recordingtest/docs/contracts/smoke2-gap-fix.evaluation.md

Evaluation — smoke2 gap fix (issue #12)

• Commit graded: 8784fec
• Evaluator: independent session (Opus 4.6 [1m])
• Date: 2026-04-07
• Note: Issue #12 used a free-form issue body instead of a Sprint Contract (docs/contracts/smoke2-gap-fix.md does not exist). Acceptable per CLAUDE.md §0.1 for follow-up bug fixes, but recorded here.

Verdict: PASS

Verdict table

Item	Required	Observed	Status
Build	0 warn / 0 err (TWAE)	Clean, 0/0	pass
Tests total	71 pass / 0 fail / 0 skip	71 pass / 0 fail / 0 skip (16+10+17+5+5+6+6+6)	pass
Gap A — full path resolver	UiaPathParser splits /, parses (Class, AutomationId?, Name?); IUiaTreeNode adapter; descend chain id→name→class; bounded fallback; UiaPlayerHost wired; null on miss	All present. UiaPathParser.cs quote-aware split, attribute parser. IUiaPathResolver.cs defines IUiaTreeNode + UiaPathResolver with MatchRoot + FindChild + DescendantsBounded(maxDepth:4) documented fallback. Matches priority AutomationId > Name > ClassName. UiaPlayerHost.ResolveElement uses UiaPathResolver.Resolve(new FlaUiTreeNode(window), uiaPath) via Retry.WhileNull, returns null on miss (engine handles throw). Old ExtractAutomationId shortcut removed.	pass
Gap B — type target inheritance	_lastFocusPath / _lastMousePath state; FlushType fallback chain (typeRes → focus → mouse)	DragCollapser adds lastFocusPath + lastMousePath locals (line 46-47); FlushType fallback typeRes ?? lastFocusPath ?? lastMousePath (line 72); focus_change updates lastFocusPath (line 332); mouse_down_l/r updates lastMousePath from downRes (line 128, 188).	pass
Gap C — window filter	ShouldKeep; mouse uses WindowFromPoint; key uses GetForegroundWindow; wired to SUT pid in Program	WindowFilter.cs: IWindowFilter, PassThroughWindowFilter, SutProcessWindowFilter with two pluggable lookups. Mouse path → processFromPoint, key path → processFromForeground, focus_change always kept, pid==0 permissive. Program.cs lines 87-107 wires SutProcessWindowFilter to app.ProcessId using WindowFromPoint+GetWindowThreadProcessId and GetForegroundWindow. LowLevelHook exposes mutable Filter and applies it in both Keyboard/Mouse procs.	pass
Gap D — UTF-8 BOM-less	Explicit UTF8Encoding(false); round-trip test with Korean strings + no BOM	ScenarioWriter.WriteToFile line 43-44: new UTF8Encoding(encoderShouldEmitUTF8Identifier: false). Round-trip test ScenarioWriter_RoundTrip_PreservesKorean asserts byte[0..2] != EF BB BF and Korean Name/Description/Path round-trip.	pass
Thread.Sleep in PlayerEngine	0	0 (grep)	pass
EG-BIM Modeler writes	none	none	pass

Regression-trap analysis

New test	Would have failed pre-fix?	Why
UiaPathParser_ParsesMultiSegment_WithClassAndId	yes — UiaPathParser did not exist	Compile-trap
UiaPathParser_ParsesNameAttribute	yes	Compile-trap
UiaPathResolver_Descend_FindsNestedElement	yes	Compile-trap; also exercises chain descent
UiaPathResolver_LastSegmentWithoutId_UsesClassName	yes	Validates ClassName fallback in Matches
UiaPathResolver_NotFound_ReturnsNull	yes	Validates null-not-throw contract
SmokeRegression_BoxV4CleanLike_ParsesAndResolves	yes — explicitly proves the resolver no longer collapses to "first descendant" (Assert.NotSame)	Direct guard against the smoke 1차 bug
DragCollapser_TypeAfterFocusChange_InheritsTarget	yes — pre-fix FlushType only used typeRes; with all-null resolver result, Target would be null. Test asserts Target.UiaPath == focusPath.	Direct guard for Gap B
DragCollapser_TypeAfterMouseDown_FallbackToMouseTarget	yes — same reason; asserts mouse path inheritance	Direct guard
WindowFilter_ExternalCoord_DropsEvent	yes — SutProcessWindowFilter did not exist	Compile-trap; also asserts drop semantics
WindowFilter_SutCoord_KeepsEvent	yes	Compile-trap; asserts keep + permissive pid=0
ScenarioWriter_RoundTrip_PreservesKorean	yes — pre-fix relied on default overload; the explicit byte-level EF BB BF assertion + Korean round-trip would only deterministically pass with the explicit encoder	Direct guard for Gap D

All new tests are load-bearing.

Notes

• docs/guides/smoke-test.md Gap D tip section confirmed (commit diff shows +18 lines).
• No partial hacks observed; all four gaps are real code-level fixes with unit coverage and reasonable abstractions (pluggable lookups for the filter, pure adapter for the path resolver).
• Recommend Generator update PROGRESS.md to mark issue #12 done.