quality-loop

One standard and one loop for "is this product good enough for a paying consumer, and does it stay that way?"

• Bar: every in-scope dimension scores >= 4 of 5, average >= 4.5 (quality >= 90%), all floors pass. A project can raise it; floors never relax.
• Quality %: the average score / 5 x 100, rounded down to a whole number. Every report opens with it and the reason for each lost point.
• Floor: a hard minimum (crash, data loss, broken core journey, accessibility minimum, deceptive design...). Any floor breach caps its dimension at 3 and blocks release.
• Receipt: evidence bound to one build (commit or build number, date, device, language). A receipt from another build is stale.
• Ledger: the project's durable quality record. Read it first, write it last.
• Ratchet: every fixed defect leaves a guard (test, lint rule, CI check, checklist line) so quality only moves up.
• Judge: a fresh-context agent that sees only receipts, the scorecard, and the bar. A self-score never closes a gate.

Green tests prove the code does what the tests say. They do not prove the product is good. Score the product a person actually gets.

Files