Reading evidence
Historical proposal/evidence: Atlas is deferred and is not included in the current Skill or CLI. Atlas paths and commands below describe the earlier implementation; they are not current usage instructions. Sequence readability improvements remain enabled.
Comparison base: 10722002bb8777ecb639d93c49586fae4adf3ae4.
The fixed input is archify/examples/cache-miss-request.sequence.json.
sequence-before.png and sequence-after.png compare its base and candidate
artifacts at 1440×900, Classic/light, READ/Still, in the same Chrome environment.
Message type increases from 9 to 11 source pixels with matching plates; the
participants, routes, labels and topology are retained. Standard inputs retain
legacy message metrics, covered by a parallel-label regression.
layered-light.png and layered-dark.png show the separate-detail reader at
2048×1320 after an offline browser run. The browser fixture uses three copies
of the checked sequence to exercise single-target and multi-target navigation
without introducing unverified domain relationships. The runnable product
sample archify/references/request-reading.atlas.json instead links the web
application's API node to its cache-miss sequence.
Reproduce interaction evidence:
ARCHIFY_CHROME="/path/to/chrome" ARCHIFY_READING_EVIDENCE="/tmp/reading-evidence" \
node --test archify/test/atlas-browser.test.mjs
That test blocks HTTP/HTTPS and exercises Enter activation, explicit detail choice, return camera/focus, shell/child theme synchronization, four desktop sizes, icon painting and canonical SVG export. Its second case deliberately shrinks message text to prove that browser readability detects the defect.
Run archify/test/desktop-reader-browser.test.mjs with the same Chrome setting
for all packaged diagrams and the production architecture readability case.
Perceptual review: screenshots were inspected with an image-capable reviewer. The enlarged messages remain distinct from auxiliary labels, no new message collision was observed in the comparison, and both atlas themes show visible controls and a contained diagram. This is bounded evidence for the fixtures, not a guarantee for arbitrary supplied HTML, all exports or every browser.
Independent review disposition: the shared output-path guard replaces bespoke alias handling; CLI tests cover non-HTML and symlink-resolved non-HTML targets. The standard-v1 parallel-message compatibility finding is fixed by retaining legacy sizing outside showcase. Atlas chrome uses the project's mono stack.