@@ -202,3 +202,29 @@ are whole commits, including ancillary changes, rather than only their headline
202202
203203These comparisons validate source-policy behavior, not rendered pixels or recall over all
204204historical PRs. Screenshot capture, AI interpretation and Slack delivery are separate stages.
205+
206+ ### Incremental smoke test on PR #7742
207+
208+ Five temporary commits were created directly on ` 458a515cbed7fbcf127ce72348ff755c5308ce13 ` ,
209+ each changing one existing source file. Comparing that commit with each temporary head
210+ isolated the test edit from the implementation PR's own dependency changes. No temporary
211+ UI edits were checked out, pushed or included in the PR.
212+
213+ | Incremental edit | Observed result after fixes |
214+ | --- | --- |
215+ | EMCN Button ` rounded-[5px] ` to ` rounded-none ` | Flagged; a ` shape-effects ` finding identifies ` buttonVariants ` |
216+ | Send button token ` bg-[#383838] ` to ` bg-[#E11D48] ` | Flagged; ` colour ` review evidence reaches the unchanged ` SendButton ` consumer |
217+ | Send button token ` p-0 ` to ` p-2 ` | Flagged; ` dimensions ` review evidence reaches the unchanged consumer |
218+ | TSDoc wording only in EMCN Button | Clean; zero findings |
219+ | Add ` translate-x-2 ` to the send button token | Flagged for review; movement is not proven harmless in this runtime context |
220+
221+ The experiment exposed and fixed dropped semicolons between CSS custom-variant statements
222+ and missing recognition of the repository's ` cn ` import from ` @sim/emcn ` . Regression tests
223+ cover both. All five overall flagging decisions matched expectations. One stricter category
224+ assertion did not: generated translation declarations currently receive the broader ` layout `
225+ category, rather than ` movement ` ; the conservative review decision is retained.
226+
227+ Report noise remains substantial: the shared shape edit generated 2 static flags and 2,205
228+ review findings; each local token edit generated 93 review findings. These counts describe
229+ potential effects, not independently verified visual regressions. Conditional consumers
230+ still require review. This validates local engine behavior, not cloud workflow activation.
0 commit comments