Skip to content

Import pilot-2, freeze guidelines v0.2, declare acceptance targets - #22

Merged
professor3333 merged 2 commits into
mainfrom
pilot/pilot-2
Nov 2, 2026
Merged

professor3333 merged 2 commits into
mainfrom
pilot/pilot-2

Conversation

@professor3333

Copy link
Copy Markdown
Owner

What

  • Origin model_draft_revised (record schema 1.2): a model draft changed with no annotation time recorded (API/script) is no longer counted as model_draft_edited, which now requires recorded working time.
  • pilot-2 (annotations/pilot/pilot-2.jsonl): 100 clean records: 22 manual, 52 model_draft_accepted, 26 model_draft_revised. All follow guidelines v0.1.
  • Gate rerun (docs/pilot_gate.md): UNCERTAIN (44) and UNRESOLVED (32) unsupported; ANY just supported (51); EXCEPTION still has no natural examples.
  • Pilot dev scores (reports/pilot/, 10 hand-labelled dev docs): rules-2 typed F1 0.095 [0.000, 0.333]; crf-pilot-2 0.231 [0.062, 0.417]. Reproduced byte for byte.
  • Guidelines v0.2 frozen for batch annotation; the pilot is documented as not migrated.
  • docs/targets.md: acceptance gates declared before any test result, judged on hand-labelled test records only.

No test-split records read; the test-access log is empty.

Checks

ruff, ruff format, mypy strict, pytest -m "not slow" (213 passed).

A model draft changed with no annotation time recorded (saved through
the Label Studio API or a script) is now `model_draft_revised`, not
`model_draft_edited`, which requires recorded working time. Record
schema 1.2 (additive).

pilot-2: pilot-1 after 43 flagged drafts were revised: 22 manual, 52
model_draft_accepted, 26 model_draft_revised; 100 clean records. Gate
rerun on pilot-2: UNCERTAIN and UNRESOLVED unsupported, ANY just
supported (51). CRF retrained as crf-pilot-2; dev scores regenerated
for the manual and all slices.
Guidelines v0.2 are frozen for batch annotation. The pilot is recorded
as labelled under v0.1 and not migrated (pilot-2 revisions cite v0.1).

docs/targets.md declares release gates before any test result: judged
on hand-labelled test records only, with gates on false-prerequisite
rate, reversed meaning, typed and span F1, role and group F1, empty-
document exact match, calibration and unflagged precision, plus
release-to-release regression tolerances.
@professor3333
professor3333 merged commit f58e168 into main Nov 2, 2026
2 checks passed
@professor3333
professor3333 deleted the pilot/pilot-2 branch November 2, 2026 08:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant