HERTZ v0.38.0 FIELD VERDICT — FAIL (one controller entry-path defect; two additional rollout symptoms) Build: CLI 0.38.0, broker_image 0.38.0, broker_stale=false after `spt update --restart`; release counter 66. PASS — attach truth 1. Perri live-session-over-stale-offline: with perri broker session confirmed ready=true/alive=true, atomically changed only persisted info.status online→offline. `spt rc perri --view` remained attached/alive after 3s and did not emit the historical "offline — nothing to attach" refusal. Restored the original online record immediately. 2. Never-bound resume: temporary `hertz-v038-unbound` launched with `--resume field-never-bound --start`; persisted record and daemon status both reported status/state UNBOUND (`ready=false, alive=false, unbound=true`), not offline. Probe was stopped and purged. PASS — explicit same-machine Take + terminal fencing ConPTY A: `spt rc deployah`; ConPTY B: `spt rc deployah --take`. Observed A output: `[displaced — 'deployah' was taken over by 14ef...85f2d]`; A terminated exit 0; B remained attached. A's post-displacement write failed `EOFError: Pty is closed`, so the displaced viewport was terminally unable to type. FAIL — same-machine newer plain rc never reaches generation ladder With B still controlling deployah, ConPTY C ran plain `spt rc deployah`. C exited immediately with: `'deployah' is controlled by 14ef...85f2d.` ` spt rc deployah --view ...` ` spt rc deployah --take ...` B remained alive and received no Displaced. This contradicts the locked field expectation that same-node windows fall through to the broker's newer-generation loud supersession. The client-side `current_driver` preflight still treats the local node's driven_by stamp as a blocker. Broker W2 may be correct; the CLI entry path prevents exercising it. PASS — render lifecycle regression seams on the exact released tree - `cargo test -p spt-daemon --test render_lifecycle`: 1 passed; production broker→serve final frame precedes Exit. - `cargo test -p spt teardown_matrix_postlude_precedes_prose_across_every_final_class`: 1 passed; dirty-display teardown precedes prose across all terminal outcomes. - `cargo test -p spt purge_transition_reconstructs_the_complete_screen_after_external_mutation`: 1 passed; stateful picker buffer reconstructs after out-of-band mutation. - `cargo test -p spt-term repaint_replays_tracked_decstbm_before_cursor_placement`: 1 passed. - `cargo test -p spt-term repaint_then_raw_scroll_matches_the_server_grid`: 1 passed; reattach repaint + subsequent raw scroll is state-equivalent. These are deterministic released-tree behavioral checks, not visual screenshot claims. PASS — substrate health during probe window Daemon log tail after v0.38.0 restart/probes: `poisoned=true` count 0; Registry/Attach poison count 0; case-insensitive redispatch records count 0. Broker status remained image 0.38.0, stale=false, stall_evict_count=0. ADDITIONAL REPRO 1 — todlando false-active projection `spt endpoint list --json` projects todlando Active in both subnets. Local daemon state is `ready=false, alive=true, unbound=false`; endpoint type ready_agent. Yet `spt rc todlando` deterministically returns `No live session for 'todlando' — not running on this node or any visible subnet.` This may be a liveness/projection issue rather than attach truth: Active/online is being shown for a ready-agent process with no broker PTY session. ADDITIONAL REPRO 2 — duplicate internal diagnostic on every rc One `spt rc todlando` invocation prints `PUMP_IPC_READER: spawned` twice before the public no-session message. Same duplicate was observed on rc attach probes. This is user-visible internal diagnostic leakage and may also indicate two short-lived pump Brain constructions; RCA agents are tracing exact call sites. RULING Do not close RC-RENDER-TRUTH as a full field PASS. Explicit Take, fence, attach truth, render regressions, and substrate pass. Plain same-node rc supersession is unreachable due client preflight. Todlando projection and duplicate pump banner are independently reproduced and under source trace.