Skip to content

Claude/eval chain sensitivity mu4kop - #36

Merged
sadacca merged 2 commits into
mainfrom
claude/eval-chain-sensitivity-mu4kop
Jun 24, 2026
Merged

Claude/eval chain sensitivity mu4kop#36
sadacca merged 2 commits into
mainfrom
claude/eval-chain-sensitivity-mu4kop

Conversation

@sadacca

@sadacca sadacca commented Jun 24, 2026

Copy link
Copy Markdown
Owner

No description provided.

claude added 2 commits June 24, 2026 12:05
Run filenames previously encoded only eval+judge (model_sig = eval_judge), so
runs differing only in triage model overwrote each other — and triage is the
model driving the Pass 0/1/3 work being tuned. Extend the signature to all
three pipeline models, in pipeline order, and surface every run for comparison.

- bias_eval / batch_eval: model_sig is now triage_eval_judge. Combined with the
  mode suffix added earlier, the full run filename is
  {article_id}_{version}_{triage}_{eval}_{judge}_{mode}.json — any single-axis
  difference lands in a distinct file.

- ingest._already_evaluated: take `mode` and build the exact path including it,
  so batch --skip-cached matches the real filename (it had silently stopped
  matching once mode entered the name).

- article_eval page: replace the mode-only toggle with a run picker keyed on the
  full (mode, triage, eval, judge) signature. The dropdown lists every stored
  run by full signature so any mode and any model chain can be compared; running
  the same mode with different models now produces a new run instead of being
  skipped. Completion polling switched to a run-file count target (robust to
  same-mode re-runs). Downloads carry a full-signature slug.

- scripts/migrate_run_filenames.py: one-off migration. Reconstructs the
  compliant name from each record's stored models+mode and renames; deletes
  legacy records whose triage/judge can't be recovered (pending re-run). Applied
  to data/processed (17 renamed, 5 legacy deleted) and index rebuilt.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01669jmknuhhZY3vBqwpDWvF
…sitivity-mu4kop

# Conflicts:
#	data/processed/bias_frequency.json
#	data/processed/model_provenance.json
#	data/processed/stats.json
@sadacca
sadacca merged commit 59587e8 into main Jun 24, 2026
1 check passed
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.

2 participants