fix(rag): reverte config FTS p/ portuguese (regressão de eval)#149
Merged
Conversation
O PR #145 casou a config do query keyword com o índice (portuguese_unaccent), teoricamente correto. Mas `npm run eval` na VPS mostrou REGRESSÃO no golden set: Recall@5 0.95 -> 0.83, MRR 0.73 -> 0.60. Unaccentuar o query alarga o leg keyword e, via RRF + rerank, piora o ranking neste corpus (o "mismatch" era load-bearing). Volta as 3 ocorrências para plainto_tsquery('portuguese', $1) e fixa a decisão num teste + comentário (não re-"corrigir" sem rodar o eval). O fix nunca foi necessário para o caso 'ervilha' (palavra ASCII casa nas duas configs); a causa real do 'ervilha vazio' era o header ausente no re-index, já corrigida e mantida. build+test verdes (1267/1267). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Hotfix: reverte o fix de FTS (regressão de eval)
O PR #145 casou a config do query keyword com a do índice (
portuguese_unaccent) — teoricamente correto. Mas onpm run evalna VPS, pós-deploy, mostrou regressão no golden set:Unaccentuar o query alarga o leg keyword e, via RRF + rerank, piora o ranking neste corpus (o "mismatch" era load-bearing). Conforme o CLAUDE.md (eval vs baseline manda), reverte as 3 ocorrências para
plainto_tsquery('portuguese')e fixa a decisão num teste + comentário.O fix de FTS nunca foi necessário para o caso "ervilha" (palavra ASCII casa nas duas configs). A causa real do "ervilha vazio" era o header ausente no re-index, corrigida e mantida (unificação
indexSinglePage→indexDocument).npm run build+npm testverdes (1267/1267). Vou re-rodar o eval pós-deploy pra confirmar volta ao baseline.🤖 Generated with Claude Code