Hosted-relay diagnostic preflight, 2026-09-11. EXECUTION NOT STARTED.
Owner hertz; results to doyle and emphasys. Separate from #304 implementation.

Pinned all-0.69.0 source: v0.69.0,16df0e41be04ca4614a792f7e70e4d1bdec60b24, detached .worktrees/hertz-relay-069. No target pool/build created there. Private released executable: bin069/spt.exe under this evidence directory. SHA256689503ad9fc6cfde637cbcacf21c42faf130ba465a0df6c356cf290e2adafb85; private --version returned spt0.69.0. Only --version and adapter/daemon-run help executed, exit0, with cwd and SPT_HOME pointing to preflight-home. Inherited SPT_*, OWL_* and OMP_* variables stripped before setting the private home. No fixture daemon, listener, endpoint, or viewer started.

Capacity/load snapshot:19,979,182,080 bytes free (~18.6GiB). No cargo.exe, cargo-nextest.exe, rustc.exe or Runner.Worker.exe matches at that sampled instant. This is not a continuous zero-load proof or a test-child census. Fresh Rust build deferred at this capacity; no automatic reclaim, no shared target deletion, no use of a foreign pool. Doyle retains infra sweep.

Isolation gap still open: released daemon network configuration uses mDNS and real interfaces (pinned daemon.rs:539-540). SPT_TEST_EPHEMERAL_ADVISORY_PORTS separates the advisory ports but does not establish loopback-only networking. A private home and private executable alone do not prove every network interaction is isolated. No daemon started on the assumption that they do. A hermetic test-hosted network can use existing NetConfig loopback/disabled-discovery controls, but requires a verified fixture/build path. Public-flow proof must not be silently replaced by an internal helper test.

Planned controlled sequence, once preflight is satisfied:
- Fresh home per arm, unique socket identity derived from that home; private adapter/source and terminal harness; no resident service manager, live identity, real adapter session or live endpoint.
- Record actual binary versions/hashes, manifest and harness hash, broker/coordinator/terminal-child/listener PID+birth+image identities and ready/bind outputs. Assert actual socket/port separation before proceeding.
- Control: real hosted bind then listener establishment with no prior relay death. Capture classification and actual local advertisement after it settles. Prove the broker-held terminal remains usable with fresh post-check output/input, not merely child liveness or replayed old output.
- Replacement: separate fresh fixture; same bind/listener baseline. Terminate ONLY the identity-verified relay process, not its parent/tree. Confirm its exit. Preserve the SAME broker and PTY/session identities, prove fresh terminal usability, and snapshot classification before replacement. Start replacement listener; retain before/after classification and settled advertisement. A remote viewer is optional and must be labelled separately from local/internal proof.
- Cleanup success/failure/timeout: record every owned process identity before intervention; terminate only owned verified identities, stop supervisor before children if forced cleanup is needed, wait/reap retained handles, then census private-image/private-home descendants. Any unproven custody or survivor stops the sequence. Run control and replacement sequentially with a clear survivor census between arms.

Mixed arm remains separate and unexecuted: broker0.67.1 source04e32c8c95cf09ddc2a44a51cd0233b0d13bdc64 with coordinator/caller0.69.0. Old release assets are available through BigscreenVR/spt-bs-releases/v0.67.1; exact binary not downloaded/verified here. Must first establish single-coordinator supervision and pinned-source compatibility. Do not imitate existing n1_pairing by running a second coordinator alongside the old one. No deployment/update of live specimens.

Retained second specimen assessed: C:/Users/decid/AppData/Local/Temp/omp-slammie-rc/findings.txt. Reported broker0.67.1, coordinator/caller0.69.0, adapter0.9.3; controller accepted05:25:43.710Z, detach05:39:36.995Z; listener restart06:01:37Z; now HARNESS ONLY. No classification snapshot spans replacement and no fresh PTY-survival proof. Earlier control acceptance prevents treating initial no-row text as permanently absent PTY. No causal transfer to/from an all-0.69.0 fixture without executed compatibility evidence.

Current result: PRE-FLIGHT HOLD, not a reproduction, not a passed control. Only private preparation files/help calls exist. No live changes or cleanup of unrelated state performed.