A prose linter combining deterministic pattern checks with Jev Noul judgments and a local undecided band; paragraph mode derives locations from known block boundaries.
Jev returns semantic writing-tell probabilities. gate.py suppresses the undecided interval and applies a separate threshold for each tell; code, not Jev, supplies paragraph line positions.
Keeps ambiguous probabilities out of the finding list and avoids asking a second model call to select a line for each semantic tell.
Fixed-version documentation and implementation material were inspected on 2026-10-24. No real document grading or benchmark was reproduced. Historical pricing and timing in source are not current account guarantees.
Detect writing problems while separating exact text rules from semantic judgments.
Material analysed：
judge.py removes the second line-location inference. Paragraph mode gets a line number from a block boundary known to code. The README’s opening sample still shows an older two-call result, so use the implementation and current behavior section to interpret it.
Inspect the supporting source ↗gate.py suppresses probabilities in the configured lo-inclusive, hi-exclusive interval before considering a tell’s threshold. Borderline output need not become a reported violation.
Inspect the supporting source ↗Choose it when you can specify actionable writing tells. Keep a reviewer for intentional prose choices and avoid treating a detected pattern as proof of AI authorship.
Read detect_blocks in judge.py and the threshold branch in gate.py. Use the code to trace a probability inside the undecided interval and another above a tell threshold; only the latter should be emitted as that finding.