Skip to content

Latest commit

 

History

History
44 lines (37 loc) · 3.42 KB

release-audit.md

File metadata and controls

44 lines (37 loc) · 3.42 KB

Release correctness audit — alpha.6

This candidate addresses two false-positive paths found during review of alpha.5 and adds an explicit keyboard-input mode. The priority is what a passing report actually proves. These are synthetic regressions, not reports about a third-party application.

Reproduction Alpha.6 result Evidence
Focus moves from the declared opener to a different button that can open the same dialog D01 fails with FOCUS_MISMATCH; observed focus is unknown; no Enter/Space is sent Both activation keys
Add 150 ms of transport delay before or after browser evaluation during a short manual-tabs window T05 is blocked with OBSERVATION_GAP; two real observations and their interval are retained; no activation follows Slow observations and a healthy short window
Cancel character keydown events on an otherwise working form type fails F06 with VALUE_MISMATCH; insertText still passes its narrower input contract Input modes
Move focus during the first typed character F06 fails with FOCUS_MISMATCH; no second character is typed Typing interruption
Request Unicode/control characters in type mode CONFIG_INVALID before submission; no silent fallback or private value in the report Input validation
Claim a passing window with missing evidence or a gap above 100 ms The CI guard rejects the report and the process exits 1 Guard subprocess cases

The 100 ms boundary has a deterministic unit test: exactly 100 ms is allowed; 100.125 ms blocks the window, even after later fast samples. Window duration is measured between actual browser snapshots, not from time spent awaiting the first snapshot. At least two observations are required. Values are checked against the JSON schema and cross-field relationships before attachment.

The dialog preflight is still a snapshot followed by a separate Playwright command. It rejects the observed wrong target; it does not make those operations atomic. Manual-tabs checks sample a declared window after focus and roving checks. A transient change entirely between reads can remain undetected. Neither input mode claims to emulate an IME or physical keyboard layout.

Ruleset 0.4.0 identifies these behavioral changes. The alpha schema-v1 extension adds sampling metrics, the gap code, form mode and value-free type steps. Use a matching package version for the assertions, schema and CI guard. The default form mode remains insertText for existing contracts; new keyboard-oriented examples select type explicitly.

The historical regression record extracts ten browser cases from the earlier 100-round series and binds them to its source hashes and raw-report checksum. It does not certify later runtime changes; current development results are listed in Verification. The injected-delay cases measured gaps above 150 ms and both stopped as blocked. These are diagnostic samples, not performance claims.

Verification and source-bound evidence are listed in Verification. The benchmark method uses an independent native implementation with the same negative-window requirements. Its form timings cover insertText.