Infrastructure: Qualify desktop lifecycle failure paths and maintain a support matrix #4451

Current status - 2026-11-09

Blocked for the qualified support baseline by F15 #4441. Evidence collection can continue; an open baseline gate does not prohibit parallel endpoint work.

The current ledger records Windows Fabric 22 Pass/1 Partial/1 Fail and Win32 Paper 22 Pass/2 Partial (2026-11-08). A later connected session passed Windows physical input (#4441). Historical macOS partial/not-run results stay at their exact source revisions. No all-pass or merged-head native qualification claim is justified.

Remaining consolidation includes:

• The per-endpoint failures and gaps enumerated in Infrastructure: Live-qualify the desktop driver and MCP foundation on macOS, Windows, and Win32 #4441, which now also carries the remaining F1-F5 and F16 native evidence
• Native ownership and containment from Infrastructure: Implement neutral macOS desktop application ownership and identity recovery #4446/Infrastructure: Implement neutral Windows desktop application ownership and containment #4447
• Cross-service exact-application exclusion and input-mutex qualification (from F2 Infrastructure: Implement settled desktop lifecycle supervision and fenced ownership records #4442)
• A real non-React integrated application and AppKit fixture
• Independent supervisor, helper, app, Metro, bridge, runner and MCP faults
• A repeated-run reliability window

Coordinate release/CI windows with #4345/#4349 rather than claiming duplicate independent evidence. Managed macOS input lanes from F16 are tracked in #4349.

• Serialized --input all (moved from F16 Infrastructure: Enforce session input policy and attribute delivered input mechanisms #4467): run the input lanes one after another and account for each lane separately. An interrupted run reports the remaining lane as Not run. Never run physical lanes in parallel on one interactive desktop.

Discovery/operation bytes and tokens, tool counts, cold/warm startup and latency, retries/recovery cost, retained resources and cleanup need exact revision/client/tokenizer/sample-count/artifact methods. Windows Fabric and Win32 Paper MCP measurements are now recorded in the MCP ledger. macOS Paper, click and cancellation costs remain Not run. Historical macOS numbers and green hosted jobs cannot fill those cells. Preserve every failed attempt and every explicit Unsupported/Skipped outcome.

Goal

Maintain a revision- and endpoint-specific support matrix for the desktop automation packages, covering injected lifecycle failures and agent-efficiency measurements. Each feature item produces its own native and consumer evidence when it lands; this item consolidates that evidence and fills the gaps that only show up across components.

Part of #4435 (plan item F14).

Scope

• Owner: cross-platform qualification; each feature owner remains responsible for its own evidence.
• Paths (under packages/agentic/): driver native fixtures and contracts, harness and Storybook consumer tests, owning app CI and scripts, and the references/current-status.md ledgers.
• Inject supervisor, helper, app, Metro, bridge, runner, and MCP failures independently. Cover same-port replacement, surviving children, stale descriptors, late renders, grant revocation, and uncertain cleanup.
• Include a non-React application, and add the AppKit fixture rather than assuming it exists.
• Report macOS Fabric, macOS Paper, Windows Fabric, and Win32 Paper separately.
• Measure identical tasks for schema overhead, tool calls and response bytes, cold and warm startup, retries, recovery cost, retained resources, and cleanup correctness, and record them in the ledgers with their method.

Dependencies

• Blocked by: F15 Infrastructure: Live-qualify the desktop driver and MCP foundation on macOS, Windows, and Win32 #4441, which establishes the baseline for the delivered foundation.
• Consolidates: each advertised feature only after its implementation and endpoint evidence land.
• Related: Infrastructure: Complete Desktop Driver production release qualification #4345 (driver production release qualification) and Infrastructure: Qualify and require native Storybook CI coverage #4349 (required native CI).

Acceptance

• The matrix lists each endpoint and feature with passed, failed, skipped, not-run, or unsupported status at an exact revision.
• Hosted physical-input or capture skips are recorded as skipped, not passed; a successful semantic activate is not evidence that a physical click works.
• Measurements record revision, client, platform, renderer, tokenizer, sample count, and artifact locator.
• Historical results are never reported as qualification of a later revision.

Validation

• Iterate with focused yarn test <name> in each changed package, then run its yarn format, yarn lint, yarn build, and yarn test. Finish with yarn clean-lage-cache then yarn lage buildci; run yarn lage test-links --verbose --grouped after markdown changes.
• Add a patch changeset for each changed published package. Update the owning references/current-status.md rows (done, work in progress, unsupported) and affected README/AGENTS guidance in the same change.
• Native, provider, or process-topology changes need matching-OS native contracts and the affected consumer lane. JavaScript tests and bundles do not qualify native behavior; hosted capability skips are not passes. Report each check as passed, failed, skipped, not run, or unsupported.

Evidence

• Current ledgers: packages/agentic/desktop-driver/references/current-status.md, packages/agentic/desktop-test/references/current-status.md, and packages/agentic/desktop-driver-mcp/references/current-status.md.
• Retired plan: driver-next.md at 0569e1b86, sections 8 (F14) and 11.