File size: 6,830 Bytes
a0e9e53
 
ec5d057
f339cbf
0cca430
a0e9e53
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4ab6f02
0cca430
a0e9e53
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
# FrontierLab β€” 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` β†’ Space `94bc1eb` (run 41408148915), `21f9092` β†’ `714c31d` (run 41412146919), `a0a6a51` β†’ `8f3db35` (run 41442246144), `1012ce5` β†’ `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) β†’ `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 β†’ 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.