release: split deterministic smoke, provider playtest, and full RRI

[100yenadmin]

Summary

Make release documentation and RRI output distinguish fast wiring proof from real-provider play proof and full release readiness.

Gate split

• Deterministic built-app smoke: fast, every provider/app wiring PR.
• Short real-provider playtest: provider/app integration PRs and latency checks.
• Full RRI: non-partial five-persona release verdict, with Mac built-app proof plus remote/support-VM heavy persona evidence as appropriate.

Acceptance criteria

• WorldOS-OPERATING-GOAL.md, WorldOS-GUI-RUNBOOK.md, and RRI help text document the split.
• qa/release_readiness.py output clearly marks deterministic smoke as non-release evidence.
• qa/SCORECARD.md rows can record smoke/playtest/RRI without implying release readiness.
• release: rerun clean non-partial 5-persona RRI after gate hardening #466 remains blocked until Codex DM should avoid safety-cancelled setup/social tool calls #479 plus app-status/scripted smoke make its evidence trustworthy.

[100yenadmin]

Architecture split-gate draft is in #489. It keeps deterministic smoke, short real-provider playtest, and full non-partial five-persona RRI as separate gates, with full RRI remaining the release verdict.

[100yenadmin]

Gate split implementation draft is up in #491, stacked on #490. now emits explicit gate-split metadata and names as the only release verdict gate.

[100yenadmin]

Gate split implementation draft is up in #491, stacked on #490. qa/release_readiness.py now emits explicit gate-split metadata and names full_five_persona_rri as the only release verdict gate.

[100yenadmin]

PR #475 now documents the gate split in docs/AGENT_GRADE_APP_TESTABILITY.md and the runbooks: deterministic built-app smoke, short real-provider playtest, and full non-partial five-persona RRI are separate gates. This PR proves the short Codex provider playtest only; #466 remains the release-grade RRI step on the support VM plus Mac built-app proof.

[100yenadmin]

Gate-split update from the takeover lane:

• main@32ca561 now has post-[codex] Add Codex DM built-app provider path #475 built-app Codex proof, but that proof is trace-noisy and should remain diagnostic, not release evidence.
• PR [codex] Add dev-gated scripted smoke provider #494 adds the deterministic built-app smoke provider from the old [codex] Add dev-gated scripted smoke provider #493 slice and has fresh built-app evidence at /Volumes/LEXAR/Codex/worldos-built-app-playtest/scripted-provider-app-proof-20260601T052807.

This confirms the intended split:

1. Deterministic smoke: scripted provider proves built app ⇄ viewer ⇄ engine ⇄ /move wiring with private art and no model/provider spend.
2. Provider playtest: Codex/OpenClaw/Claude path proves real-provider gameplay quality and tool-contract behavior; Codex DM should avoid safety-cancelled setup/social tool calls #479 remains active until a current-head provider trace is clean or properly bucketed.
3. Full RRI: release: rerun clean non-partial 5-persona RRI after gate hardening #466 should wait for complete persona/evidence inputs and support-VM preflight; neither [codex] Add Codex DM built-app provider path #475 nor [codex] Add dev-gated scripted smoke provider #494 by themselves are release verdicts.

Important evidence detail: the #494 scripted proof included real image-route probes (location:loc-lower-city, Abby portrait, class art), so it is stronger than the earlier #475 Codex proof on private-art rendering.

[100yenadmin]

PR #496 is the recovered gate-split/RRI-metadata slice on current main. It adds explicit RRI gate-split metadata: deterministic built-app smoke and short real-provider playtest are evidence inputs, while only full_five_persona_rri is a release verdict. It also strengthens built-app playtest output with failure_bucket/failure_detail/original_result fields and score_failed classification for score.json red outcomes. Keeping this issue open until #496 merges and the next #466 sweep consumes the metadata.

[100yenadmin]

PR #496 merged as 080497e. RRI output now includes release_verdict_gate=full_five_persona_rri and gate_split_contract metadata that explicitly marks deterministic smoke and short real-provider playtests as evidence-only, not release verdicts. The next #466 sweep should consume this metadata and still fail if persona/evidence artifacts are missing or partial.