Infrastructure workstream #4275

Goal

Make desktop behavior and package-cost claims for the agentic component library provable through one supported desktop-driver/Storybook workflow, truthful native coverage, reproducible package boundaries, and optional reviewer-facing visual evidence.

Status - 2026-10-01

In progress: 19 direct sub-issues; 9 completed delivery milestones and 10 open qualification/follow-up/bug items.

This reconciliation audits merged main at c3a89b0b7 and current native CI evidence. The original ten children were reviewed individually. Completed delivery is now separated from remaining original acceptance criteria: #4253/#4254/#4255/#4256/#4260 are closed as delivered milestones, with unfinished work preserved in new direct sub-issues. Existing completed #4257/#4258/#4259/#4279 remain closed. #4328 remains open.

Closure means the described milestone landed, not that its later production qualification is complete.

Current repository evidence

• Desktop Driver: public Appium-independent W3C package, source-built macOS and shared Windows/Win32 native providers, verified native artifact selection, exact app ownership, cancellation/input recovery, normalized accessibility, native capture, and shared Storybook integration are merged (Infrastructure: land the public desktop driver #4294/Infrastructure: add native desktop driver providers #4306).
• Story authoring: named, typed wdio callbacks replace component-catalog static plans; authenticated readiness, callback source verification, isolated workers/sessions, and grouped owned smoke are implemented (Add wdio test support in stories #4313/Fix desktop focus management and accessibility action handling #4314). Current discovery is 29 cases across 17 stories and ten component/primitive families on each endpoint.
• Catalogs: current generated manifests contain 168 macOS, 158 Windows Fabric, and 152 Win32 Paper stories. Fabric excludes Accordion/Callout; Paper excludes Accordion/ListItem because of native fail-fast problems, now explicitly tracked in Infrastructure: Restore Storybook stories excluded by native desktop crashes #4344.
• Native CI: all three Storybook jobs passed in October 1 run 36927750317, each with 7 authored passes / 22 skips. Windows/Win32 disable keyboard, physical click, and wheel; hosted macOS disables physical click. These results prove only executed assertions/capabilities, not full interactive qualification.
• Required-status policy: main currently requires PR; that aggregate excludes Storybook and bundle-size jobs. Green optional jobs or a green aggregate are not required desktop/package-cost gates (Infrastructure: Qualify and require native Storybook CI coverage #4349/Infrastructure: Enforce package-cost budgets and production dependency boundaries #4341).
• Package cost: seven macOS/Windows esbuild scenarios, baselines, workspace attribution, optional-module rules, and trusted persistent advisory PR reports exist. Two current-main measurements produce identical inputs for all 14 combinations (Faster and more concise bundle-size reports #4335). This measures tree-shaken package JavaScript with native/React peers externalized, not total application size. Required budgets, complete dependency enforcement, primitives/archives, and promotion policy remain Infrastructure: Enforce package-cost budgets and production dependency boundaries #4341.
• Capture/visuals: native window/element capture is implemented, but Windows/Win32 fidelity/DPI/provenance/session qualification remains Infrastructure: Qualify Windows/Win32 capture fidelity and provenance #4348. Native baseline/candidate/diff reports are not implemented (Infrastructure: Add native Storybook visual comparison reports for Windows, Win32, and macOS #4328).

portablePlanDigest hashes legacy static plans, currently an empty set in the component catalog. Matching it is not proof of WDIO callback equivalence; production reconciliation must include actual callback/source and endpoint catalog identities.

Open work items

Completed delivery milestones

Beta exit criteria

• Public Desktop Driver and shared Storybook CLI are merged and included in repository build/test integration.
• macOS, Windows Fabric, and Win32 Paper PR jobs explicitly build/verify helpers, run native stories-and-tests, and retain diagnostics.
• Authored semantic cases actually execute through each native provider without a fake fallback.
• Authored discovery extends beyond the initial Button/Checkbox/Input foundation.
• Cross-platform package-code measurements and tested optional-module boundaries are reproducible and advisory reports are published.
• Complete portable accessibility role/name/state/value and keyboard coverage on authoritative endpoints (Infrastructure: Complete portable Storybook accessibility coverage and reporting #4343/Infrastructure: Fix native accessibility-state projection gaps on macOS and Win32 #4346).
• Preserve all original delayed focus-survival and movement coverage in active native CI (Infrastructure: Restore Windows Storybook focus-regression parity #4347/Infrastructure: Restore Storybook stories excluded by native desktop crashes #4344).
• Eliminate the all-skipped success path and distinguish product, app, helper, authority, infrastructure, and coverage outcomes (Infrastructure: Reject all-skipped WDIO runs as completed coverage #4342/Infrastructure: Qualify and require native Storybook CI coverage #4349).

Production exit criteria

• All required authored cohorts run on authoritative macOS/Windows/Win32 environments with auditable callback/catalog/helper/app identity (Infrastructure: Qualify and require native Storybook CI coverage #4349).
• Retry/quarantine/expiry, retention/redaction/access, escalation, fork safety, and runner capacity have named owners (Infrastructure: Qualify and require native Storybook CI coverage #4349).
• At least 100 representative runs per endpoint over 14 days meet at least 99% infrastructure success, with zero leaked processes or unreleased input and complete failure metadata (Infrastructure: Qualify and require native Storybook CI coverage #4349/Infrastructure: Complete Desktop Driver production release qualification #4345).
• Desktop Driver passes its remaining compatibility, authority, security, cache, performance, packaging, external-consumer, and publishing gates (Infrastructure: Complete Desktop Driver production release qualification #4345).
• Package-cost/dependency gates are required, measurement coverage is complete, and increases require reviewed rationale (Infrastructure: Enforce package-cost budgets and production dependency boundaries #4341).
• Native accessibility and restored regression coverage are required without counting skips/quarantine as passes (Infrastructure: Complete portable Storybook accessibility coverage and reporting #4343/Infrastructure: Restore Windows Storybook focus-regression parity #4347).
• Windows/Win32 publish qualified capture evidence with provenance and truthful locked-session diagnostics (Infrastructure: Qualify Windows/Win32 capture fidelity and provenance #4348).
• Native visual comparison pilots/reports are delivered on all three endpoints; pixel changes remain advisory (Infrastructure: Add native Storybook visual comparison reports for Windows, Win32, and macOS #4328).

Evidence and operating boundaries

• Public driver delivery #4294, native providers #4306, WDIO #4313, focus/accessibility #4314, and bundle measurement #4335.
• PR workflow, CI authority model, remaining driver work, platform discovery, and bundle fixture.
• Windows retained native reports and Win32 retained native reports substantiate both the executed/skipped cohort and the all-skipped per-story report bug.
• Fake runs prove contracts only; render traversal proves renderability only. Hosted skips do not qualify input, TCC, DPI, or native accessibility gaps.
• Interactive Windows qualification requires a logged-in unlocked desktop; authoritative macOS qualification requires a managed Aqua session, stable signer, and explicit privacy authority.