feat(corpus): ask for cross-article relationships + wire the scope question (27 S.1/S.2)#176
Closed
bryanmatthewsimonson wants to merge 1 commit into
Conversation
…estion (27 S.1/S.2)
Phase 27 slices S.1/S.2, from the maintainer report that corpus
analysis never suggests cross-article claim relationships.
Root cause (JOURNAL 2026-07-16): the reduce deliverables list omitted
proposals entirely — the only prompt mention of relationships was the
20.6 restriction ("OMIT rather than guessing"), so the optional slot
predictably came back empty. And the case scope question NEVER reached
any LLM call: synthesis-block reads dossier.scope.question, which
buildCaseDossier never emitted.
- Reduce prompt: proposals join the deliverables list with an
affirmative cross-article directive ("scan the claims index for
pairs from DIFFERENT articles… propose EVERY such pair"), the
human-gate stated in-prompt, knots excluded from re-proposal. The
20.6 never-guess rule and the no-verdict rule stay verbatim.
- Digest: claims carry a short per-article key (art: A1/A2…) plus an
articles key→hash map, replacing the 64-hex hash on every entry —
cross-article pairs become identifiable at a glance and the digest
shrinks.
- buildCaseDossier now emits scope.question from the case entity's
authored_fields.scope_question — synthesis map/reduce (and the
Phase-26 hypothesis-edge pass via its case-view read) actually
receive the author's framing.
- CORPUS_PROMPT_VERSION → corpus-v2 with the institutionalized
discipline note (20.6 changed the prompt without bumping it, so
broken-era briefs never showed the stale chip). All stored briefs
correctly go stale; recompute stays human-triggered.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This was referenced Jul 16, 2026
Owner
Author
|
Landed on main via the #183 consolidation merge — this PR was auto-closed by the base-branch deletion cascade, but its commits are all in main. |
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.
Summary
Phase-27 slices S.1/S.2 — the first of the fixes from today's investigation into "corpus analysis never suggests cross-article claim relationships." Stacked on #175 (the Phase-26 chain); merge in order.
Diagnosis (verified, in JOURNAL)
synthesis-blockreadsdossier.scope.question;buildCaseDossiernever emittedscope. Every synthesis ran unsteered.Fix
art: "A1") + anarticlesmap — pairs identifiable at a glance, digest smaller.buildCaseDossieremitsscope.questionfromauthored_fields.scope_question.CORPUS_PROMPT_VERSION→corpus-v2with the institutionalized bump discipline — all stored briefs correctly go stale (decision D2); recompute stays human-triggered.Tests: 1,790 green (3 new) · build clean · lint 0 errors.
Next
S.3/S.4 (visible zero-proposals + durable triage + local link chips) follows; the forensic F ladder after that.
🤖 Generated with Claude Code