Skip to content
Permalink

Comparing changes

Choose two branches to see what’s changed or to start a new pull request. If you need to, you can also or learn more about diff comparisons.

Open a pull request

Create a new pull request by comparing changes across two branches. If you need to, you can also . Learn more about diff comparisons here.
base repository: binarybison/proofreader
Failed to load repositories. Confirm that selected base ref is valid, then try again.
Loading
base: main
Choose a base ref
...
head repository: binarybison/proofreader
Failed to load repositories. Confirm that selected head ref is valid, then try again.
Loading
compare: jev-support
Choose a head ref
Checking mergeability… Don’t worry, you can still create the pull request.
  • 2 commits
  • 22 files changed
  • 2 contributors

Commits on Sep 20, 2026

  1. Experimental Jev (TypeSafe System One) decision layer -- not for merge 

    Adds an optional layer in which Jev answers narrow typed questions about
text the pipeline has already produced, code owns thresholds, and every
decision is persisted. Gates: domain, route, fp-filter, cx-gate (opt-in,
reorder only), quote-check, adjudication-check, match-results,
proof-changed. Stdlib-only client; falls back to Claude-only behavior.

Validated against the precursor pipeline's adjudicated corpus. Findings
(docs/jev-architecture.md, Obs. 1-8):
- Jev cannot judge a proof from the proof text (AUC 0.49-0.54, chance).
- Surface features of a proof do not predict flaws (max AUC 0.62).
- From the audit, Jev predicts counterexample-hunt success at 0.81 (the
  audit's own falsifiable flag: 0.72), but the arbiter confirms 79% of
  findings with a counterexample and 5% without, so skipping hunts loses
  flaws; the gate only reorders.
- Jev predicts the arbiter at 0.945, but the label is circular and the
  errors dismiss true positives; the arbiter is not replaced.
- Opposite framings of a question are not mirror images.

Decision (2026-09-19): keep this branch, do not merge. The decisions Jev
can make are not where the pipeline's cost lies, and the one that would
matter is beyond the current model. scripts/jev/validation/README.md has
the runbook, the aggregate baseline, and the criteria for revisiting.

Also gitignores .env/ (API keys).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
    @binarybison @claude
    binarybison and claude committed Sep 20, 2026
    Configuration menu
    Copy the full SHA
    bf19ddc View commit details
    Browse the repository at this point in the history

  2. Validation runbook: write summaries as .summary.txt so they are not g… 

    …itignored

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
    @binarybison @claude
    binarybison and claude committed Sep 20, 2026
    Configuration menu
    Copy the full SHA
    31aa732 View commit details
    Browse the repository at this point in the history
Loading