docs(design): approve Phase-26 designs — hypothesis map + structural counterfactual (H.0/CF.0)#169
Merged
Merged
Conversation
…or review) Two design docs for the speculative half of the corpus-analysis review — written for maintainer review, NOT to be built without sign-off (this PR is left unmerged deliberately). - HYPOTHESIS_MAP_DESIGN.md: competing answers to the case question with captured claims attached as supporting/undermining structural evidence, side by side, model never picks, NO score/probability (grep-tested). Promotes synthesis positions; local-derived first (dossier posture); LLM edge-suggestion via the codified gate/ground/filter/human-accept firewall. §6 names the three ways it could cross a red line and the guard for each. - COUNTERFACTUAL_DESIGN.md: the maintainer's 'Monte Carlo' ask reframed to clear the firewall — a STRUCTURAL 'what depends on this claim' over the existing case graph, counts-with-derivation never probabilities (a measurement, not an estimation, per TRUTH_ADJUDICATION §1). Deterministic graph walk, no LLM, no wire kind. §4 refuses the probabilistic version as a separate, constitution-amending decision not taken. - ROADMAP Phase 26 stub: safe wins shipped (T1.1-T1.3), these two design-only pending review. Docs only — no code, no wire. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017SZCGu5qxuF2QCuGyJrNcH
Flip HYPOTHESIS_MAP_DESIGN.md and COUNTERFACTUAL_DESIGN.md from draft to approved (2026-07-16) and resolve their open-questions sections with the maintainer's decisions: - Hypothesis map stays local-derived (no wire kind; H.5 deferred). - Neutral per-section edge counts are acceptable, never cross-compared (guard test required). - Verdict-state chips render as context with an explicit "verdict does not weight the edge" note. - Counterfactual: inline per-claim expander; CF.1 ships remove + negate; the probabilistic Monte Carlo refusal is confirmed. ROADMAP Phase 26 moves from design-only to in-progress (H.1–H.4 then CF.1–CF.2, one PR per slice). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This was referenced Jul 16, 2026
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
2 participants
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-26 slice H.0/CF.0: brings the two overnight design drafts from
claude/epistack-plan-review-krh2wrontomain, flipped to approved with the maintainer's decisions recorded, plus the ROADMAP Phase-26 entry moved to in-progress.docs/HYPOTHESIS_MAP_DESIGN.md— competing answers to a case's scope question with claim-levelsupports/underminesedges; computed on read, no new wire kind, no score/weight/probability anywhere (grep-tested), model never picks a winner. §8 decisions: local-derived v1 (wire kind stays deferred), neutral per-section edge counts OK but never cross-compared, verdict-state chips shown as context with an explicit "verdict does not weight the edge" note.docs/COUNTERFACTUAL_DESIGN.md— "what depends on this claim" as counts-with-derivation over the case graph; the probabilistic Monte Carlo remains refused (constitution-amending, not planned). §7 decisions: inline per-claim expander; CF.1 ships bothremoveandnegate.docs/ROADMAP.md— Phase 26 entry updated: designs approved 2026-07-16, ladder H.1–H.4 then CF.1–CF.2, one PR per slice.Docs-only; no code changes.
Next slices
H.1 (pure model + assembler + tests) follows in its own PR.
🤖 Generated with Claude Code