Loom routes at distinct cognitive boundaries: task to graph, modality to perception, memory query to tier, search query to ranking path, problem structure to reasoning form, theory to solver portfolio, domain signature to specialists, and evidence state to verification or refusal. Each decision works with its own candidate type and scoring method. One universal router would lose the meaning of those different choices.","product-loom-secondary-technical.7f16a43919.second":"A routing record can retain the full candidate set, scores, thresholds, rejected alternatives, canonical tie-break order, selected path and component versions. That makes a route observable as a decision rather than a hidden preference. With the same complete pinned state and canonical ordering, the same record selects the same path, which is the deterministic condition required for replay.","product-loom-secondary-technical.297b00b3d0.first":"Merge rules depend on the artefacts being combined. Candidate lists can be unioned, intersected or reranked; constraint sets can be combined or declared inconsistent; specialist findings can require quorum, arbitration or verification. Reasoning modes can retain competing hypotheses when collapsing them would hide a real disagreement.","product-loom-secondary-technical.297b00b3d0.second":"Where the contract requires exactness, an exact solver result can dominate a predictive suggestion. An invalid combination is rejected because its types do not unify, rather than being averaged into fluent language. The merge operator is declared in the graph, allowing a reviewer to see why one result prevailed and why an incompatible input did not pass onward.","product-loom-secondary-technical.04506015fe.first":"The deployed representation participates in training. Transformers, micro models, embeddings, routers and specialists use binary, integer or fixed-point machinery appropriate to the component, without a floating-point master model being quantised only for release. The resulting one-bit artefact is trained, evaluated and sealed for the numerical contract it will execute.","product-loom-secondary-technical.04506015fe.second":"Each learned family can be trained and evaluated on its own responsibility and measurement surface. That allows a new specialist or model to enter the catalogue additively, with its own version and contract, instead of forcing a full catalogue retrain. Independent training also keeps a later change local enough to compare, register and replay.","product-loom-secondary-technical.08374dcb29.first":"A weave pins the identities of the components that made its result: model artefacts, memory snapshots, solver builds, constraint sets, specialist definitions, policy bundles and knowledge state. The manifest records the versions that actually ran, not a catalogue of versions available at the time. A reviewer can therefore reconstruct the graph from its concrete boundaries.","product-loom-secondary-technical.08374dcb29.second":"An upgrade changes a named boundary and leaves an explicit diff. Replacing a ranker, for example, does not silently change the identity of the perception, memory, solver or policy state around it. Earlier answers retain the manifest that produced them, while a later weave can show exactly which component version changed and what must be compared.","product-loom-secondary-technical.5133b955da.first":"Replay pins controlled state: input bytes, preprocessing, encoders, memory snapshots, counters, versions, indices, knowledge, constraints, solver options, routing thresholds, candidate sets, tie-breaks, Core build and hardware target. Learned, symbolic and numerical paths use canonical ordering without a random seed. The manifest records the dependencies that gave the run its meaning.","product-loom-secondary-technical.5133b955da.second":"Captured replay reuses the recorded external evidence and dependency outputs, reproducing the controlled weave. Live replay calls mutable sources again and identifies the first node and artefact whose content identity changed. The deterministic core remains fixed, so a changed external boundary is visible as an external change rather than mislabelled as model drift.","product-loom-secondary-technical.7c617944c9.first":"The weave trace records observable execution in the order it occurred. It can name activated nodes, typed inputs and outputs, recalled memory, retrieved evidence, routing decisions, reasoning modes, solver calls and statuses, specialist contributions, verification outcomes, refusal reasons and the final expression path. Artefact identifiers connect each event to something a reviewer can inspect.","product-loom-secondary-technical.7c617944c9.second":"The trace deliberately does not claim to expose private model chain-of-thought. Generated internal prose is not treated as evidence, while source spans, assignments, cores, bounds and verification results remain available at the boundary where they matter. That gives a reviewer an operational record of what Loom did without turning a hidden thought stream into a false proof surface.","product-loom-secondary-technical.78b66810bd.first":"Loom composes cognitive faculties into one task-specific graph and returns one intelligence result with its trace. Nexus composes agents with identity, lifecycle, authority, tasks and communication, which may call Loom as one capability among others. They can share perception, memory, reasoning, solvers, knowledge and verification without becoming the same product.","product-loom-secondary-technical.78b66810bd.second":"A Loom thread has typed input and output inside a weave; it