Skip to content

feat(spec): a source claim that fails measurement routes like a conflict - #955

Open
leeovery wants to merge 1 commit into
remit-rulefrom
incoherence-reality-trigger
Open

feat(spec): a source claim that fails measurement routes like a conflict#955
leeovery wants to merge 1 commit into
remit-rulefrom
incoherence-reality-trigger

Conversation

@leeovery

Copy link
Copy Markdown
Owner

Why

Idea 40, observed live: resolve-source-incoherence.md triggers only on source-vs-source disagreement, so a single internally-coherent discussion carrying seven false claims about the tree extracted cleanly — and the one falsehood construction did catch was patched in the spec alone as a "validated correction", leaving the completed discussion asserting it on disk.

What

  • Trigger widened (spec-construction.md §A and §B, the reference's attribution): source material disagreeing with the codebase or toolchain it describes routes exactly like a conflict — never patched in the spec alone.
  • New first-match classify branch in resolve-source-incoherence.md: re-run the measurement before classifying (an asserted-as-verified figure is not a measurement); value-only mismatch whose citing conclusions survive → lands in the owning document via §C with a one-line notify, no gate; a mismatch that undermines a conclusion → conversational stop that overrides auto, then the existing settle/gap ladder.
  • §C in-place repair clause for corrections that revise no decision, and the Trigger: line may now cite a failed measurement.
  • Construction principle (specification-principles.md): load-bearing tree claims are verified before extraction and extracted with their command and result.

Stacks on #954. Part 2 of the idea-40 stack.

🤖 Generated with Claude Code

The incoherence trigger named only source-vs-source disagreement, so a
discussion that was internally coherent but empirically wrong extracted
cleanly — observed on a real feature where seven false tree claims rode
into the spec and the headline one was 'corrected' in the spec alone,
leaving the completed discussion asserting it on disk. The trigger now
names disagreement with the codebase the sources describe; a new
first-match classify branch re-runs the measurement, lands a value-only
correction in the owning document with a one-line notify, and stops
(overriding auto) when the corrected value undermines a conclusion.
Construction principles gain the measure-before-extract duty, with the
command and result travelling into the spec.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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