AI Knowledge Governance and Provenance | Dweve Spindle

Choose the audience that matches your question

The page contains three selectable readings of the same subject.

Coding agent and operator assistance. Demo metrics are illustrative.

Terminal, search, lint, test, git, and more.

Remembers your codebase and team context.

Specialised agents collaborate across concerns.

Review diffs, request changes, final sign-off.

Replay any session bit-for-bit when something needs review.

Autonomous agents that write code and keep receipts

to handle network timeouts, 5xx responses, and idempotent-safe conditions. Pure helper, fully unit-tested.

Spindle turns your sources into facts you can defend. An audit trail a regulator will read. Scope a pilot, map the integration into your existing knowledge systems, and see the lineage output before you commit.

Point Spindle at your sources. Get a graph of Canonical atoms with six-dimensional quality, 47-type bias reports, and forward and backward lineage on every fact.

Every fact cited. Every confidence honest. Every disagreement visible. Meet Spindle through the Fabric workspace.

Move the boundary without changing what knowledge means.

Trigger downstream work via webhooks, or query the state machine through the API.

Subscribe to atom promotions, threshold breaches, bias flags, or conflict detections.

SharePoint, Confluence, data warehouses, ticketing, and collaboration platforms, each a scoped source class.

High-dimensional embeddings for similarity search.

Partitioned metrics with continuous aggregates.

ACID storage with batch operations and pooling.

Sharded semantic lineage with traversal queries.

A reviewer can rebut a tag. The rebuttal lands in the chain. No silent override.

Adding a category is a versioned change with an effective date. Old atoms keep their tags.

The bias engine runs as part of certification, not as an after the fact audit.

Each agent mimics a focused human expert.

Missing context blocks progression below threshold.

Held to the configured policy gate before certification.

Target population coverage validated against the declared policy.

Supports, contradicts, extends, depends.

Source chain, authors, dates, credibility.

A fact, claim, method, question, or result.

Export Compiler emits training-ready formats from the verified graph.

Governance Council applies DMBOK and the configured AI governance policy.

Quality Guard runs Byzantine consensus and 47-type bias detection.

Connection Network resolves entities and maps typed relationships into the graph.

Analysis Division decomposes into atomic facts, scores quality, scans for bias.

Extraction Corps turns documents into structured information, preserving headings, tables, and figures.

Discovery Brigade casts the net across academic, industry, code, regulatory, and community sources.

Spindle runs managed through Fabric on the public Dweve Mesh, on your own hardware, or air-gapped. There is no re-platforming between tiers. The same pipeline, the same gates, and the same lineage guarantee hold in every posture, so you choose your posture for control, not for capability.

Spindle parallelises at three levels. Different stages process different items at the same time. Multiple instances of the same agent process different items in parallel. Inside a single agent, independent tasks such as multi-source checking fan out. Each stage scales horizontally on its own worker pool, and a multi-store backend keeps the graph, the atoms, the metrics, and the embeddings each in the right place.

Spindle is not a monolith. It is a military-inspired hierarchy of 32 agents under a Master Orchestrator, organised into a Discovery Brigade, an Extraction Corps, an Analysis Division, a Connection Network, a Quality Guard, a Governance Council, and a Export Compiler. Specialisation mimics human expert evaluation, lets work run in parallel, keeps maintenance focused per agent, and degrades gracefully when one agent fails.

Bias is not a vibe. It is a typed property of the atom. The bias engine inspects every atom at certification time against 47 documented bias categories spread across statistical, cognitive, linguistic, and systemic dimensions, then tags the atom accordingly. A bias report is then a query against the graph, "show me selection-biased atoms in this corpus", not a separate audit run after the fact.

Quality is not a single number. Every atom carries six independent dimensions: verifiability, objectivity, completeness, consistency, temporal stability, and source diversity. Each dimension is queryable on its own, so an auditor can inspect representativeness, accuracy, or completeness separately. Certification applies the configured evidence and governance policy for the stated use; no single score stands in for that decision.

The Knowledge Atom is the indivisible unit Spindle works with: the content itself, its epistemological state, a six-dimensional quality score, complete provenance, and typed graph relationships. Provenance carries the source chain, authors, dates, credibility scores, and transformation history. The same fact stored from the same source is deduplicated in the graph, so a Canonical atom appears once.

Seven stages and hard gates exist for one reason: a fact that cannot be traced back to its source is not a fact. Spindle is a pipeline, not a pile. Each fact enters as a candidate and advances only when it meets the quality, consensus, and provenance requirements of the next stage. The Certified gate applies the configured evidence, rights and governance policy for the declared use. Consensus is a claim-specific evidence requirement, not a universal source count or diversity score. If a fact fails a gate it stays where it is, flagged with a reason. It does not leak into training or answers.

Spindle is an epistemological state machine. Each fact advances through Candidate, Extracted, Analysed, Connected, Verified, Certified, Canonical, with explicit quality thresholds at every transition. Nothing reaches Canonical without clearing them. Entity resolution, multi-source consensus, 47-type bias detection, and full provenance on every atom.

Spindle assembles this chain as the work happens

The audit package is created as facts are certified, not assembled when the question arrives. Every item carries the source it came from, the rule applied at the gate, the certifier who signed off, and a replay path for the inspector.

Trace, provenance, and replay are product properties, not compliance decorations added after the regulator starts asking harder questions. That is what makes the knowledge base hold under scrutiny.

Think about how an auditor accepts a fact.

Managed through Fabric on the public Dweve Mesh, on-prem, or air-gapped.

Six dimensions and 47 bias types, per fact.

Source to Canonical, every step logged and signed.

Counter-evidence flags affected facts and notifies downstream. Both the assertion and its withdrawal are preserved.

Every applicable regulatory framework is validated before the fact is certified.

Forty-seven bias types inspected. A flagged fact lands tagged, never silently.

Six dimensions scored. A fact below the threshold does not advance, and the reason is recorded.

Any answer walks back to its source, end to end.

The lineage moves with the fact wherever it is used.

Provenance is built when the fact is created, not reconstructed when the question is asked.

The audit package is not a screenshot of a dashboard. It is the lineage, the quality scores, the bias reports, the consensus inputs, and the signatures that together let a DPO, an AI officer, or an external auditor reconstruct any decision the system made.

Governance is not a dashboard bolted onto a finished corpus. It is how facts get from Candidate to Canonical. Quality thresholds, bias detection, and compliance rules are enforced at the gate, not audited after the fact. If a fact fails a gate it stays where it is, flagged with a reason.

A fact challenged in a board room or a regulatory review is only answerable if the lineage was attached when the fact was created, not reconstructed when the question was asked. Spindle attaches provenance to every fact at certification time. The lineage travels with the answer.

Spindle turns your fragmented internal sources into a queryable knowledge base where provenance is attached to every fact. Seven stages, thirty-two specialised agents, one documented lineage from the original document to the answer. When the auditor asks where a number came from, the answer is already attached.

The team relitigates the same question every quarter.

Old versions linger uncorrected until someone notices.

Nobody can point to the source of a number.

Each tool pulls from its own scratch space. Conflicts surface in meetings.

Save a workflow. Send it to a colleague. They get the same steps and the same cited sources.

Drop a file. Spindle reads it, links it to what it already knows, and surfaces what you need.

Type a question. Spindle finds the source, checks it, and cites it.

When your team uses Spindle together, everyone draws from one verified pool. No more conflicting answers from different tools. When a document changes, the team is told. The conversation moves from who is right to what should we do.