# FrontierLab — release & submission checklist **Target public release:** v0.40 on main (runtime code from merge `6f0f7d2`, PR #37), redeployed to the existing HF Space from the latest main merge and runtime-verified. **Status 2026-11-03 Asia/Taipei:** submitted on the Cerebral Valley portal on 2026-11-02 (Indie Developer Track, Push the Frontier, team FrontierLab, solo); demo video public at https://youtu.be/agYFr2HgZ5Y (3:03). The boxes below are kept as the original working checklist. **Public event deadline rechecked 2026-11-01 Asia/Taipei:** Nov 4, 2026 at 9:00 PM ET = **Nov 5, 2026 at 10:00 AM Asia/Taipei**. Organizer dates may change. Official public references: - https://dev.meta.ai/events/global-hackathon - https://cerebralvalley.ai/e/meta-ai-developer-hackathon ## A. Repository freeze - [ ] All intended PRs are merged. - [ ] Main Git SHA is recorded. - [ ] Main test workflow is green. - [ ] README version matches `frontierlab/version.py`. - [ ] README test count points to a real CI run. - [ ] No open PR is required for the recorded demo. - [ ] No secret/token/API key is committed. - [ ] Historical run records are left immutable. ## B. Hugging Face deployment **Current deployment reference (2026-11-03 Asia/Taipei):** the deployed baseline is **v0.40**, synced to the latest main merge by the deploy workflow (407 software tests passed in the workflow before upload). The public Space API then reported runtime stage `RUNNING`; the served [`deployment.json`](https://huggingface.co/spaces/Jaycee766/FrontierLab-Meta-Hackathon-2026/blob/main/deployment.json) records the exact source SHA and per-file hashes. Earlier v0.40 deploys: `6f0f7d2` → Space `94bc1eb` (run 41408148915), `21f9092` → `714c31d` (run 41412146919), `a0a6a51` → `8f3db35` (run 41442246144), `1012ce5` → `6041356` (run 41448120728), all 2026-11-02, then a docs-only hand edit `f339cbf` on 2026-11-03 that did not regenerate `deployment.json`, then `e6a81fa` (PR #42) → `ec5d057` (run 41596184744, 2026-11-03 15:25). Main CI run 41408148161 passed 407 software tests and matched main canary run 41408148311 is HEALTHY with advisory DRIFT_REVIEW_REQUIRED / HEALTHY_DRIFT (`requirements.txt` hash differs from the accepted Nov 1 baseline). Any later merge needs one more workflow deploy before claiming exact Space/main SHA parity; do not hand-edit Space files, because that leaves `deployment.json` stale. The deployment workflow is manual (`workflow_dispatch`). - [ ] Manually trigger **Deploy FrontierLab to existing HF Space** from current main. - [ ] Confirm the deployment workflow's software tests pass. - [ ] Confirm `deployment.json.version` equals the current central version. - [ ] Confirm `deployment.json.source_sha` equals the intended main SHA. - [ ] Confirm the HF Space commit receipt matches the new deployment. - [ ] After the image rebuild, verify runtime status is RUNNING. - [ ] Verify Streamlit health endpoint responds successfully. - [ ] Verify rendered UI shows the current version and source marker. - [ ] Verify CPU Basic hardware was not changed. - [ ] Verify no NVIDIA credential or model opt-in was transferred to the Space. - [ ] Preserve a new dated deployment receipt. Do not equate "source committed to the Space repo" with "runtime verified healthy." ## C. Demo-state verification Before recording: - [ ] Published-literature pilot loads. - [ ] RECORDED REAL META RUN panel displays. - [ ] Gate-ablation replay shows 8 decisive / 4 abstaining / 0 new post-gate abstentions. - [ ] Completed two-claim human-review record displays. - [ ] Reviewed Skeptic → Hypothesis record is available. - [ ] Method Evaluation shows M2 selected. - [ ] Planning-readiness panel shows data/code/environment open. - [ ] Planning-readiness panel shows Planning approved = NO. - [ ] Evidence Graph renders or provides graph.json. - [ ] Trajectory Audit Graph renders its statistics and provides trajectory.json. - [ ] Complete provenance JSON downloads. - [ ] No panel labels a recorded replay as live inference. ## D. Optional experiments These improve the evaluation story but are not allowed to destabilize the submission. ### 12-case researcher review - [ ] Download reviewer packet. - [ ] Edit only labels and rationales. - [ ] Upload it through the app. - [ ] Verify source/case binding passes. - [ ] Download canonical review packet. - [ ] Report researcher corrections as intervention, not independent accuracy. ### Two-call prompt-level gate ablation - [ ] Trigger only if NVIDIA NIM usage is explicitly authorized. - [ ] Confirm exactly two calls were attempted. - [ ] Confirm same model + same fixed payload. - [ ] Preserve the artifact. - [ ] Validate result before adding it to README/video. - [ ] Do not retry automatically if the run fails. ## E. Stage 5 decision Current default: - [x] M2 selected. - [x] Method Evaluation current. - [x] Planning readiness contract current. - [ ] **Researcher explicitly approves Stage 5 Planning.** - [ ] Planning Agent runs. - [ ] Draft plan reviewed. - [ ] Individual plan steps pass Evidence Graph gate and researcher approval. If Stage 5 is not approved before submission, say exactly that. A visible human boundary is part of the product design. ## F. Video - [ ] Target approximately 3 minutes unless the logged-in portal specifies another requirement. - [ ] Use the demo runbook. - [ ] Show at least one genuine Meta execution record. - [ ] Show a gate/human decision, not only happy-path output. - [ ] Show one provenance view. - [ ] Show the Stage 5 readiness/approval distinction. - [ ] Check text readability at the final upload resolution. - [ ] Do not use unsupported "autonomous scientist" language. ## G. Submission form Public pages do not expose every logged-in final-submission field. - [ ] Open the authenticated participant portal. - [ ] Recheck mandatory fields and accepted link/video formats. - [ ] Select Indie Developer track. - [ ] Select / describe Push the Frontier accurately. - [ ] Paste final project description only after matching it to the shipped revision. - [ ] Add public GitHub link. - [ ] Add verified public HF Space link. - [ ] Add demo video link/file as required. - [ ] Accept any required rules/waivers. - [ ] Submit before the portal deadline. - [ ] Capture submission confirmation. ## H. Claims audit Before submission, search the README/video/copy for these risky phrases: - "scientifically verified" - "autonomous scientist" - "accuracy" (unless independent gold actually exists) - "live" next to recorded replay - "deployed" when only a source upload is known - "completed plan" when Stage 5 remains unapproved Replace them with the narrowest true statement: software-verified, model-executed, researcher-reviewed, source-anchored, prepared, or scientifically unverified.