visual 기준 이미지는 2026-08-18(cf45bcc)이 마지막이다. 그 뒤로 화면을
바꾼 커밋이 130개 들어왔고, 기준은 한 번도 갱신되지 않았다. 그래서
test:visual 133건 가운데 2건(평문 NOT_FOUND)만 통과했다.
달라진 이유는 아래와 같다(이미지별 표는 PR 본문).
- 공개 화면: 9월 개편(b9d0585 바탕화면 디자인 수정본, 0b0c5cb 홈의 길,
fe1f235·d3bf392 기록 머리와 읽기 품질, e88ed5e·5399e19 조사 상태,
aca4f18·de4cb8b 주제 목록과 탭 등), 질문 라벨 대비(cd4fa8c).
- Studio 화면: 로그인한 Studio를 찍는다(500dbc6). Studio 재설계와 v4 블록
편집기(9545860, d72eae2 등), MOCK 관리 데이터(이 브랜치의 관리 mock:
대시보드 「지금 집중하는 것」).
- 모르는 slug 20장: SPA 안의 not-found(6784eb1, 이 브랜치의 visual 수정).
- 편집기 360px: 블록 손잡이 여백(이 브랜치의 935px 이하 수정).
검토한 방법:
- visual이 찍는 56개 화면을 같은 조건(고정 시계, Studio 로그인)으로 열어
오류 화면, aria-busy, 로그인 관문, 콘솔 오류를 셌다. 의도한 not-found
화면 말고는 없었다.
- 옛 기준과 새 기준을 나란히 놓고 범주마다 봤다. 의도와 다른 차이가 하나
있었다. 공개 화면 h1에 파란 초점 테두리가 생긴 것이다(이 브랜치의 도착
초점 처리가 h1에 data-route-heading을 달지 않았다). 그것은 기준을
바꾸지 않고 코드에서 고쳤다(f78ce58에 합침).
- desktop 호스트(chromium, CI env)에서 test:visual:update. 그 뒤
test:visual 두 번 연속 132건 통과(즉시 미리보기 1건은 지웠다).
한계: 이 호스트에서 만든 기준이다. 8월 기준도 같은 호스트에서 만든 것이고
평문 NOT_FOUND 2장이 이번에도 바이트까지 같다. CI 컨테이너
(playwright --with-deps)에서 픽셀까지 같은지는 확인하지 못했다.
「Studio immediate preview」 visual은 편집기의 「즉시 미리보기」 영역을
찍는다. d72eae2(09-16)가 편집기의 편집/미리보기 모드를 없애며 이 영역을
뺐고, 단위 시험은 그때부터 「없음」을 확인한다. visual만 남아 영역을 찾지
못하고 떨어진다.
시험과 기준 이미지를 지운다. 공개 렌더는 Public Preview 화면의 visual
(current-preview 등)이 계속 본다.
`/studio/assets` was the one route in the TechLog route contract missing
from `TECH_LOG_CANONICAL_ROUTES` and `TECH_LOG_STUDIO_STATE_PATHS`, so it
had no golden, no responsive-overflow check, and no Axe scan. That gap was
not hypothetical: the responsive overflow just fixed in the Asset search row
shipped in two places and was only caught in the Picker, which has a golden;
the Library's identical copy would have gone unseen.
Verifying the render before capturing anything turned up a second defect.
`studioSpaPathPatterns` in the production serving contract never listed
`/studio/assets`, so a hard navigation or reload of that URL was answered
with the in-shell Studio 404 -- the screen was reachable only by client-side
navigation from another Studio page. A golden taken then would have frozen a
404. With the pattern added, the route serves 200 text/html like its
siblings, and the page renders correctly at both breakpoints:
scrollWidth == clientWidth == viewport at 360 and 1440, `main` and the `h1`
visible, the search label/input/button all inside the 360px viewport, and no
critical or serious Axe violations in chromium.
Three new goldens, no existing golden regenerated:
tech-log-studio-assets-{360,1440} from the canonical route list and
tech-log-studio-asset-library-1440 from the Studio state list, matching how
every other Studio route appears in both. `pnpm test:visual` is 133 passed,
up from 130.
The demo profile seeds no assets, so the golden captures the empty state --
which still covers the header, heading, search row and status region, the
surface the overflow regression lived on.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Stale because of the search UI this task added to the CASE editor's asset
panel, so they belong to this work. Each diff was inspected before
regenerating; every difference is attributable to the added control.
Four render `.../edit` for a CASE (2999 -> 3066px tall, +67px):
document-edit-1440, case-editor-1440, conflict-editor-1440,
dirty-leave-dialog-1440 -- the comparator flags one 401x95 region at
(131,2795), the "Asset 검색" label, input and 검색 button. Above it the
maximum per-channel delta is 1/255, i.e. visually unchanged.
conflict-editor was not in my first estimate: document ...115 is a CASE
too, so it renders the same panel.
document-edit-360 (3627 -> 3694px, +67px) shows more because the layout
is single-column: byte-identical above y=3032, then the added control,
then the status rail below translated down (67px, 68px past the 저장
button -- sub-pixel rounding, same content).
