# Modular Runtime Phase 3 — Live Competitive Foundation Canary 01

Date: 2026-08-01

Status: accepted competitive UI slice

## Accepted outcome

One genuinely unresolved Helios foundation moved through the bounded competitive path without operator input:

1. One accepted read-only `visual-contract:helios@v1` artifact admitted one competitive `ui_foundation` graph node against the exact project `integrationHead`.
2. The controller admitted exactly two named directions—`airy-editorial` and `structured-utility`—from the same pinned base commit.
3. Two Sol/high planning calls independently produced bounded packets that owned only `public/foundation.css` and preserved the same product, visual, contract, foundation, and base lineage.
4. Two Luna/xhigh workers implemented those packets concurrently in separate worktrees. Neither worker held Chrome, integration, or Git authority.
5. Deterministic host code inspected and sealed each candidate, reran its offline and trusted-host checks, and persisted both immutable candidates before comparison.
6. One central `browser:chromium` resource captured the two running products serially at exactly 402×874. Both captures passed readiness, font, console, overflow, and required-visible-selector checks.
7. A fresh read-only Sol/high context compared the ordered evidence against the same contracts and selected `airy-editorial` autonomously.
8. Only that winner entered the existing `IntegrationQueue`. The authoritative repository advanced by exactly one commit, the loser was not an ancestor of `HEAD`, and the graph's `integrationHead` became the winner commit.

The live test completed in 351.430 seconds (351.70 seconds including harness startup). It used two Sol/high planning calls, two genuinely overlapping Luna/xhigh implementation calls, and one fresh Sol/high selection call. It required no repair attempt or human checkpoint.

## Decision evidence

The autonomous selector chose `airy-editorial` because its type-led hierarchy, true-white space, outlined identity row, substantial focus surface, and singular warm-yellow action were calmer and more product-specific. It found `structured-utility` contract-compliant but denser and more conventional.

Post-run pixel inspection independently agreed with that ranking. This inspection audited the runtime decision; it did not participate in selection or integration.

| Direction | Runtime result | Immutable capture | PNG SHA-256 | Candidate commit |
|---|---|---|---|---|
| `airy-editorial` | selected and integrated | [`airy-editorial-402.png`](./evidence/MODULAR_RUNTIME_PHASE3_COMPETITIVE_CANARY_01/airy-editorial-402.png) | `1201c2d35c4e6a8eac3d55505d234403b3856e2f0be481d5ed59b72e0b4953d0` | `fb47c5fd93e6e2977aeaa1a68a70926250a7169d` |
| `structured-utility` | rejected; never integrated | [`structured-utility-402.png`](./evidence/MODULAR_RUNTIME_PHASE3_COMPETITIVE_CANARY_01/structured-utility-402.png) | `bfad39e25185753fe6f6a37b93520b83fc6361b690925ef45be7a60c9ffddd5e` | `8f50691f24ddfc5c40a1cd6441ec7260ec2462fd` |

The complete immutable decision is [`foundation-selection.json`](./evidence/MODULAR_RUNTIME_PHASE3_COMPETITIVE_CANARY_01/foundation-selection.json), SHA-256 `33b162beee4eb775a45fcbf1ece1a24154d43a90a40bd54d75837fe79d1ce650`.

## Concurrency and authority proof

- Deterministic conformance instrumentation observed `maxImplementing = 2` and `maxCapturing = 1`.
- The live resource ledger contains exactly two Chrome leases; the second grant is no earlier than the first release.
- Both candidates were based on `72a27dccbe16ac0d4ae2a3ff7a872dc85f042fce`.
- The integration ledger contains exactly one job and names only the selected commit.
- `git rev-list --count <base>..HEAD` returned `1`.
- `git merge-base --is-ancestor <loser> HEAD` failed as required.
- The authoritative worktree ended clean at the selected commit.
- Replaying the completed module returns the same selection, integration receipt, and graph state without planning, implementing, capturing, selecting, or integrating again.

## Deliberate mechanism limit

This is not a tournament framework or workflow language. The admitted mechanism is deliberately narrow:

- exactly two directions;
- only for an explicitly unresolved or rejected visual foundation;
- one durable competition record;
- existing attempts, resource broker, candidate sealing, visual capture, selection schema, and integration queue;
- no candidate fan-out after selection;
- no loser merge, hidden synthesis, or operator taste checkpoint.

One direction remains the default for ordinary UI work.

## Verification evidence

Commands:

```text
pnpm --filter @orchestrator/modular-builder-core build
pnpm --filter @orchestrator/modular-builder-core test
pnpm --filter @orchestrator/modular-builder-core check:source
RUN_LIVE_BROWSER=1 pnpm --filter @orchestrator/modular-builder-core exec vitest run test/live-chrome-running-product-capture.test.ts --reporter=basic
NO_COLOR=1 RUN_LIVE_CODEX=1 PHASE3_COMPETITIVE_EVIDENCE_DIR=../../docs/audits/evidence/MODULAR_RUNTIME_PHASE3_COMPETITIVE_CANARY_01 pnpm --filter @orchestrator/modular-builder-core exec vitest run test/live-competitive-foundation-canary.test.ts --reporter=verbose
```

Current-revision results:

- 179 deterministic tests passed across 33 active suites;
- ten explicitly gated real-model or real-browser tests were skipped by the deterministic suite;
- the live Chrome regression passed 1/1 in 1.776 seconds;
- the competitive live canary passed 1/1 in 351.430 seconds;
- TypeScript completed with no diagnostics;
- 50 source files average 170 logical lines, with p50 136, p90 370, and a 492-line maximum;
- no source file crosses the 500-line cohesion-review threshold.

## Gate consequence

This canary closes the remaining competitive requirement of Phase 3. The complete backend, cooperative UI, read-only, and competitive evidence is consolidated in [`MODULAR_RUNTIME_PHASE3_GATE.md`](./MODULAR_RUNTIME_PHASE3_GATE.md).