Skip to content

iOS: UX polish (graph-modeled) + production readiness#522

Open
1jehuang wants to merge 9 commits into
masterfrom
ios/ux-production
Open

iOS: UX polish (graph-modeled) + production readiness#522
1jehuang wants to merge 9 commits into
masterfrom
ios/ux-production

Conversation

@1jehuang

@1jehuang 1jehuang commented Jul 20, 2026

Copy link
Copy Markdown
Owner

UX improvements gated by the TestHarness reward model (interaction ActionGraph + 16 scorers over a 15-cell device/type-size/scenario matrix): top-anchored transcript, compact single-line header, starter prompt chips (tap_suggestion graph edge), calmer status pill, collapsible reasoning, grid-aligned padding. Reward 89.6 vs 88.7 baseline, worst cell 84.0 vs 83.2, 71 tests pass.

Production: PrivacyInfo.xcprivacy, foreground reconnect, PRODUCTION_CHECKLIST.md + check_production.sh (13/13), CI signing reworked (archive unsigned, sign at export).

Remaining account-holder-only step: fix ASC cloud signing permissions per ios/PRODUCTION_CHECKLIST.md.

Related production-readiness context: #464

1jehuang added 9 commits July 19, 2026 16:29
…actions) + calmer status, collapsible reasoning, tool summaries
- PrivacyInfo.xcprivacy privacy manifest (UserDefaults CA92.1, no tracking)
- Foreground scenePhase reconnect so returning to the app is instant
- PRODUCTION_CHECKLIST.md with executable pass conditions
- TestHarness/check_production.sh local gate (13 checks, all passing)
- Grid-aligned tool card padding, full-width code blocks
Transcript now reads top-down like every major chat/agent app
(iMessage short threads, Slack, ChatGPT). Removes the flexible
bottom-anchor spacer, keeps autoscroll-on-overflow. Also snaps the
last off-grid padding literal (2pt -> 4pt) to the 4/8pt grid.

Reward gate: 88.8 vs 88.7 baseline; worst-category floor moved from
space_efficiency/rhythm to perf (launch-time noise).
One-tap suggestions fill the composer, cutting first-interaction cost
from M+12K+TAP to M+TAP. Modeled as the tap_suggestion edge in the
interaction graph (imputed prior 0.02).

Reward gate: 89.2 vs 88.7 baseline (+0.5), worst cell 83.7 vs 83.2.
Title and model share one baseline (model shown without the auth-route
prefix), halving header chrome height. Transcript gains the space.

Reward gate: 89.6 vs 88.7 baseline (+0.9), worst cell 84.0 vs 83.2.
Cloud-managed signing defaulted to a development profile, which fails
on a team with no registered devices. Forcing the distribution identity
makes xcodebuild request an App Store profile instead.
Archive-time automatic signing wants a development profile, impossible
on a device-less team. Export re-signs with cloud-managed Apple
Distribution using the ASC API key.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant