quality-eval-core

name	quality-eval-core
description	Shared reference for the quality-eval cluster: the four quality gates (write → commit → CI → ship), the evidence ladder that says what proves a pass at each, pass@k vs a single green run, and the AI write-and-review blind spot. USE WHEN deciding what counts as done, choosing which gate catches a failure class, or wiring a measurement loop — the conventions every quality-eval spoke shares.
cluster	quality-eval
version	1.0.0

Quality-Eval Core

Shared model for the quality-eval cluster. The TDD, E2E, verification, benchmark, linting, and audit spokes all depend on these conventions — keep them consistent here so no spoke contradicts another about what "done" means.

1. The four gates (the cluster's defining model)

Quality is enforced at four stages, cheapest-and-earliest first. Each gate catches a different failure class; a failure that escapes one is meant to be caught by the next.

Edit ──write-time──> Commit ──commit-time──> CI ──build-time──> Release ──ship-time──> Prod
   plankton-          verification-loop        e2e / windows-      production-audit
   code-quality       tdd-workflow             desktop-e2e
                      ai-regression-testing    browser-qa · benchmark

Gate	Stage	Catches	Spokes
Write	every file edit