[smishra3]

What changed

• README opening rewritten and a new What to expect section: what the model was trained on, where it generalised in the paper, where it fails (texture-like signals such as EZH2, semantic not instance), and when fine-tuning is worth it (nuclear envelope and lysosome numbers). All facts are from the manuscript PDF.
• New Maintenance section with the checklist's "active public code" statement, who maintains and reviews, response time, release rule, planned work, and the wind-down after publication.
• New How this code was developed section: cyto-dl origin of the model, the package as a standalone re-implementation, AI assistance disclosed, Assisted-by: trailer policy. One new RELEASE TODO: verify masks against the cyto-dl predictions.
• CONTRIBUTING: short AI-assisted contributions section following the Institute's coding-agent guidelines. PR template: matching checklist item.
• Existing RELEASE TODO comments are left in place.

Why

Items 1 (maintenance plan), 4 (AI usage disclaimer) and part of 5 (docs edit pass) from the public-release review. The new prose is written from the paper rather than generic, which is also the fix for the "reads AI generated" comment; please edit it into your own words where it does not sound like you.

How it was tested

Docs only. Markdown previewed locally; no code changes.

Checklist

• Tests pass locally (pytest) (unchanged code)
• Docs updated if behaviour or commands changed
• CHANGELOG.md has an entry under Unreleased (not user-visible behaviour)
• No weights, outputs or notebook outputs committed

🤖 Generated with Claude Code