has no organisational identity of its own. A Nexus agent owns a role and can coordinate work across people, tools and other agents. Loom can serve one agent, several agents or Fabric directly. The two products may reuse the same cognitive faculties without blurring that semantic boundary.","product-loom-secondary-technical.3425ccb07f.first":"Loom decides which cognitive graph is required for a request. Core executes the learned and algorithmic operations that graph contains, supplying numerical representations, kernels, dispatch and hardware backends. The boundary is practical: cognitive planning remains in Loom, while operation lowering and machine execution remain in Core.","product-loom-secondary-technical.3425ccb07f.second":"That execution can use native binary and integer CPU paths, larger GPU or distributed backends, and the browser or edge capabilities allowed by the deployment contract. Selected strategic deployments place Core on Kera for a deeper graph-native execution foundation. The choice of backend changes how operations run, not the typed graph contract that selected them.","product-loom-secondary-technical.b6ba549d25.first":"The same cognitive contract can run as managed Fabric on the public Mesh or in licensed customer-controlled, edge, distributed and air-gapped postures. Location, hardware, direct operating rights and available integrations vary with the agreement and policy. The graph, memory, solver, trace, refusal and version contracts remain recognisable across those postures.","product-loom-secondary-technical.b6ba549d25.second":"External sockets are an explicit policy boundary. They may be enabled where a posture permits external dependencies or sealed for local and isolated work, with those conditions visible in the deployment contract. Moving the weave does not change what its typed artefacts mean; it changes the environment and dependencies under which the same cognitive graph executes.","section_st_a2_split-spec-sheet.7c9a7c0610":"Detail","section_st_a6_split-scenario-triad.f75fd3a884":"No runtime deps","section_st_a6_split-scenario-triad.143ac4ef35":"Single binary","section_st_a6_split-scenario-triad.c61f134d87":"Three surfaces","section_st_a6_split-scenario-triad.215475056e":"WHERE IT RUNS","section_st_a6_split-scenario-triad.a794a586f0":"Sealed-room ready","section_st_a6_split-scenario-triad.42d52c32ea":"Local model","section_st_a6_split-scenario-triad.fe6168e59f":"No network","section_st_a6_split-scenario-triad.7d93f83c3e":"Volume-mounted code","section_st_a6_split-scenario-triad.bee0cb05f8":"Kubernetes","section_st_a6_split-scenario-triad.9f1141549f":"Docker","section_st_a6_split-scenario-triad.5315129cfe":"macOS, Linux, Windows","section_st_a6_split-scenario-triad.a89193f674":"No install","section_st_a6_split-scenario-triad.f600b9e709":"One binary","section_st_a7_split-connector-grid.4e6822a6d4":"Hot-reload","section_st_a7_split-connector-grid.19e35f0609":"Set in TOML","section_st_a7_split-connector-grid.d9de50b343":"Caller-driven inventory","section_st_a7_split-connector-grid.8be1bb51b7":"CONNECTORS","section_st_a8_split-scaling-profile.7e0861e04c":"Tunable in TOML","section_st_a8_split-scaling-profile.56d6732417":"Per-project ceiling","section_st_a8_split-scaling-profile.27a454915b":"Active dots reflect concurrent agents per tier.","section_st_a8_split-scaling-profile.0ea215ea0d":"BACKGROUND","section_st_a8_split-scaling-profile.9fc6fb7e03":"TEAMS","section_st_a8_split-scaling-profile.018d2cd1a5":"PARALLEL","section_st_a8_split-scaling-profile.16623c5d4a":"CHAINS","section_st_a8_split-scaling-profile.0fb79c96ed":"Solo → Team → Network","section_st_a8_split-scaling-profile.068f6b3ecc":"TEAM SCALE","section_st_a8_split-scaling-profile.53ebc572b4":"Network","section_st_a8_split-scaling-profile.218887269a":"Team","section_st_a8_split-scaling-profile.9fc93acaa4":"Solo","section_st_a9_split-time-lost-donut.7c9a7c0610":"Detail","section_st_b5_request-lifecycle.47e364232a":"audit #4521","section_st_b5_request-lifecycle.4d3535f846":"patch · session.rs","section_st_b5_request-lifecycle.6c2519cd62":"Tool execution","section_st_b5_request-lifecycle.f6d1c823ab":"session #128 · 14:02","section_st_b5_request-lifecycle.636de0d0c3":"> \"fix this bug\"","section_st_b5_request-lifecycle.0208131a7d":"User request","section_st_b7_audit-chain.9239506454":"0d4e · 8f2c · 51a9 · 73be","section_st_b7_audit-chain.a383df4adb":"a015 · 7c93 · 3e6f · d28b","section_st_b7_audit-chain.efd3faebc5":"9e1b · 22a7 · 8d50 · 6b3c","section_st_b7_audit-chain.4df252b458":"4f2a · b7d1 · 1c83 · e0a4","section_st_b8_defense-stack.eee0f04cba":"YOUR CODEBASE","section_st_b8_defense-stack.89b6e2a784":"DEFENCE IN DEPTH","section_st_c2_dual-loop-flow.9d29573ea6":"tool / step","section_st_c2_dual-loop-flow.08ab1ac55a":"plan / replan","section_st_c5_integration-spoke.55d0677ac3":"All connectors funnel through the same audit + permission layer.","section_st_c5_integration-spoke.c29eb7e425":"Local agent","section_st_c5_integration-spoke.14afd82a7f":"AURA","section_st_c5_integration-spoke.ed0afbab5a":"GitLab CI","section_st_c5_integration-spoke.39540900bb":"GitHub Actions","section_st_c5_integration-spoke.a2feefe9af":"Jenkins","section_st_c5_integration-spoke.ab4115a7b2":"External tools","section_st_c5_integration-spoke.70c123381f":"JSON-RPC 2.0","section_st_c5_integration-spoke.dc99d54d99":"Local","section_st_c5_integration-spoke.6466df1d3d":"Mistral","section_st_c5_integration-spoke.a19ee5a9fd":"OpenAI","section_st_c5_integration-spoke.b780a23b53":"Anthropic","section_st_f1_cta-dark.90e40d5043":"Get started","toc-rail.f5cbdf6bfb":"Contents"}