Conformance Framework

TSS-CF Conformance Framework

A differential conformance framework for threshold-signature and MPC systems: every release executes against the machine-checked formal model, and the two behaviours are compared axis by axis, producing a three-valued verdict with replayable evidence.

A threshold-signature deployment rests on two artifacts: a formal model with proven security invariants, and an implementation that evolves with every release. TSS-CF binds them. Each scenario in the conformance corpus executes twice — once on the production runtime and once on the executable formal model — and the observable behaviour of both runs is projected onto named comparison axes.

Every comparison resolves to exactly one verdict. Conformant: every required axis matched at its stated maturity. Indeterminate: the comparison signal is incomplete — an axis was unsupported, immature, or a projection failed. Violated: implementation and model diverged on a checked axis, and a replayable counterexample pack accompanies the verdict.

The framework runs in continuous integration. Model drift surfaces at merge time, on the release that introduced it, with content-addressed evidence attached.

Design principles

Four rules hold across every runner, family profile, and formal backend.

Observable behaviour. Conformance is judged on externally observable projections — publications, accepted ingress, evidence classes — independent of internal state.
Typed absence. A missing observation is a typed state (Unsupported, NotRun, ObservedEmpty, ProjectionError) and enters the verdict as such; agreement requires positive evidence on both sides.
Scoped claims. Every claim names its scenario corpus, family profile, compared axes, runner and adapter versions, and per-axis maturity.
Replayable evidence. Reports and counterexample packs are content-addressed and reproduce their verdict offline, for any party.

Position in the assurance stack

Conformance strength is a product of independently evidenced factors. TSS-CF makes each factor explicit and separately tracked.

Cryptographic hardness, side-channel behaviour, RNG quality, secure erasure, and formal-model fidelity are named obligations with their own evidence tracks; the conformance verdict covers the model-to-implementation binding. The secret boundary itself is verified twice — statically at compile time by the taint-analysis lint suite, and behaviorally by the leakage axis.

Assurance composition

text

assurance = model fidelity
          x scenario coverage
          x axis coverage
          x projection faithfulness
          x crypto binding
          x production hardening

Document map

The specification pages define the scenario language, observation model, family profiles, and verdict engine. The evidence pages define the report, counterexample, and attestation formats. The assessment pages define assurance levels, the assessor procedure, and regulatory alignment.

Assurance levels Verdict engine Claims & attestation