Its width returns to 360 here; the previous run captured it at 382,
which was the overflow fixed in d2574a3, not a golden to bake in.
`corepack pnpm test:visual`: 130 passed. `git diff --stat` for this
commit lists exactly these five PNGs.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The prior commit (ef17bca) fixed the 7 of 13 test:visual failures that
were a pure dimensional consequence of 79e9aa8's evidence-figure width
rule. The other 6 were left failing pending investigation of their own
diffs rather than being folded into that theory.
Per-file inspection of each diff/expected/actual triple confirms all 6
are stale content from 0355b64 ("feat: add TechLog project decision
authoring"), also an ancestor of main's merge-base -- a "Decision"
document type was added without refreshing the affected goldens:
tech-log-studio-documents-360 360x2883 (unchanged) -- subtitle
copy gained ", Decision" before "을 찾고", rewrapping the sentence.
tech-log-studio-documents-1440 1440x1493 (unchanged) -- same
subtitle copy change, no wrap at this width.
tech-log-studio-document-list-1440 1440x1493 (unchanged) -- same
page/route as documents-1440 (identical diff), same copy change.
tech-log-studio-document-new-360 360x1000 -> 360x1130 -- a new
"Decision" radio option/row added to the document-type picker;
narrow viewport can't absorb the extra row without growing.
tech-log-studio-document-new-1440 1440x1000 (unchanged) -- same
new "Decision" radio row; existing bottom whitespace absorbs it at
this width.
tech-log-studio-new-document-1440 1440x1000 (unchanged) -- same
page/route as document-new-1440 (identical diff), same new row.
None of the 6 diffs show anything resembling a rendering defect (no
overlap, no clipped/broken layout) -- each is legitimate new copy or a
legitimate new control from the shipped feature.
test:visual: 124 passed/6 failed -> 130 passed/0 failed. lint and
check:types unaffected.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
79e9aa8 ("fix: align TechLog article content widths") narrowed
.evidence-figure/.code-block/.data-table-wrap from
min(61rem, calc(100% + 15rem)) to min(var(--body-copy), 100%), shrinking
every page whose article body contains an evidence-figure diagram. The
goldens were never refreshed after that commit landed, so test:visual
carried 13 failures inherited from main.
Per-failure diff inspection (not just the aggregate count) showed the 13
split into two unrelated causes:
- 7 are a pure dimensional consequence of the width rule -- full-page
diffs starting at the evidence-figure diagram, page height down by a
fixed ~93px offset in matched pairs. Regenerated here:
tech-log-case-1440 6506 -> 6412
tech-log-public-fixture-cases-collection-... 6506 -> 6412
tech-log-studio-document-preview-1440 1904 -> 1811
tech-log-studio-current-preview-1440 1904 -> 1811
tech-log-studio-publication-preview-1440 1687 -> 1594
tech-log-studio-publication-snapshot-1440 1687 -> 1594
tech-log-studio-immediate-preview-1440 1733 -> 1640
- 6 are unrelated content staleness from a later commit
(0355b64, "feat: add TechLog project decision authoring") that added a
"Decision" document type/option without refreshing its goldens. These
are left failing intentionally -- they are not a dimensional
consequence of 79e9aa8 and regenerating them here would silently bake
an unreviewed content change into the baseline:
tech-log-studio-documents-360/1440,
tech-log-studio-document-new-360/1440,
tech-log-studio-document-list-1440, tech-log-studio-new-document-1440
test:visual: 13 failed/117 passed -> 124 passed/6 failed (the 6 above).
lint and check:types unaffected.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Attribution was measured, not inferred: the visual suite was run at the
merge-base `9e5fbd1` in a detached checkout (13 failed / 117 passed) and the
`-actual.png` each run produced was compared by SHA-256 against HEAD's. For
13 of the 18 failures the rendered bytes at HEAD and at the merge-base are
identical, so those failures belong to `main`, not here.
Regenerated (all Case-editor surfaces, all grown by the Task 10 Asset panel;
the +293px band was inspected in the new golden and is the "EVIDENCE / 본문에
Asset 삽입" panel):
tech-log-studio-document-edit-360 360x3313 -> 360x3627
tech-log-studio-document-edit-1440 1440x2706 -> 1440x2999
tech-log-studio-case-editor-1440 1440x2706 -> 1440x2999
tech-log-studio-conflict-editor-1440 1440x2706 -> 1440x2999
tech-log-studio-dirty-leave-dialog-1440 1440x2706 -> 1440x2999
`--update-snapshots` was restricted to those five tests with `--grep`; a
blanket update would have absorbed the 13 inherited failures and destroyed
the distinction. `test:visual` now reports 13 failed / 117 passed, exactly
the merge-base's set.
Two earlier records are corrected in the spec: the five "pixel-only, not
investigated" failures are all inherited, and `TECH_LOG_STUDIO_DOCUMENT_NEW`
360px was wrongly attributed to the Asset Picker -- it fails at the
merge-base with the identical 28,413 differing pixels.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>