Skip to content

GCL-CEX-01-OBS-001 — Observational Validation and Simplification

Status: PREPARED FOR PROTECTED ADMISSION Governing issue: grandchallenge/MATH-PROGRAMME#1033 Standard under observation: GCL-CEX-01 v0.1.0 at grandchallenge/gcl-standards@efe06a27aa594c63bd0489929ffd9fff1e2daeb9 Programme starting base: 7cfa047c820b08b9a7cd97dae5898bb25fa84b66

Objective

Determine from protected operational evidence whether GCL-CEX-01 is delivering its intended properties in practice and identify the smallest evidence-backed corrections or simplifications needed before any normative revision.

This is an observational tranche. It does not modify the admitted v0.1.0 normative bytes, certify mathematics, alter MATHCERT authority, or create new agent authority.

Phase 0 — non-mutating reconnaissance

Before changing any governed campaign machinery:

  1. Re-fetch live protected heads for gcl-standards, MATH-PROGRAMME, and MATHSOLVE.
  2. Read the admitted standard, Programme adoption, current status record, external-execution profile, and the current protected CEX campaign state in MATHSOLVE.
  3. Enumerate protected CEX operation, freeze, completion-receipt, and preflight artifacts.
  4. Identify status drift, missing observability, and duplicated preflight logic.
  5. Do not alter MATHSOLVE campaign code or the admitted standard during this phase.

Lane A — status coherence

Repair only non-normative status/presentation surfaces where they are stale. Preserve the admitted GCL-CEX-01 v0.1.0 blob unchanged. Record cross-repository adoption state by exact protected identity.

Lane B — telemetry

Produce a machine-readable snapshot derived only from protected evidence. At minimum attempt to measure:

  • operation count;
  • disposition counts;
  • frontier-moving fraction: CLOSED/BLOCKED/FALSIFIED versus all terminal dispositions;
  • MAINTENANCE/RECONNAISSANCE fraction;
  • admission-cycle churn where reconstructible;
  • stale-evidence catches where explicitly recorded;
  • deterministic-preflight catches where explicitly recorded;
  • missing-observability fields.

Unknown values remain unknown. Do not infer absent events from silence.

Lane C — blind cold-start benchmark

Run at least one reproducible zero-chat-history trial. Input is limited to repository identity plus campaign or operation identifier. Score:

  • correct current frontier;
  • correct protected authority/adoption anchor;
  • correct stale-evidence handling;
  • correct allowed next action;
  • claim-boundary preservation;
  • steps and elapsed operator interactions to usable state.

Preserve the exact prompt, starting identities, resulting reconstruction, scoring record, and any failure.

Lane D — preflight consolidation reconnaissance

Inventory campaign-specific preflight scripts. Split each check into:

  1. generic CEX invariant;
  2. campaign/domain invariant;
  3. historical compatibility shim;
  4. obsolete or duplicated logic.

A rewrite is not authorized merely because duplication exists. Produce a bounded consolidation design first.

Lane E — review-independence annotation

Observational reporting must distinguish:

  • role separation;
  • distinct logical passes;
  • distinct durable review records;
  • distinct runtime/session, when present;
  • stronger epistemic independence, only when actually supported.

Historical compliant reviews are not retroactively invalidated.

Terminal disposition

End with exactly one evidence-backed disposition:

  • NO_NORMATIVE_CHANGE
  • NORMATIVE_ERRATUM_CANDIDATE
  • V0_2_CHANGE_CANDIDATE

A v0.2 candidate is not itself authorized by this tranche.

Terminal condition

Protected evidence is measured; at least one blind cold-start trial is reproducibly scored; preflight duplication is classified; status coherence is repaired or precisely blocked; limitations are explicit; and the terminal disposition is recorded without widening claim authority.