Archive: superseded ProtNLM2 bootstrap files
These files are kept only as a record of how the first ARGO-ProtNLM-50 prediction reviews were
seeded, plus the first slide deck. They are superseded. Do not run or cite them as current results. Archived on 2026-09-27.
| File | What it was | Why it is archived |
|---|---|---|
generate_prediction_reviews.py |
One-time script that wrote genes/<ORG>/<GENE>/<GENE>-protnlm-predictions-review.yaml from the bench50 CSVs |
It hardcodes per-accession verdicts that contradict the current curated YAMLs, and by default it overwrites existing YAMLs (only --only-missing skipped them). Its __main__ now exits with an error instead of running. |
bench50_novel_review.csv |
Early agent triage of 32 "novel" ARGO-50 predictions, with its own labels (e.g. TRIVIALLY_DERIVABLE, PHMMER_TRANSFER_PLAUSIBLE) |
Its labels are not the VDCL categories, and many of its judgments were later revised in the per-gene reviews. |
protnlm_evaluation_slides.md, _embedded.md, .html, .pptx |
First slide deck, built from the pre-release XML export and the early ARGO-50 triage | Its counts and several case-study verdicts (e.g. "trivially correct" for the wheat patatin, now NPI) predate the current reviews. Replaced on 2026-10-19 by an HTML slide deck (claude.ai artifact) whose figures are generated from benchmark-funnel.json. Its image links were rewritten to ../slides_figures/... so the archived Marp source still renders; the archived .html embeds its images. |
The calls of record are the per-gene *-protnlm-predictions-review.yaml files. Current totals come
from build_benchmark_summary.py and are reported in the
cross-cohort results.