lukstei released slop-grader, a rule-based text grader. It uses Jev to check every rule against every line in parallel. It caches unchanged lines to avoid repeated API calls. It batches lines to fit model context limits. It supports custom rulesets in Markdown or JSON. It outputs terminal reports for agents and JSON for pipelines. It works with TypeSafe AI and OpenRouter. It costs cents to grade hundreds of lines. It targets AI writing habits, tech docs, grammar and more. It is open source under MIT license.

• 16 lines, 53 rules, $0.0053
• 300 lines, 42 rules, $0.076
• 255 lines per batch max