Module causal_reasoning 

v0.40: Causal reasoning over the schema landed in v0.38.

v0.38.0 made causal_claim and causal_evidence_grade first-class fields on Assertion. v0.38.1 folded a soft compatibility multiplier into the confidence formula. v0.38.2 let aggregate queries filter by claim type. v0.38.3 caught the most common structural error (supports across claim-strength mismatch).

v0.40.0 lands the reasoning move: a hard identifiability verdict. Given a finding’s (claim, grade), can the design — as declared — support the claim being made? This is Pearl’s identifiability question at level 1: does the rung-of-the-ladder match the evidence type?

Doctrine:

• Identifiability is a function of (claim, grade), not of the confidence score, the citation count, or any soft signal. Either the design admits the claim or it doesn’t.
• The kernel records the verdict; the kernel does not auto-correct. v0.40.1+ will surface remediation proposals so a reviewer can downgrade the claim or strengthen the evidence.
• Findings without typed claims (causal_claim = None) are Underdetermined — the kernel knows it doesn’t know.

Structs

AuditEntry

One row of the causal-audit report.

AuditSummary

Summary counters for an audit pass.

Enums

Identifiability

v0.40: hard identifiability verdict for a finding’s causal claim against the declared study-design grade.

Functions

audit_finding

v0.40: audit one finding against the identifiability matrix.

audit_frontier

v0.40: audit every finding in a frontier. Return entries sorted so reviewer-attention items (Underidentified, then Conditional) surface first; identified findings sink to the bottom.

is_identifiable

v0.40: hard identifiability check on (claim, grade). Pure function; the matrix encodes the Pearlian doctrine documented above.

summarize_audit