Skip to main content

Epistemic Contracts

Fornax v0.0.8 adds Epistemic Contracts — a versioned schema that says, for a given class of claim (tests_passed, deployment_healthy, security_defect_fixed, and others), exactly what evidence would need to exist for that claim to be justified.

The core idea​

Before v0.0.8, Fornax's five-state verdict (VERIFIED / UNVERIFIED / CONTRADICTED / REVIEW / UNAVAILABLE) told you whether a claim matched the evidence. It didn't tell you what evidence a claim of that type should have had in the first place.

An Epistemic Contract makes that explicit:

  • A claim class (e.g. tests_passed, versioned — v1, v2, ...).
  • A list of proof obligations — each one names a required evidence kind, a minimum trust class, a freshness window, and (new in v0.0.8) an independence rule so that two pieces of evidence that ultimately trace back to the same underlying observation don't get counted as two separate confirmations.
  • A satisfaction state for each obligation — one of seven, never collapsed into fewer: Satisfied, Unsatisfied, Unavailable, Stale, Contradicted, NotApplicable, Unknown.

Contract satisfaction​

Given a claim, its contract, and the evidence actually captured, Fornax's contract-satisfaction engine produces a SatisfactionReport: which obligations are met, which aren't, and why. This report feeds a critical-obligation safety floor — a recommendation can never reach PROCEED merely because aggregate evidence looks strong while a single required obligation is unsatisfied.

Six representative claim classes ship in v0.0.8: tests_passed, build_succeeded, file_changed, commit_push_completed, deployment_healthy, and security_defect_fixed.

What this does not do​

  • It does not invent evidence. An obligation with no matching evidence reads as Unavailable, never as an implicit pass.
  • It does not imply calibrated real-world accuracy. Whether a satisfied contract correlates with real-world correctness at some measured rate is a separate, harder question this release does not claim to answer (see Reliability).

See also​

  • Adaptive Verification Budgets — how much verification effort a claim gets is informed by its contract's obligation-satisfaction state.
  • Assurance Cases — a human-readable explanation built directly from a contract's satisfaction report.