Eight-Cell Product Cap Promotion

A — PROMOTED — READY FOR 0.12.0 RELEASE. Normal product cap: 8.

The selected-network product maximum is promoted from6 to8 in the canonical policy and normal Go/React/OpenAPI builds. Recommendation remains5, Measurement Validation remains6. The frontend Recommendation guard uses its independent constant; request builders/persistence guards and bounded-follow-up accounting are preserved.

Qualification remains the exact eight frozen paired Ankara2026.07 W1 layouts at2.6/28GHz,120rays/400m/30dBm/120deg/legacy two-pass, shipping Go1.26.6 Linux/arm64 CGO0/cgroupv2 Profile A2CPU/4GiB. No W2/W3, arbitrary hardware/datasets or multi-instance extrapolation.

Promotion and accounting

Normal Evaluate/Optimize/Interference/Building Entry/Explanation accept8/reject9. Normal Auto reports8 without adaptive behavior. No source-copy cap patch or environment switch is used. Historical audit binaries/reports retain their original six-cap context.

All16 frozen F28 cases succeed;448 required RF calls. Worst 11.788444160s, minimum headroom 48.211555840s, process kernel peak 1217204224 bytes, OOM/kill/max0. The matrix completed before an additional scientific probe failed because it omitted a historically included optimization field; retained completed HTTP rows were independently verified, and corrected supplemental comparisons passed. No failed probe substitutes for evidence.

Exact eight-Cell JSON outputs match800 frozen audit-eight baseline responses;32 Explanation/Building Entry comparisons match historical evidence, excluding only the pre-existing volatile Building Entry diagnostics.elapsed_ms. Search still proposes578 legacy states. Distinct-IP Optimize+Optimize, Optimize+Evaluate and Evaluate+Evaluate overlap; qualified async/sustained interval, producer/drain and final-tail guards pass. Real cancellation and budget/replay/forgery checks retain the validated policy behavior.

Product presentation and persistence

Actual normal eight-Cell F28 browser test passes; numbered badges1..8 render. App interaction tests refuse a ninth without RF and retain selection. Eight responsive light/dark presentation tests cover normal Setup/list/scrolling, with RF mocked only for presentation. The actual F28 test opens Review Results and Solutions, verifies eight per-Cell rows and scrolls to the last row. Reports retain all eight IDs. Eight network-result-only projects round-trip; larger ray-bearing projects can still reach existing guards at either cardinality. These guards and schemas remain unchanged.

Release and historical evidence

All promotion gates pass. Canonical VERSION and package-visible references are0.12.0. The final normal shipping image completes F28, matches28 frozen outputs and rejects the next unrelated request. No tag/release/push is performed here. W1 qualification, invalid/valid audits, design and policy implementation studies are hash-preserved. New tooling builds normal cap8; historical cap-only audit mechanisms remain immutable history.

Lock SHA256 2b493404f0ccfc5396416b9c41a4c8c66015a84b77937ee1d84cc524c24f2346. Raw evidence lives outside Git at /tmp/atom-eight-cell-product-cap-promotion; source/tests/locks/manifests/report summaries are tracked.

