AntiRot v0.2.0

AntiRot v0.2.0

This release turns AntiRot into a more defensible Markdown evidence checker instead of a thin line-by-line heuristic pass.

Highlights

• paragraph-aware Markdown parsing instead of raw per-line scanning
• inline links, DOIs, arXiv ids, and footnote references count as evidence anchors
• references can now resolve from the same Markdown document when a References section is present
• unresolved citations no longer count as evidence when no references source is loaded
• common numeric identifiers like Section 3, GPT-4, and [32, 64] no longer trip claim checks
• new absolute-claim warning for universal or zero-risk wording
• fenced code blocks are ignored instead of producing false positives

Quick check

python3 -m pytest -q
python3 -m antirot.cli lint examples/sloppy_paper.md --references examples/references.md --format text

AntiRot v0.1.1

AntiRot v0.1.1

This patch release cleans up the public package and the first-run CLI experience.

Highlights

• clearer README with a direct tool-first pitch
• safer antirot init starter config
• clean error for missing references files
• removal of internal launch-planning material from the public repo

Quick check

python3 -m pytest -q
python3 -m antirot.cli lint examples/sloppy_paper.md --references examples/references.md --format text

AntiRot v0.1.0

AntiRot v0.1.0

AntiRot is a local-first evidence-hygiene linter for AI-written research artifacts.

It catches:

• unsupported claims
• numeric claims without citations
• citation drift
• hype language without evidence
• comparative claims without benchmark grounding
• TODO and draft markers left in publishable text

Highlights

• zero-key local CLI
• Markdown-first workflow
• text, JSON, Markdown, and SARIF output
• .antirot.toml config support
• GitHub Actions CI with SARIF upload path

Quick start

python3.10 -m venv .venv
source .venv/bin/activate
pip install -e '.[dev]'

antirot lint examples/sloppy_paper.md \
  --references examples/references.md \
  --format markdown

Why now

Research agents can generate polished text faster than authors can verify it. AntiRot is a narrow guardrail for the last mile: the artifact that actually gets shipped.

Known limits

• heuristics are intentionally simple in v0.1.0
• claim detection is conservative and markdown-focused
• config parsing currently supports scalar values only