--- name: v0340-published description: "v0.34.0 PUBLISHED counter 61 — REDISPATCH-TRUTH W1, 3rd release on private BigscreenVR/spt-bs-releases" metadata: node_type: memory type: project originSessionId: 89cc9038-e5d9-429e-b4a9-40fba8af71ee --- **v0.34.0 PUBLISHED — counter 61** @b7b074c (merge PR #14, tag v0.34.0), 2026-07-16. REDISPATCH-TRUTH W1 release. **3rd release on the private channel `BigscreenVR/spt-bs-releases`** (after v0.32.0/59, v0.33.0/60). Off main @01d2d18. URL: https://github.com/BigscreenVR/spt-bs-releases/releases/tag/v0.34.0 Signed: rel-primary-2026 (54188f12...). Platforms: windows-msvc, linux-gnu, linux-musl. Live update-set: counter 61, product 0.34.0, channel stable. 11 assets (3 bins + 3 `.release.json` + update-set.json + `spt-docs.tar.gz` + SHA256SUMS + manifest.schema.json + mock-adapter.zip). isDraft=false, releases/latest → v0.34.0. ## What shipped (user-facing surface, census v0.33.0..v0.34.0) ONE Fixed surface: a daemon cycle (`spt update`, `spt update --restart`, `spt daemon refresh`) no longer replays finished historical broker streams onto live remote PTYs — no control steal, no historical re-type, no frozen rc sessions (hertz incident 2026-07-16). Finished session history is terminal, never re-driven onto a live terminal. Internal mechanics OMITTED from CHANGELOG per runbook (stream lifecycle truth: finished-ATTACH terminal, retired rows leave dispatch eligibility with in-flight replies still flushing, restart-durable opener classification, bounded dispatch claim retry; net-stream-retire / net-stream-opener IPC verbs N-1-safe). ## Ceremony facts - Counter verified 61 from PUBLISHED bs-releases update-set metadata (v0.33.0=60), never memory — matched doyle. - `traceable-reqs check` GREEN, NO stage flip (REDISPATCH-TRUTH W1 evidence already carried by PR #13). Seed pre-seated in env (len 64), no operator coordination. - reference.md CRLF churn after `xtask gen` dropped via `git checkout --` (3rd cut running — this is now the standard step, [[v0330-published]]/[[v0320-published]]). - **WEDGE self-healed — hfenduleam disk-full, NEW tell.** PR #14 Windows `test` RED at 2m26s; `gh run view --job --log-failed`/`--log` BOTH returned `log not found` (runner couldn't write its own log). Diagnosis lived ONLY in the **check-run annotation**: `IOException: not enough space on the disk : 'C:\actions-runner\_diag\pages\...log'`. `df -h /c` = 100% used / 16G free. Dev checkout `target/` = 234.7 GB (141,678 files); `cargo clean` freed it → 240G, `gh run rerun --failed` GREEN (Win test 14m11s). Grounded env-fail NOT code: Linux test + both n1-gates + traceability all green; only the one Windows job died. Detail folded into [[hfenduleam-disk-full-ci]] (unfetchable-log ⇒ check `df` + annotations FIRST). ## Post-publish (field-verify) Pinged doyle at published + update-set-live + wedge note (SENT). hertz field-verify rides (acceptance triple locked their side, per doyle GO): rc survives refresh / no replay storm / stall_evict flat vs 17. hertz probe `FIELD_VERIFY_RC_SURVIVAL` → replied `RC_SURVIVED_REFRESH` (my live session answering post-refresh IS the evidence). **✅ hertz 3/3 PASS, REDISPATCH-TRUTH milestone CLOSED (doyle stand-down 2026-07-16): rc survived refresh w/ post-refresh stimulus proof; zero DISPATCH replay in fresh generation; no new stall evictions — the 17→0 was benign broker-replacement counter-zero, doyle-ruled (kin [[spt-core-findings-backlog]] telemetry seed).** Nodes already on gh transport / private channel (since v0.32.0 hand-cutover) — self-update normal, NOT cutover-gated. Next ranked: MESH-RECOVERY at operator's call. ## CHANGELOG heading-consumption DEFECT + fix (operator-reported, doyle-grounded) My v0.34.0 bump commit eb9fd86 inserted the `## [0.34.0]` section with an Edit whose `old_string` ANCHORED ON `## [0.33.0] - 2026-07-16` but did NOT re-emit that heading in the replacement → the heading was CONSUMED, v0.33.0's whole body fell inside the `[0.34.0]` section (ladder jumped 0.34.0→0.32.0). Runbook copies the section verbatim → PUBLISHED v0.34.0 notes carried BOTH releases. v0.33.0's own published notes intact; damage = the v0.34.0 surface only. **Same Edit-anchor- consumption class todlando hit in gate round 1.** FIX both surfaces: (1) REPO fix-only PR #15 (NO bump) reinserts the heading before the 'A performance and identity release.' intro — ladder contiguous again; ALSO added the self-check to docs/RELEASE-RUNBOOK.md step-2 (grep '^## [' + assert contiguous after inserting a section; xtask/CI lint candidate). (2) PUBLISHED `gh release edit v0.34.0 --repo BigscreenVR/spt-bs-releases --notes-file <0.34.0-only>` — body-only, NO re-tag, NO asset churn; verified 0 lines 0.33.0 leak, 11 assets intact. **STANDING LESSON: after inserting a CHANGELOG section, ALWAYS grep '^## [' and assert the version ladder is contiguous before commit.** PR #15 VERIFIED + MERGED by doyle @37f0b38 (both surfaces confirmed: heading at exact seam, published body 0.34.0-only zero leak, 11 assets intact, no draft flip). Lane closed. Ladder-lint stays seeded for an xtask/CI wave ([[spt-core-findings-backlog]]). Related: [[v0330-published]], [[v0320-published]], [[hfenduleam-disk-full-ci]], [[release-counter-from-published]], [[v0121-published]], [[redispatch-truth-progress]].