neveshtar 0.2.0: naturalness mode

neveshtar 0.2.0: naturalness mode

New: neveshtar natural FILE finds stiff, template-like Persian prose and, for each finding, gives the location, why it reads that way, a rewrite or a question for the author, and a confidence tier. safe mode applies only unambiguous trims.

• 11 checks (R101–R601): stock openers and phrases, empty intensifiers, stacked connectors, uniform rhythm, repeated frames, three-item lists, vague generalities, English calques, nominalisation, passive pile-ups
• neveshtar rhythm FILE: plain numbers (sentence length, variation, opener diversity, connector / three-item / passive share). No score, no verdict
• Output: text, --json, --html diff, and a .docx copy with tracked changes plus Word comments for the open questions
• 55 new tests (99 total); still standard library only, offline

Honesty: a writing-quality aid whose rewrites you review. It makes no claim about AI detectors, is not built to evade them or hide authorship, and you stay responsible for your text and your university's AI-assistance rules. Tracked changes and comments have not been opened in Microsoft Word (checked by reading the XML back). Linux only tested.

Install: pipx install git+https://github.com/mrzroot/neveshtar · Live page: https://mrzroot.github.io/neveshtar/ · Changelog: CHANGELOG.md

neveshtar 0.1.0

First release. Offline, rule-based, standard library only.

• neveshtar check finds spoken forms (about 1,300 rules in three confidence tiers) and thesis-style issues; each finding has a reason.
• neveshtar fix writes a copy with Word tracked changes (safe mode by default; quotes, code and URLs are protected).
• neveshtar scaffold prints fill-in-the-blank frames per thesis section; --json, --strict, self-contained HTML report.

Limits: not an AI author, not designed to disguise machine-written text or evade AI detectors, no guarantee of academic quality; never opened in real Microsoft Word; Linux only; no accuracy benchmark. See the README.