feat: highlight search matches and document the working UI
Core CI / core (ubuntu-latest) (push) Failing after 3m29s
Core CI / core (macos-latest) (push) Canceled after 0s
Core CI / core (windows-latest) (push) Canceled after 0s
Core CI / desktop (ubuntu-latest) (push) Successful in 7m23s
Core CI / desktop (macos-latest) (push) Canceled after 0s
Core CI / desktop (windows-latest) (push) Canceled after 0s

This commit is contained in:
karl committed 2026-10-12 19:57:34 +02:00
1 parent b8400c4fbf
commit 79553ce921
26 files changed
+4124 -50

No files matched your search

+36 -4
View File
@@ -1,16 +1,48 @@
# Design QA
final result: blocked
final result: browser captures checked; native desktop visual gate remains open
## Browser-rendered implementation evidence (2026-10-12)
Four 1440×1000 captures now show the production frontend against the actual
shared Rust/Google CPU engine: document search in both themes, an indexed photo
preview, and measured performance details. The selected PDF result really maps
to page 2 and its notice clause. Photo pixels are loaded from the indexed file;
the trace includes actual worker inference counters. Capture assertions passed,
and every final PNG was opened and inspected.
The brand mark, local typography, two-pane structure, selection, passage text,
photo preview and diagnostics render visibly. No primary selected content is
clipped. Windows namespace prefixes are hidden only in display labels; file
operations retain the original indexed paths. The ordinary summary uses clear
Search/Shown timing labels. The last partially visible result is inside the
scrollable result list.
The final captures also check literal query-word marks in filenames, snippets
and the extracted passage. A small **Why this result?** disclosure stays
collapsed by default in the selected preview. Its page-2 explanation was opened
and checked during capture, then closed for the ordinary screenshots. Semantic
passage emphasis identifies the retrieved text; it does not claim model-derived
word importance. Photo preview text contains no invented literal matches.
These captures omit native window chrome and do not verify native WebView
behavior, OS integration, animation, playback or native keyboard/focus behavior.
The runner uses reduced motion and an owned CPU-rendered headless browser.
Its responses come from the live engine without result replay or UI mocks.
See [the capture workflow](tools/desktop-capture/README.md) and
[capture manifest](docs/screenshots/capture-manifest.json).
## Native desktop gate
Target: Karl's selected first search/preview concept, combined with Raycast-style
compact interaction and the supplied searchi brand board. Source images were
opened and inspected. The frontend implementation builds and its state tests pass.
Rendered comparison is unavailable: the in-app browser bootstrap fails because
Native rendered comparison is unavailable: the in-app browser bootstrap fails because
its browser-service module cannot be found. Prior native Windows capture attempts
returned GetCursorPos access denied and monitor capture error 0x80070057.
No matching-state implementation screenshot has been captured. No side-by-side
visual comparison or visual pass is claimed.
No matching-state native implementation screenshot has been captured. A native
side-by-side visual comparison and visual pass remain unverified.
The native development app launched in forced-CPU mode. Its accessibility tree
could be read without screenshots and exposed a missing font-import error, which