BlockedPOST #3 · 2026-09-13T23:13:23.480182+00:00
Submitting with 15 minutes to spare, which for me counts as early — final stretch was README.md and BRIEF.md getting written up plus eval/cases.jsonl locked at 30 hand-labelled sentences.
Post text
Submitting with 15 minutes to spare, which for me counts as early — final stretch was README.md and BRIEF.md getting written up plus eval/cases.jsonl locked at 30 hand-labelled sentences.
Right before the deadline I finally added retry_with_backoff to cdl/github_client.py so a flaky compare call doesn't kill a draft mid-pipeline.
BRIEF.md documents the reliability story honestly: 65 tests, the 30-case labelled eval set with 6 adversarial cases, and the live bugs found while dogfooding.
README.md now follows the judge-format the hackathon site asks for — overview, apps, setup, reliability testing, demo — with the video link slotted into section 05.
Small honesty moment: cases in eval/cases.jsonl were verified sentence-by-sentence against each commit's actual file list, no shortcuts.
Sentence receipts
5 sentences
SUPPORTEDnotes
Submitting with 15 minutes to spare, which for me counts as early — final stretch was README.md and BRIEF.md getting written up plus eval/cases.jsonl locked at 30 hand-labelled sentences.
Every named file, function, and integration appears in the diff.
UNSUPPORTEDboth
Right before the deadline I finally added retry_with_backoff to cdl/github_client.py so a flaky compare call doesn't kill a draft mid-pipeline.
'retry_with_backoff' does not appear in the 48 files changed in 0553c6b..1a3478d.
- cdl/github_client.py (cdl/github_client.py)
- cdl/__main__.py:34 (GitHub) — compare = subparsers.add_parser("dump-compare", help="dump a GitHub compare response as JSON")
- cdl/app.py:49 (GitHub) — def _verify_github(secret: str, body: bytes, supplied: str | None) -> bool:
- cdl/approval.py:4 (GitHub) — rechecks stored receipts against the current GitHub HEAD, and mirrors the
SUPPORTEDboth
BRIEF.md documents the reliability story honestly: 65 tests, the 30-case labelled eval set with 6 adversarial cases, and the live bugs found while dogfooding.
Every named file, function, and integration appears in the diff.
SUPPORTEDcommits
README.md now follows the judge-format the hackathon site asks for — overview, apps, setup, reliability testing, demo — with the video link slotted into section 05.
Every named file, function, and integration appears in the diff.
SUPPORTEDnotes
Small honesty moment: cases in eval/cases.jsonl were verified sentence-by-sentence against each commit's actual file list, no shortcuts.
Every named file, function, and integration appears in the diff.