2.0 KiB
Part Document View Services
2026-09-16. Both Part:part.documentViewCommand and Part:part.viewCommand
now have executable business coverage, each with 15 passing scenarios. Scope
is the view service protocol and owner/document behavior, not every desktop
command reachable through that service.
Evidence: business-evidence/c2be76cf68b327965002a269-20260916T083246908283.json.
Cases exercise actual native geometry, draw styles, hide/show, face alignment,
invalid owner/document/style/reference recovery, cancellation before execution,
FCStd restoration, same-named document isolation across two runtimes, mounted
service controls and real Three.js rendering. Screenshots and actual worker
downloads are recorded. Each owner loads once in Node and downloads its WASM
once in the browser; the same instances serve all operations.
The new negative browser cases revealed a real document-loss defect: malformed
requests returned the sentinel request ID invalid, causing ModuleRouter to
destroy the session. The two Part workers now attach a supplied request ID to
request validation failures. Only genuinely uncorrelatable errors retain the
sentinel. A bad request no longer discards healthy documents or reloads WASM.
Final selected execution took 12.483 seconds. Repeating both IDs with --reuse
took 0.153 seconds, read one proof and executed zero cases. No native rebuild
was required. TypeScript, geometry ownership checks, production build, four
router recovery tests and the continuous-planner tests passed.
The continuous planner now reserves uncovered development after two acceptance batches without a first business pass. Explicit user priorities and observed runtime failures stay ahead of that opportunity. Provider waits do not spend the allowance; a development turn or a genuinely new pass resets it. This prevents repeated evidence refreshes from indefinitely delaying missing cases.
Continue remaining requirements; keep all 1508 in scope. Do not infer acceptance of untested desktop actions from a passing service envelope or rendered model.