feat(spec): a claims-verification pass measures the spec against the tree - #958
Open
leeovery wants to merge 1 commit into
Open
feat(spec): a claims-verification pass measures the spec against the tree#958leeovery wants to merge 1 commit into
leeovery wants to merge 1 commit into
Conversation
…tree Both review agents compare documents against documents, so a false empirical claim present in the spec and its source read as a perfect fidelity pass — seven such claims survived six observed cycles, and the one pass that found them was an off-remit rewrite. A third agent now opens every cycle: it re-runs each load-bearing claim's recorded command (specification-format now records measurements as command → result), verdicts hold/fail/unreproducible, and classifies a failure by where it lives — source-carried failures become Source defect findings that route back, spec-only ones carry the corrected measurement. The review loop's two prose menus migrate to the engine (render spec-review-gate) per the touching-a-file rule, with tests and a simulation extension. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
leeovery
force-pushed
the
spec-claims-verification
branch
from
August 19, 2026 14:20
a4a1891 to
f763baf
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Why
Idea 40: both review agents compare documents against documents. A false empirical claim present in both the spec and its source is a perfect fidelity match — seven such claims survived six observed review cycles, and the only pass that caught them was an unauthorised remit rewrite. For a spec whose content is largely codebase measurements, nothing prescribed ever touched ground truth.
What
workflow-specification-review-claims: opens every review cycle (Phase 1). Collects the spec's empirical claims, re-runs each load-bearing one's recorded command (never trusting the documents, prior cycles, or asserted-as-verified figures), and verdicts hold / fail / unreproducible. A failing claim that lives in a source becomes a Source defect finding (routes back via feat(spec): review findings that indict a source route back, never apply #957's lane); a spec-only failure carries the corrected measurement; unreproducible load-bearing claims must be restated measurably or removed. Read-only measurement is a hard rule.spec-review.mdbecomes three sequential phases — claims first, so a false-in-both claim routes before fidelity review reads it as a match. Sections relettered C–G.specification-format.md: empirical claims are recorded with their measuring command (cmd → result) at their home — what makes re-verification mechanical; open-decision markers are banned from the document outright.render spec-review-gate --variant continue|reloop, with render tests and a pipeline-simulation extension;convergence-analysis.mdandspec-completion.mdpick up the third stream.Stacks on #957. Part 4 of the idea-40 stack.
🤖 Generated with Claude Code