Description
Current hardened per-tab checks prove placed query-bearing elements and field selectors against genuine envelopes, not an independently declared tab-specific product coverage contract. Enum selectors may include values while aggregating away their identities. This is disclosed coverage debt, not separately visible enum proof.
Acceptance Criteria
- #1 Owner grades tab-specific field/enum ownership and whether distinct enum granularity must survive aggregation, with explicit dispositions for explanatory-only tabs and intentional aggregate headlines.
- #2 An independent contract and meaningful seeded counterexamples prove each required tab/field/enum display; structural placement or query-name presence alone cannot claim that stronger coverage.
Definition of Done
- #1 just test
- #2 just tf-validate
- #3 just check-identifiers and just no-em-dashes both return clean
Implementation Plan
Build frozen D-LAY23 section2 independent field/enum ownership contract on accepted layout source. Owner tabs display each required field and enum value distinctly; seed meaningful per-value counterexamples across assembled JSONata/Prometheus/Grafana transformations rather than structural query-name or selector existence alone. Explicit explanatory-only and titled-aggregate exemptions; existing source/lineage/optional-state/privacy/OnCall/raw-unit guards unchanged. Dashboard files only, no source/schema/new-series/identity/reader expansion; root landing after full gate/review then exact composed gate and CI.
Final bounded packet: repair the older dashboard_coverage_inventory only to recognize the exact type-preserving two-field JSONata projection with negative controls against arbitrary replacement/filter; establish actual datasource range-table identity/cardinality before claiming retention behavior (do not infer deployed bug from adapter); add explicit route_parked/not_evaluated and Excluded-rules relationship seed; verify full47row contract, review delta, complete gate and exact CI only after an owner-raised ceiling. Existing source/schema/reader/series/identity fences remain.
Loop24 D-CAP24 resume: total ceiling8=4 consumed+4 conditional future ladder. No new attempt until boundary readiness is proved. (1) supply versioned genuine Prometheus range-table DataFrame fixture or exact local converter witness for A/B, >=2 identities and >=6 repeated samples per side; assert one intended row per stack, paired values, zero, unequal values, units; negative collapse/duplication/swapped/lost-values/scalar controls. Adapter72rows is NOT deployed bug proof; no instant-query substitution. (2) source/fixtures prove route_parked is not_evaluated and overlaps Not evaluated; Excluded rules counts only parked same-stack/signal rows. Future paired owner consumer-path seed with parked+missing_input plus not_applicable/deadline/truncated controls must display Reason and coverage count, reject remap-to-excluded and wrong-signal counts. (3) repair older dashboard_coverage_inventory only for exact two-field type-preserving tenant selector; actual assembled ten-field lineage must pass, fabricated key, literal true:false replacement, appended enabled filter, destructive object/scalar/filter and unknown-column controls must fail. Preserve older optional-root positive. Reconcile retained candidate old base to current main before any new ladder.
2026-10-13 direct-main follow-through authorized by Rob: genuine pinned Grafana v12.2.0 converter plus actual transforms reproduces 72 rows for two stacks with six A/B samples each. Fix locally by retaining range queries and reducing each identity to latest values before joining; prove pairing, zero, unequal values, units and negative controls. Reconcile retained ownership candidate with current main, repair exact legacy lineage recognition and paired exclusion display proof, then independent review, CodeRabbit and full just check. Existing four consumed attempts remain recorded; this is the first conditional continuation attempt, not a reset.
Implementation Notes
Loop23 four implementation attempts, one review-repair round, ceiling reached. Final uncommitted candidate at caf04314cf1da919f520ef9941f615daf9e39c4c has patch60df2aee20672219f647dedc4253544000b5825babab77159bb9f0cf31b900c7. Candidate gate red: older exact-lineage recognizer rejects two-field tenant projection (10 ai_tenant_config gaps); 2713 passed,2skips but Tofu not reached. Actual owner runtime65/66: retention range-table identities/cardinality unresolved at disclosed datasource-frame adapter; explicit route_parked/Excluded-rules pair proof missing. Boolean projection28/28 and tables181/181 passed; CRcomplete0/all5 does not clear red/fullAC. NO LAND. Resume requires owner ceiling increase and exact guarded lineage negative controls, proper retention datasource boundary and exclusion proof. Keep candidate/worktree/evidence; no weaker tests or new producer/source authority.
Loop24 read-only boundary research accepted, zero implementation attempts. Root inspected retention builder and evaluator/summary semantics; retained five hashes match historic candidate. Source/fixture boundaries2-3 established, but genuine retention datasource/converter frame is missing, so D-CAP24 readiness gate remains closed. Paired consumer display proof remains future acceptance, not missing producer semantics. Do not blindly replay historic red candidate. Remaining evidence blocker and exact controls in /Users/rob/repos/grafana-cloud-org-insights/codex/loop24-evidence/field-boundaries/lane.md. Prior4 implementation attempts and1 review-repair remain; total8 authorized conditionally, not reset.
2026-10-13 direct-main implementation accepted. Pinned Grafana 12.2.0 converter reproduced retention 72-row duplication, corrected to per-identity latest range values then join: two paired rows including zero. Reconciled prior candidate, exact legacy tenant projection lineage with destructive negatives, independent 47-row ownership, reader-state display, boolean unknown handling, fixed severity and separate resolution bands. Runtime proof: 71 metric owners, 185 table cases, 28 boolean cases, four producer-to-display exclusion scenarios plus six rejection controls. Independent reviewer PASS; CodeRabbit complete0, all changed files. Configured just check exit0:2739 Python passes,27906 subtests,2 skips;37 OpenTofu passes; current/history identifier and text/format checks clean. Initial missing-pattern gate and separate short identifier timeout retained as failures, not passes. Evidence /Users/rob/repos/grafana-cloud-org-insights/codex/direct-20261013/dashboard-implementation/report.md and just-check-configured-result.json. Local runtime proof, no deployed/browser claim. One conditional implementation continuation used; prior four attempts retained.
Final Summary
Completed all 47 field/enum ownership rows and repaired retention sample multiplication while preserving range queries. Exact reviewed candidate passed full local gate and real local Grafana processing; no deployed-browser claim.