Repository navigation
docs(dr-math): structural DR to reusable mathematical compilation [0/5] #1360
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
| name: Citations have not decayed | |
| on: | |
| pull_request: | |
| paths: | |
| # The gate's inputs are BOTH the citing files and the cited ones -- a | |
| # prepend to `.claude/board/EPIPHANIES.md` moves targets under citations | |
| # that live in other, unchanged files on unchanged lines. That is the | |
| # motivating case, and it is why this runs on the whole board/plan tree | |
| # rather than on the diff. | |
| - .claude/plans/** | |
| - .claude/board/** | |
| # ...and the gate itself. | |
| - .claude/tools/citation_decay.py | |
| - .github/workflows/citation-decay.yml | |
| concurrency: | |
| group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }} | |
| cancel-in-progress: true | |
| permissions: | |
| contents: read | |
| jobs: | |
| citation-decay: | |
| runs-on: ubuntu-latest | |
| steps: | |
| - uses: actions/checkout@v4 | |
| with: | |
| # The gate builds a worktree at the merge base and compares verdicts | |
| # against it, so it needs both sides of the history, not a depth-1 tip. | |
| fetch-depth: 0 | |
| - name: Self-test the gate | |
| # A gate whose own falsifiers are broken cannot be trusted to report on | |
| # anything else. Both halves are asserted, in both the single-revision | |
| # and the two-revision arm: it FIRES on a decay this PR introduced or | |
| # exposed, and STAYS SILENT on the pre-existing backlog. | |
| run: python3 .claude/tools/citation_decay.py --self-test | |
| - name: Check for citation decay introduced by this PR | |
| # A REGRESSION comparison, not a diff filter. `--since` scans the whole | |
| # corpus at the merge base and at HEAD and fails ONLY where a citation | |
| # is decayed now and was not decayed at the base. | |
| # | |
| # The first version of this step filtered to the lines a PR ADDED, and | |
| # was structurally blind to its own motivating case: prepending to | |
| # EPIPHANIES.md decays citations in files the PR never touched. It also | |
| # computed a file list with `|| true`, so an unresolvable merge base | |
| # became an empty list and a GREEN run -- bypassing the script's own | |
| # fail-closed handling. Both are gone: there is no file list to compute, | |
| # and the script fails closed on any git failure. | |
| # | |
| # The pre-existing backlog (measured 148 decays on 2026-09-04, up from | |
| # 124 as the corpus grew) is reported and never fails the job. A gate | |
| # that fires on everything carries exactly as much information as one | |
| # that never fires. | |
| run: python3 .claude/tools/citation_decay.py --since "${{ github.event.pull_request.base.sha }}" |