feat(v0.6.1): real "detailed" answer style — reproduce source detail, not a summary#1074
Merged
Merged
Conversation
… not a summary Operator: after an image's full text was ingested, asking about it in detail only ever got a one-line summary back. Root cause: the document's full content (e.g. a schedule table) is stored + retrieved fine, but `response_style="detailed"` collapsed to NATURAL (the v2 redesign mapped brief/standard/detailed→NATURAL), so there was no style that tells the synth LLM to REPRODUCE the source content instead of condensing it. The chat UI also never sends response_style — the engine auto-classifies it, and the classifier only emitted terse/natural. Fix (no chat.js change — keyword-driven, automatic): - answer_style_classifier: FAST_DETAILED_PATTERNS checked first — a query asking for full detail (상세히/자세히/구체적으로/낱낱이/전체 내용/원문/ 그대로 · in detail/verbatim/full table…) classifies as "detailed". - response_style: add a real DETAILED_PRESET (rule_text: reproduce the relevant source verbatim — keep tables/rows, every number/date/count/ place; do not summarise/compress) and map "detailed" → it. brief/ standard still → NATURAL; empty/unknown → NATURAL (byte-identical, no regression). Verified e2e: "전장연 7월 1일~2일 전체 일정 상세히 알려줘" → [STYLE] …→ detailed (fast) → the answer is now the FULL schedule table (7 rows: times, 인원/휠체어 counts, 서울지방조달청/잠수교/한국재정정보원…) vs the prior one-line summary. - core/answer_style_classifier.py + core/response_style.py - tests updated (brief/standard→natural kept; detailed→DETAILED_PRESET) + new detail-request classification test core/retrieval + core/graph traversal + core/reasoning 0 lines (these are the response-style/format layer). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01B7gaaBkpM3xYTzxhYrGHnR
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 subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
Summary
After an image's full text was ingested, asking about it in detail only ever got a one-line summary back.
Root cause: the document's full content (a schedule table) is stored + retrieved fine, but
response_style="detailed"collapsed to NATURAL (the v2 redesign mapped brief/standard/detailed→NATURAL), so no style told the synth LLM to reproduce the source instead of condensing it. The chat UI also never sendsresponse_style— the engine auto-classifies it, and the classifier only emitted terse/natural.Fix (no chat.js change — keyword-driven, automatic):
answer_style_classifier:FAST_DETAILED_PATTERNSchecked first — a query asking for full detail (상세히/자세히/구체적으로/낱낱이/전체 내용/원문/그대로·in detail/verbatim/full table…) →"detailed".response_style: add a real DETAILED_PRESET (rule_text: reproduce the relevant source verbatim — keep tables/rows, every number/date/count/place; do not summarise) and mapdetailed→ it.brief/standardstill → NATURAL; empty/unknown → NATURAL (byte-identical, no regression).Verification (e2e)
전장연 7월 1일~2일 전체 일정 상세히 알려줘→[STYLE] …→ detailed (fast)→ the answer is now the FULL schedule table (7 rows: times, 인원/휠체어 counts, 서울지방조달청/잠수교/한국재정정보원…), vs the prior one-line summary.test_response_style27/27 (updated: brief/standard→natural kept, detailed→DETAILED_PRESET),test_answer_style_classifier17/17 (+detail-request test),test_measurement_critical_surfaces33/33.How to use
Just phrase the question with a detail cue — "상세히", "전체 내용", "원문 그대로", "구체적으로", "in detail". No UI toggle needed.
Quality delta
Quality delta: exempt (label: feat) — response-style/format layer (verbosity), opt-in, default path byte-identical;
core/retrieval+core/graphtraversal +core/reasoning0 lines.🤖 Generated with Claude Code