docs/RELEASE_VALIDATION.md

Release Candidate Validation — v189.7

This archive is based on agentic-space-factory-v188-7.zip and hotfixes partial progress plus Space log diagnostics before deployment for the v188 sequence.

Validation date

2026-06-06

Automated checks

• Pytest suite: expected full pass from the archive root.
• JavaScript syntax:
  • web/static/app.js
  • web/static/runs.js
  • web/static/progress.js
  • web/static/api.js
  • web/static/components.js
• Python compile check:
  • app.py
  • src/*.py

Critical v188 regression checks

• full_inference_candidate_health_passed and health_only are partial validation states, not green full-success states.
• /health validation from gradio_client maps to health_passed=true in the inference gate and eval record.
• Generation smoke tests prefer INFERENCE_CONTRACT.json payloads, then Gradio schema introspection, then generic fallbacks.
• Smoke test payloads include required text/image arguments such as negative_prompt when discovered in the schema.
• Smoke failures classify validator request errors separately from generated-app runtime errors.
• Eval records use schema 1.2 and expose process_completed, verdict, outcome, validation, failure_owner, and failure_reason.
• Backend eval archive publication is observable through eval_publish_status.json and is not confused with run-local eval_record.json creation.
• Active Run hides cancel actions for terminal/partial states and renders partial validation, eval publish, report, recovery and hardware fallback consistently.
• The default ASF app/eval version is centralized in src/version.py and defaults to v189.7.
• Space log collection writes logs/space_logs_index.json, logs/space_runtime_snapshot.json, and logs/space_log_diagnostics.json.
• Runtime/build logs are marked unavailable with a clear reason when neither SDK nor REST log collection works.
• Backend progress payloads keep partial terminal states visually partial/warn, not success.
• Space log diagnostics only extract first_error from real available runtime/build logs, never from [ASF_LOG_UNAVAILABLE] placeholders.

Live-run scenario covered

The v188 sequence was driven by the live run:

universal-20260606-181620-5f6004b8

Expected honest verdict for that class of run:

Completed with partial validation.
Space created and repaired.
/health passed.
Generation smoke failed only if the real generation call still fails after schema-aware payload construction.
Eval archive publication must be separately observable.

Manual validation recommendation

After deploying this archive, run the same Z-Image style scenario again and check:

• Active Run top badge says partial only when generation remains unverified.
• The smoke payload file includes negative_prompt.
• health_passed is true when /health passes.
• The eval archive bucket receives a dated record, or eval_publish_status.json explains why it did not.

v189.7 validation

Validate that timeline_model is present on run view payloads and correctly distinguishes full-inference success from partial validation. Eval archive sidebar restoration from v188.9 must remain intact alongside the per-run Eval archive card.

v189.7 semantic progress validation

• Active Run progress uses timeline_model.progress when available.
• The progress bar represents process completion; the badge/result represents the verdict.
• A 100% partial validation renders as warning/amber, not success/green.
• Legacy payloads without timeline_model continue to render through the previous fallback.

v189.7 mobile timeline validation

• Active Run renders the canonical timeline_model as phase steps plus timelinePhaseDetails.
• Mobile layouts switch the canonical timeline to a vertical single-column list.
• Phase details expose the active phase summary, details, and “Needs attention” warnings without changing worker/runtime behavior.
• Legacy timeline fallback remains available when no canonical model is present.

v189.7 timeline state validation

• Terminal full-inference success runs must not leave early phases such as Start in running because of historical started events.
• Resolved log collection warnings must not keep Deploy in warning when the final logs index reports full quality.
• The active phase should focus the most useful terminal proof, especially Live test for full-inference success, while non-blocking warnings remain grouped separately.

v189.7 runtime sanitation validation

• README python_version is quoted and normalized to a safe value.
• torch is added when app.py imports torch.
• failed timeline phases prefer confirmed eval publication and final build blockers.