Skip to content

Plain-language list fusion; two fresh holdouts (cobra, axios); baseline path-match fix - #136

Merged
ParsaVictor merged 3 commits into
mainfrom
phase-o/concept-holdout3
Oct 9, 2026
Merged

ParsaVictor merged 3 commits into
mainfrom
phase-o/concept-holdout3

Conversation

@ParsaVictor

Copy link
Copy Markdown
Owner
  • S1: short-question localization = RRF(packet order, whole-question ranking). R@3 ripgrep 0.208→0.500, click 0.750→0.917, cobra 0.792→0.875; fresh axios holdout unchanged (0.833 vs BM25 0.583).
  • concept-holdout3 (cobra, Go) and concept-holdout4 (axios, JS): gold locked before any run.
  • Eval fix: Python baselines matched gold with plain endswith; now segment-boundary. Published tables re-run: unchanged.
  • Reranker v2 rejected on the fresh holdout (0.792 → 0.708).

Fourteen benchmark sets unchanged. Details: contributions-log §8.6.

🤖 Generated with Claude Code

ParsaVictor and others added 3 commits October 9, 2026 10:16
…d before any run

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ld locked before any run

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…aseline path-match fix

- localization for a short question: RRF of the packet order and the whole-
  question ranking. R@3: ripgrep 0.208 -> 0.500, click 0.750 -> 0.917, cobra
  0.792 -> 0.875, this repo and the fresh axios holdout unchanged. List order
  only; packets unchanged.
- Two plain-language holdouts with gold locked before any run: cobra (Go) and
  axios (JavaScript), wired into benchmark-fast.sh (packet 0.833/0.357 and
  0.667/0.528).
- Python baselines matched gold with a plain endswith (`zsh_completions.go`
  "was" `completions.go`): now on a path-segment boundary. Published numbers
  re-run unchanged; only the new cobra set was affected (BM25 R@3 0.875).
- jina-reranker-v2 rejected on the fresh cobra holdout (R@3 0.792 -> 0.708).

Fourteen sets unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@ParsaVictor
ParsaVictor merged commit 2feca12 into main Oct 9, 2026
3 checks passed
@ParsaVictor
ParsaVictor deleted the phase-o/concept-holdout3 branch October 9, 2026 07:41
@ParsaVictor ParsaVictor mentioned this pull request Oct 9, 2026
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