---
name: v0490-published
description: "v0.49.0 counter 84 published 2026-07-31T22:13:15Z @2a66d78 — access-control W4/W5 (releases#53); 3 golden attempts, all reds resource-class not code"
metadata: 
  node_type: memory
  type: project
  originSessionId: 19af1a2f-5196-41e5-8233-8bb5241f8280
  modified: 2026-07-31T22:13:52.333Z
---

v0.49.0 (counter **84**) published 2026-07-31T22:13:15Z @`2a66d78d853abfd862c92e8112785ae0839c0b76` (== main, ff by doyle). Milestone `releases#53`: knocking (W4) + monics/trust-warning (W5). **MINOR** bump justified from observable behavior: knock/monic verbs are additive opt-in (patch-level alone), the **trust warning** carries it past patch because it changes an existing surface — a message from an access-rule-admitted sender the recipient holds no note about now arrives with a caution, rewordable but not hideable.

Hashes: win `d13f8b5515cfb810adc90aa94c6315d2782443bcaedca1da20971ce35e818f16` · linux `ea831b9fb87ef16c54445fd45254a1dcd5ab5a92dcf9a534e21cffb4a2e9ba74` · musl `9199fac401e87b9ff6388308f59880a9befc2c1efddb5946246bfbcefb7b7fef`. update-set v84, key `rel-primary-2026`, channel stable, sig 128 hex, compat constants 1/1. Verified after publish: `isDraft=false`, 11 assets, `/releases/latest` → v0.49.0.

**Golden took 3 attempts; NO attempt re-rolled a real red.** Run 30660216982. Attempt 1: both Windows legs died on the 32 GB free-space floor. Attempt 2: floor passed, Windows `test` died `rustc-LLVM ERROR: out of memory` building spt-daemon lib test. Attempt 3 (bounded): full green, all 9 jobs. In every red **no test binary was produced** — platform UNTESTED, never failing. See [[free-space-floor-blocks-golden]].

**Two release-shape saves worth keeping:**
1. **The hand-off said "bump + changelog ride the release PR" — the runbook says the opposite** (step 1: bump goes IN the golden candidate; the dedicated release-PR form is retired; no post-golden edits). Stop-and-refer was correct — doyle confirmed the sentence was stale from his own notes. Running golden as instructed would have burned a full cycle on a sha that could never ship.
2. **Changelog novelty audit caught two near-misses:** the `spt endpoint access`/`engine-room`/`role` blocks appear as `+` lines in the CLI-reference diff but are only reflow — they shipped in v0.47.0; and endpoint `deny`/`remove` looked breaking until the full command block showed `revoke` survives. Diff the reference `main..head` AND read the whole block, never the `+` lines alone.

**Instrument trap logged:** `gh api repos/…/releases/tags/<tag>` returns **404 for a draft**, so a body-vs-changelog diff run pre-publish silently compares against a 133-char error payload and reports a false mismatch. Use `gh release view <tag> --json body` for drafts.

**Still open at publish:** `[build] jobs = 4` stopgap in `C:\Users\decid\.cargo\config.toml` (global to user decid, header names its removal condition) — remove once the operator lands machine-scope `CARGO_BUILD_JOBS` on the runner or `releases#56` is ruled. Two operator asks on #56: pagefile sizing, elevation. My docs lanes PR #140 (runbook step 6 — board-not-body association) and PR #141 (ring perch-safety contract) rode this batch.