Required final report

1. Starting baseline: main @ a10284a708661795b1e31bd1f40c614acf55e8f4; VERSION0.11.0; validated policy changes already uncommitted and preserved.
2. Promotion-lock SHA256: 2b493404f0ccfc5396416b9c41a4c8c66015a84b77937ee1d84cc524c24f2346
3. Product-cap inventory: Canonical policy, generated Go/JS, frontend selection/copy, validators/errors, OpenAPI/product docs, Auto and cap-specific tests inventoried; historical/scientific/fixture six values classified separately.
4. Every cap location changed: policy network_towers_max, generated Go/JS, networkSelection binding, main/interference errors, ControlPanel copy, Network/BuildingEntry OpenAPI maxItems, product docs and cap tests.
5. Every independent limit intentionally retained: Recommendation5, Measurement6, experiment64/worker1/queue16, ray/radius/sample/profile limits and all persistence guards retained.
6. Backend max before/after: MaxNetworkTowers6→8; unchanged minima.
7. OpenAPI max before/after: Product network/BuildingEntry6→8; Recommendation5/Measurement6 unchanged.
8. Frontend max before/after: MAX_NETWORK_CELLS now generated policy8; ninth refused.
9. Validation messages: Exact main/interference messages now2..8; independent1..6 message retained.
10. Auto metadata result: Normal shipping Auto max_cells8, observation-only; other admission/resource metadata unchanged.
11. Standard 8 acceptance: Normal8 accepted across applicable APIs.
12. Standard 9 rejection: Normal9 rejected before RF.
13. Per-endpoint 8/9 results: Evaluate/Optimize/Interference/BuildingEntry/Explanation8 accept9 reject; frozen full workflows and additional contracts prove both.
14. Recommendation cap result: Recommendation5 accepts;6/8 rejects. Frontend independent guard corrected to5 instead of product-max minus1.
15. Measurement Validation cap result: Measurement6 accepts;7/8 rejects.
16. Frontend select-8 result: Actual normal browser selects8 and completes F28.
17. Frontend ninth-selection result: App interaction and selection helpers refuse ninth, keep8 and dispatch no RF.
18. Eight marker/list presentation: Eight badges/list and responsive light/dark Setup/Inventory pass; actual Review Results/Solutions render8 rows and scroll to the last; established layout unchanged.
19. Request builder order/count: Eight IDs/coordinates/profiles/azimuths preserved in order; no six truncation; unit regression passes.
20. Eight Evaluate/maps accounting: Fresh Evaluate/maps ordinary1/extra8, holds0.
21. Eight Optimize/maps accounting: Fresh Optimize/maps ordinary1/extra8, holds0.
22. Eight full-cycle accounting: Full cycle ordinary11/extra8.
23. Eight F28 accounting: F28 ordinary20/extra8, held0;16 frozen layout/band cases complete28 calls.
24. Post-F28 denial: Unrelated29th protected attempt429/no RF.
25. Worst request latency: 11.788444160s
26. Minimum deadline headroom: 48.211555840s
27. Peak cgroup memory: 1217204224bytes process-lifetime kernel peak covering matrix; below3GiB.
28. OOM events: oom0/oom_kill0/max0.
29. Optimize+Optimize: Two Optimizers both succeed and overlap with global2.
30. Optimize+Evaluate: Optimize+Evaluate both succeed and overlap.
31. Evaluate+Evaluate: Evaluate+Evaluate both succeed and overlap.
32. Async preservation: Qualified background interval guard passes normal async Optimize; computing worker overlaps request.
33. Sustained async preservation: 30-second producer/at-least95% active/final-tail/drain guards pass sustained Evaluate.
34. Cancellation/slot cleanup: Real normal eight Optimize cancellation releases slots; same-client and other-client requests remain usable.
35. Workflow cleanup: Claimed child cancellation retains charge and retires siblings; pending-root callback/owner cleanup tested; no leaked holds.
36. Scientific invariance: 800 matrix JSON matches frozen audit-eight baselines;32 Explanation/BuildingEntry historical canonical matches; science sources unchanged apart from cap/error text.
37. Search proposal count: Legacy proposal count578 confirmed by normal engine timing regression; legacy algorithms unchanged.
38. Budget invariance: Ordinary20+extra8/anchored60 unchanged; independent21 and postF29 deny; same policy source hashes.
39. Replay/forgery preservation: Real replay/foreign/mutation and full race suite preserve fail-closed behavior.
40. Persistence observation: Eight network-result-only project round-trip passes. Existing large ray-bearing guards still apply at six/eight; not raised.
41. Report/export result: All eight IDs render in HTML/Markdown reports; no export-summary truncation.
42. Documentation changes: Current README/getting-started/architecture/API/OpenAPI updated; scope and persistence guard caveats added; historical reports immutable.
43. Changelog: Promotion and qualified scope recorded under Unreleased.
44. VERSION: 0.12.0
45. Backend tests: Full backend suite plus normal8/reject9/independent/578/F28 tests pass.
46. Race detector: Full go test -race ./... passes.
47. Frontend tests: Full frontend suite plus selection/builders/report/persistence/ninth-interaction regressions pass.
48. Real E2E execution counts: Four named real E2Es each discovered1/executed1/pass1/skipped0;8 presentation tests pass with0 skipped.
49. Build/lint/vet: Production build,lint,go vet and generated-policy check pass.
50. Docs/OpenAPI validation: Docs,OpenAPI semantic caps/schema validation and version consistency pass.
51. Methodology violations: Failed development probes/selector assumptions preserved and corrected; no skipped/failed probe counted as final evidence. Completed matrix retained and reverified despite later supplemental fixture omission.
52. Missing evidence: No required evidence missing after final audit; bounded promotion confirmation reuses qualified evidence.
53. Final A/B/C/D/E promotion verdict: A — PROMOTED — READY FOR0.12.0 RELEASE
54. Whether product cap is now 8: Yes: normal backend/frontend/OpenAPI cap8; no audit override.
55. Exact qualification scope: exact frozen eight paired W1 layouts, 120 rays/400m/30dBm/120deg/2.6 and28GHz/legacy two-pass; no W2/W3 extrapolation
56. Known limitations: No W2/W3/arbitrary datasets/hardware/replicas/general cardinality claim. NAT shares budgets; large ray exports retain guards.
57. Release recommendation: 0.12.0 ready; no tag/publish/push.
58. Files changed: Focused cap/error/selection/OpenAPI/docs/test changes plus promotion lock/report/validation tooling; inherited policy patch preserved.
59. Raw evidence policy: Raw traces/screenshots/binaries/resources outside Git, SHA-indexed/local archive; manifests and source/tests/reports tracked.
60. Freeze decision: Frozen cap8/version0.12/20+8/ProfileA-W1/concurrency2/1/deadline60/worker1/queue16; all gates pass.
61. Single recommended next action: Review and release0.12.0 through normal release workflow; no additional capacity expansion.