Rubrics are an editorial choice
The README’s custom rules demonstrate that “slop” is defined by the ruleset, not a universal model property. A technical-document rubric should differ from a narrative-writing rubric.
Inspect the supporting source ↗REPOSITORY ANALYSIS
slop-grader
A configurable writing-grading CLI that turns line rules and document rubrics into typed flags and scores for an editor or revision agent.
E2 · source reviewedPublic material or source evidence was reviewed. A fixed-version source review requires checking a pinned revision; this is not a reproduction, performance endorsement, or safety guarantee.
- Stars
- 22
- Forks
- 2
- License
- MIT
- Last commit
- 2026-09-22
What Jev does here
Jev evaluates the declared writing rules and document rubrics; the grading output is input to a separate review-and-revision workflow.
Makes the rule and its flagged evidence inspectable before a generative agent proposes edits.
Fixed-version documentation and implementation material were inspected on 2026-10-24. The grading CLI and editing agent were not run. Model grades are not evidence that a document was AI-written.
What task does this project fit?
Grade a document against explicit writing rules before asking a generative agent to revise it.
Material analysed：
From input to action
- A ruleset defines line-level violations and document-level rubrics.
- Jev produces typed flags and grades; the CLI exposes them as review material.
- A separate agent proposes edits, which the user reviews before applying.
Implementation details that matter
The wire tests prove mapping, not detection quality
jev.test.ts injects fixed responses for Noul, Score and Choice and checks request/answer handling. These synthetic values cannot establish real false-positive rates.
Inspect the supporting source ↗When to choose it
Useful for repeatable editorial checklists. Keep specialized terminology and deliberate stylistic choices open to human exceptions instead of automatically deleting every flagged line.
How to inspect it yourself
Inspect one line rule and one document rule in README, then compare them with the typed response fixture in jev.test.ts. Verify that each grade’s meaning comes from the declared criteria, not from a generic quality score.
Limits of this review
- The grading CLI and editing agent were not run.
- Model grades are not evidence that a document was AI-written.