VRS — Vigency Resolution Score (benchmark spec, stub)

Canonics reproducibility surface. This stub replaces the earlier CAVA outline (cef-bench-outline.md); CAVA is retired — see the paper's canon log. This document defines the metric, the corpus structure, and the mandatory adversarial modules so the claim is falsifiable before any result exists.

What VRS measures

Vigency Resolution Score (VRS) is the rate at which a system correctly resolves the vigency state of stored knowledge across the four question types below — i.e. whether it treats each retrieved interpretation as vigent, historical, non-vigent (pending re-derivation), or revoked-as-guidance, relative to the currently active canon.

VRS is an application metric, not a retrieval metric: a system that retrieves the right document but applies a dead-regime interpretation scores zero on that item. This is deliberate — the STALE line of evidence shows models routinely fail to act on updates that are plainly in context.

Corpus structure

Longitudinal scenarios (synthetic operator scenarios plus anonymized real logs). Per scenario:

• multiple domains; at least 3 epochs per domain;
• canon events with scoped regime diffs (retired/introduced assumptions, blast radius);
• question sets whose correct answers change across epochs while remaining semantically similar.

The four question types

1. State Resolution — which interpretation is currently operative?
2. Premise Resistance — reject a question that presupposes a dead epoch.
3. Reinterpretation — reuse an old fact correctly under the current lens.
4. Audit Mode — report what was rational to believe under a past epoch, without hindsight contamination.

Types 1–3 lift STALE's dimensions from fact level to regime level; type 4 is Canonics-specific.

Mandatory adversarial modules

A VRS run is not valid without both:

• Undeclared-shift module. Scenarios containing regime shifts that are not declared in the corpus. Measures (a) how badly κ-based freshness misleads absent declaration — quantifying the κ caveat — and (b) scan precision/recall for the detection hypothesis.
• Age-superiority condition. Scenarios where no canon is declared, or declaration lags materially behind the real shift, so chronological age outperforms κ on invalidity prediction. Such a result does not falsify Canonics globally, but falsifies any unconditional reading of the κ hypothesis, and requires reporting κ alongside declaration-process health and volatility/age baselines.

Submetrics (reported with VRS, never in place of it)

Stale-interpretation avoidance; epoch-attribution accuracy; blast-radius precision/recall (over- and under-invalidation); retroactive-canon handling; audit fidelity. From longitudinal deployment: declaration lag (t_declared − t_effective), discrimination quality, proposal rejection rate (read as a symptom, never a target), and ritual survival.

Baselines the framework must beat

• No-epoch baseline.
• Strategy-document baseline — the current strategy summary always in the system prompt, plus "distrust notes older than it." The cheapest plausible alternative and the null hypothesis VRS must beat to justify the machinery.
• Human-factors-hardened baseline — the strategy document plus cognitive forcing, independent pre-judgment, and a provenance checklist, but no canon events, epochs, or κ. Isolates the ontology's contribution from the ritual's.
• Stamp-only / stamp + canon diff in system prompt / materialized reprocessing — per-mechanism ablations.

Status

Spec only. No results are claimed. Naming was gated on a targeted prior-art search for "Vigency Resolution Score" / "VRS benchmark" before adoption. A reference harness and scenario seeds are future work (paper §10, Track A).