Ship bar
A PR is not done when it first works. A PR is done when the relevant quality gap is closed well enough to justify final signoff.Functional severity definitions
critical- the primary flow is broken, unsafe, or misleading
high- the feature technically exists but cannot be trusted for normal use
medium- the core flow works but important friction, inconsistency, or missing resilience remains
low- polish-only gap that does not change trust in the result
Runtime and reliability expectations
- targeted checks must be green
- no known high-severity regressions may remain open
- retries and failure paths should remain believable when they are part of the scoped feature
UI quality expectations
- hierarchy should be understandable at a glance
- spacing and density should support fast scanning
- interaction cues should make state and actions legible
- UI work is not ready if it still obviously needs manual taste cleanup
Final signoff checklist
- the PR goal and acceptance criteria are met
- the relevant quality gap is explicitly recorded as closed
- latest QA judgment is acceptable
- latest UI judgment is acceptable when UI is in scope
- wiki writeback is current
- open risks are either resolved or consciously accepted
Discard instead of continue when
- three consecutive runs fail to improve the same axis
- the next meaningful improvement is outside the current boundary
- the PR goal no longer looks worthwhile relative to the current wedge