Mind · Inference
TruthseekingProtocol
Layered epistemic verification (cache → coherence → decomposition → specialists → empirical), with verified claims persisting as content-addressed records — the accretion loop
Definition
Mechanism
A protocol for evaluating factual claims through layered epistemic decomposition, where each verified claim persists as a content-addressed record that later checks can reuse, so the store of verified claims expands through use (the accretion loop). Unlike monolithic claim evaluation (which entangles source credibility, logical consistency, and cross-referencing in one pass, letting strong evidence on any one axis suppress scrutiny on the others), the protocol applies ConceptualDecomposition to verification-depth, producing five layers governed by MarginalValueRule: Layer 0 cache (has this exact claim been verified? a cached verdict is reused only when its recorded scope, time and provenance fit and no superseding evidence is known), Layer 1 structural coherence (does this contradict the verified cache?), Layer 2 epistemic decomposition (ClaimExtractorSolver isolates falsifiable statements, ProvenanceSolver evaluates source independent of content, CoherenceSolver checks internal consistency independent of source, CorrespondenceSolver queries external data, ContestabilitySolver constructs the strongest counter-case), Layer 3 cross-domain specialists in the Cognitive Commons, Layer 4 empirical verification via primary data gathering. Each layer's Solvers operate behind the contract, blind to the others' outputs until they compose at a typed boundary. A claim that survives deep verification is recorded as a content-addressed verified-claim entry, with its verification provenance and boundary conditions, and enters the Layer 0 cache. Only recurring verification strategies are promoted to Sema patterns; individual claims accrete as records. Composes with Validate at each layer boundary.
Invariants
- Layer-blind composition: Solvers within a layer do not see each other's outputs until they compose at a typed boundary — no suppression of scrutiny by a strong signal on a parallel axis.
- Accretion: a claim that survives deep verification enters the Layer-0 cache as a content-addressed record of the claim, its verification provenance and its boundary conditions. A record whose verdict no longer qualifies stays as history and is not reused.
- Conditional reuse: a cached verdict is reused only while its recorded scope, time and provenance fit and no superseding evidence is known; a claim whose cached verdict does not qualify is checked again through the layers.
- Marginal-value depth: an eligible cached verdict resolves a claim at Layer 0; any other claim continues past it, and MarginalValueRule sets how deep it goes.
- Typed outcome: a check may stop without a verdict, and its typed output then reports the warrant deficit, the support still missing, not proof that the claim is false.
Risks
Failure modes
None stated.
Caution
Contamination vector — one mis-verified claim recorded in the shared Layer-0 cache persists and later checks defer to it; gate recording on independent re-verification.
Relationships
Builds on
Composes with
Finding what a concept is made of: locate it, carve it into the dimensions every case it claims needs, test each, search for what is missing, route real cases through it, say how the parts make the whole, and carve again below
Deepen only while expected improvement plus an exploration credit, divided by incremental cost, exceeds a domain threshold
References
Infrastructure
Verifying inputs conform to expected schema
Nothing in this library builds on this pattern yet.