| name | exeris-benchmarks-comparative-strict-gate-review |
| description | Comparative strict-gate review for exeris-benchmarks. Use on every PR making a cross-target comparative claim — verifies 4 gate artefacts present + `claim-status.json = comparison_eligible` + `track_id` consistent. |
Exeris Benchmarks Comparative Strict-Gate Review
Purpose
Enforce: comparative runtime runs fail closed. Comparative math is valid only when claim-status.json = comparison_eligible AND strict gates pass. track_id is an isolation boundary.
When to Use
- Any PR making a cross-target / cross-protocol / cross-tier comparative claim.
- Any PR aggregating across
track_id.
- Any PR running
scripts/run-comparative.sh, run-entity-read-by-id-campaign.sh, run-e2e-shop-order-saga-campaign.sh, run-tls-matrix.sh, run-iso-budget-sweep-ab-ba.sh, run-full-triad-ab-ba.sh.
Required Inputs
- Comparative output directory.
- Stated claim (what is being compared, on what axis).
track_id assignments per row.
Review Procedure
- Artefact presence — confirm all four artefacts present:
stage7-gate-report.csv
stage7-gate-summary.json
claim-status.json
rejection-codes.json
- Claim status —
claim-status.json final status MUST be comparison_eligible. non_eligible → reject the claim (but non_eligible labelling itself is honest and OK).
- Gate rows — every row in
stage7-gate-report.csv should be pass. fail rows MUST have rejection_code populated; comparative math against fail rows is invalid.
track_id isolation — every aggregated row carries the same track_id. Cross-track aggregation is a hard reject.
- Axis + track labelling — report outputs MUST include explicit axis labels (tier, protocol, family, comparison axis) AND the track label.
- Run scripts — confirm
scripts/validate-comparative-readiness.sh and scripts/aggregate-comparative-results.sh were used in the right order.
- Decision and report —
APPROVE / CONDITIONAL / REJECT.
Decision Logic
- APPROVE: All four artefacts present,
comparison_eligible, gate rows pass, track_id consistent, labels present.
- CONDITIONAL: Sound run but missing one specific label / report stamp — propose the specific addition.
- REJECT: Missing gate artefact,
non_eligible status, fail rows used in math, cross-track aggregation, missing labels.
Completion Criteria
- Artefact presence checked.
- Claim status confirmed.
- Gate rows audited.
track_id isolation confirmed.
- Labelling confirmed.
- Verdict and remediation recorded.
Review Output Template
- Scope analysed (comparative directory / claim)
- Artefact presence (4/4 present?)
- Claim status (
comparison_eligible / non_eligible)
- Gate rows (pass / fail count + rejection codes)
track_id isolation (consistent / cross-track aggregation)
- Labelling (axis + track labels present)
- Verdict (
APPROVE / CONDITIONAL / REJECT)
- Required actions (precise and minimal)
Non-Negotiable Rules
- Never approve a comparative claim without all four gate artefacts.
- Never approve a claim with
non_eligible status as if it were eligible.
- Never approve cross-
track_id aggregation.
- Never approve a comparative claim that bypasses
scripts/validate-comparative-readiness.sh.