[trissim]

Scope

Paired backend correction for OpenHCS PR159/issue135 and its independent review:
OpenHCSDev/openhcs#159 (comment)

ViewerStateManager remains the only viewer registry and atomic acquisition owner.
An application-owned nominal ViewerReuseAdmissionABC now admits a healthy
existing instance under its lock, before returning it. The generic declaration
requires only the existing health contract; OpenHCS always supplies its concrete
process-launch admission. There is no viewer-kind switch, second cache, callback
policy predicate or acquisition pre-check.

Rejection leaves the existing viewer registered, running and unchanged. It does
not silently replace it or kill a process. A factory returning an already-live
external lifecycle is registered without calling start; callers retain the
existing readiness check. Ordinary new/stale acquisitions still start normally.

Evidence and limits

• Six focused provider-free viewer-state tests pass in0.04s.
• A second thread cannot acquire the manager lock while admission executes.
• Rejected reuse invokes no factory/restart/stop and preserves the existing entry.
• External attachment is registered without start; original startup-failure,
  force-stop, readiness and subscription cases remain passing.
• Paired OpenHCS source journey:14 focused lifecycle/process-launch tests pass
  in4.84s, exercising real atomic acquisition for Napari and Fiji. Physical
  process/network operations are controlled in those tests, not live readiness.
• After normal current-main integration,50 OpenHCS lifecycle/Fiji/control and real
  data/control socket tests pass in5.34s (57 deselected); no GUI/JVM started.
• Source base1d3b32f; independent waiter PR6 edits disjoint files. No source merge,
  package installation or active blind runtime mutation has occurred.

Fresh actual viewer/MCP acceptance is pending the current frozen blind run's
released runtime slot. This draft does not claim all concurrent fresh/release
races or foreign endpoint shutdown authority are solved.

[trissim]

Parent paired integration checkpoint: normally merged main3374aa8 into existing PR7, exact2c68a114.26 viewer-state/ACK/progress source cases PASS2.61s. Installed this reviewed source with paired OpenHCS159, offline in the existing venv; real fresh MCP stream/state/screenshot on isolated:88/5794 works with actual loopback data/control listeners. Second fresh MCP successfully externally attaches, streams and captures; exposed close succeeds, owned PID/listeners independently gone.113 paired final installed checks PASS6.45s. Opened actual bitmaps; synthetic transport only, not biological QA/presentation preservation. Original packaged158-metric screening still flags one ForeignAbsenceProbe at the original nominal visualizer.is_running query; source classification is owner-reported health, not raw private absence reconstruction. No screening-pass/guard waiver/global NRA claim. Full archive remains unfinished. Exact source, original failures, receipt limits and live evidence: https://github.com/OpenHCSDev/openhcs/blob/plan/issue-135-viewer-bind-20260929/docs/validation/viewer_current_main_integration_20260930.rst. Locally accepted installed/live checkpoint can ship without optional hosted CI under the explicit owner override; wider lifecycle races remain separate work.