Evidence ledger

Tagged transcript baseline, held-out split, corpus 2026-09-07.1

MeasuredEV-001

Produced by running code in this repository and preserved as an artifact.

Scope
Twelve synthetic held-out transcripts, run once with the deterministic Tagged transcript baseline.
Observation
Action precision 1.00 (8 of 8 predicted actions matched). Action recall 0.24 (8 of 33 gold actions). Owner accuracy 0.00 (0 of 6 matched actions that have an explicit gold owner). Decision status correctness 0.07 (1 of 14 matched decisions). Zero critical defects. Zero of twelve cases met the acceptance criteria.
Limitations
Twelve cases is a small sample. The gold annotations are drafts that no human has reviewed, so every number here could move after review. This describes one narrow tag-reading method and nothing else.
Source
experiments/runs/tagged-transcript-baseline--held-out--2026-09-07.1/report.json
Source date
2026-09-07
Retrieved
2026-09-07
Reviewer status
Pending review. Written down, not yet checked by a person.

How measured records are produced