Skip to main content

Documentation Index

Fetch the complete documentation index at: https://wuweism.com/llms.txt

Use this file to discover all available pages before exploring further.

Scientific and legal work demands that claims be traceable: where did this conclusion come from, how confident is it, and what would it take to falsify it? The Claim Ledger is Wu-Weism’s answer to that requirement. Every hypothesis, observation, and counterfactual generated in Wu-Weism is automatically written to the Claim Ledger with a full provenance record — the session it came from, the feature that produced it, its confidence score, and whether it passed the epistemic gates designed to filter out unfalsifiable or redundant claims.

​
What a claim is

A claim is a discrete, structured assertion produced during an analysis session. Claims are not informal notes — they are machine-readable records with specific fields that make them queryable, filterable, and auditable.

​
Claim kinds

Wu-Weism produces three kinds of claims:
KindDescriptionProduced by
hypothesisA causal proposition that could in principle be testedCausal Chat, Hybrid Synthesis
observationA factual statement grounded in available evidenceAll features
counterfactualA claim about what would have happened under a different scenarioCausal Chat (intervene mode), Legal Causation

​
What a claim record contains

Each claim in the ledger includes the following fields:
FieldTypeDescription
idstringUnique identifier for the claim
claim_kindstringhypothesis, observation, or counterfactual
source_featurestringWhich feature generated the claim: chat, hybrid, or legal
statusstringCurrent state of the claim in review
confidence_scorenumberNumeric confidence from 0 to 1
uncertainty_labelstringHuman-readable uncertainty: low, medium, or high
session_idstringThe session during which the claim was produced
trace_idstringLinks the claim to a counterfactual trace, if applicable
created_atstringISO 8601 timestamp of claim creation

​
Uncertainty labels

The uncertainty_label translates the numeric confidence score into an actionable interpretation:
  • low — The claim is well-supported. The causal structure is identified, the adjustment set is valid, and the evidence base is strong.
  • medium — The claim rests on assumptions that are plausible but not fully verified. Treat with care before acting on it.
  • high — Significant uncertainty remains. The claim may reflect an unidentifiable causal effect, missing confounders, or sparse evidence.
Uncertainty labels do not mean a claim is wrong. A high uncertainty counterfactual may still be the best available answer — it tells you how much epistemic confidence to place in it.

​
How claims are created

You do not create claims manually. Wu-Weism generates them automatically at the end of each analysis:
1

Complete a session

Finish a Causal Chat session, run a Hybrid Synthesis analysis, or complete a Legal Causation analysis. Claims are not created mid-session — they are finalized when the session produces a governed output.
2

Epistemic gates run

Before the claim is written to the ledger, two gates evaluate it:
  • Falsifiability Gate: Is the claim testable in principle? Claims that cannot be falsified by any possible observation are flagged or rejected.
  • Novelty Gate: Does the claim add something not already in the ledger? Redundant restatements of existing claims are identified and linked rather than duplicated.
3

Claim written to the ledger

The claim is written to the Claim Ledger with its full provenance record, including the SCM that was loaded, the session and trace IDs, and the gate decisions.
4

Evidence links attached

If the claim was generated with reference to source documents — uploaded PDFs, cited literature, or prior claims — those sources are linked to the claim record.

​
Gate decisions

Every claim record shows whether it passed the two epistemic gates: Falsifiability Gate — A claim passes if it can, in principle, be refuted by evidence. Unfalsifiable claims (e.g., “consciousness cannot be explained”) are flagged with a gate decision of failed and a note explaining why the claim cannot be tested. Novelty Gate — A claim passes if it is not a restatement of an existing ledger entry. If a similar claim exists, the new claim is linked to the prior entry and marked as a duplicate or extension. Gate decisions are visible in the ledger and accessible via the API. They are part of the claim’s permanent record.

​
Viewing your claims

Navigate to /claims in the workbench to see the full ledger for your account. The claims view shows:
  • All claims across all sessions, sorted by most recent
  • Filter controls for source feature, session, and claim kind
  • The confidence score and uncertainty label for each claim
  • Gate decisions and any flags applied during evaluation
  • Links to the originating session and, for counterfactuals, the counterfactual trace

​
Evidence links

Each claim references the sources that support it. Evidence links connect a claim to:
  • Source documents: PDFs or papers that informed the response
  • Prior claims: Existing ledger entries that the new claim builds on or contradicts
  • Counterfactual traces: The deterministic computation record for any do(·) intervention
Evidence links make it possible to trace a claim back through the chain of reasoning to the original inputs.

​
Filtering claims via the API

The Claim Ledger is fully accessible via the API for programmatic auditing and integration. 
GET /api/claims
Query parameters:
The number of claims to return. Accepts an integer between 1 and 100. Defaults to 20.
GET /api/claims?limit=50
Filter claims by the feature that produced them. Accepted values: chat, hybrid, legal.
GET /api/claims?sourceFeature=legal
Return only claims from a specific session. Use the session ID from the workbench or the causal chat API response.
GET /api/claims?sessionId=sess_abc123
Return the claim associated with a specific counterfactual trace. Each trace produces at most one counterfactual claim.
GET /api/claims?traceId=trace_xyz789

​
Example response

{
  "claims": [
    {
      "id": "clm_01j8kq3p",
      "claim_kind": "counterfactual",
      "source_feature": "chat",
      "status": "active",
      "confidence_score": 0.74,
      "uncertainty_label": "medium",
      "session_id": "sess_abc123",
      "trace_id": "trace_xyz789",
      "created_at": "2026-03-14T09:22:11Z"
    }
  ],
  "total": 1,
  "limit": 20,
  "offset": 0
}
Use traceId filtering to cross-reference a claim with its counterfactual trace. The trace record contains the intervention details, adjustment set, and identifiability result that produced the claim. See Counterfactual reasoning for the trace schema.

​
Related concepts

Pearl's causal ladder

How claim kinds map to the three rungs of causal reasoning.

Counterfactual reasoning

The traces that back counterfactual claims in the ledger.

Structural Causal Models

The domain models that constrain what claims can be made.