Changelog

What we shipped this week.

Model updates, new style guides, new features. We publish every release, every regression, every bug we hit. Subscribe at the bottom to get the digest weekly.

All Model Style guides Features Fixes
v3.2.1

Detector pass rate up to 95.8% (from 94.7%)

Adapter retrained on the latest 200-essay rotation. Largest gains in the personal-statement category (96.2%, up from 94.4%). Existing users get the update automatically, re-polish any document to use the new model.

v3.2

Voice fingerprint, per-user voice preservation

Polishes now constrain to your existing documents’ sentence-length distribution, lexical diversity, and 4-gram profile. Median ‘voice match’ score across users: 93% (was 78% before).

style

Three new journal style guides: PLOS, AGU, Lancet

Added on user request. We add 1–2 journal guides per quarter. Up next: ICLR / NeurIPS workshop style (mid-June).

v3.1

Per-section audit log for research papers

When you polish a research paper, you now get a PDF audit log per section, what edits were applied, what strategies they implemented, and which sentences we refused to touch.

fix

Citation extraction handles Vancouver-style superscripts

We were occasionally treating numbered superscript citations (e.g. finding12) as regular text. Fixed. Affects ~3% of medical-journal manuscripts.

v3.0

v3 launch, reviewer rubric, detector preview, voice fingerprint

The biggest release in a year. New polish methodology, new in-app workflow, new pricing. Full notes →

feature

Detection preview unlimited for paid users

Removed the 5-checks-per-month cap. Detector preview is now unlimited on the Researcher plan and above.

v2.7.4

Methods-section guard rails strengthened

Active-voice conversion is now hard-disabled in Methods sections by default. Previously it was a soft default that could be over-eager. Existing users won’t see this change unless they explicitly toggle it on.

Load older entries ↓

Subscribe to the changelog digest.

Weekly summary of model updates, new style guides, and shipped features. Plus the proof-test results from each Monday run.