---
name: v0393-published
description: "v0.39.3 counter 71 PUBLISHED 2026-07-21; docs-only frames encoding fix; template for naming a prior note's error"
metadata: 
  node_type: memory
  type: project
  originSessionId: 0cc427d2-c078-40d7-86d1-52aeb724e14c
  modified: 2026-07-21T15:48:26.711Z
---

v0.39.3 DOCS-ONLY PUBLISHED — counter **71** @bc561c7 (PR #50), 2026-07-21. Off main @b246493 (PR #48, doyle gate PASS FINAL). First cut under the standing dedicated-release-PR rule, and first cut driven entirely from `.worktrees/release-lane`.

**Why it existed:** flynn's conformance diff found the v0.39.2 frames page specified **no body/attribute encoding**. doyle ruled docs-only v0.39.3 rather than riding v0.40.0 — see the framing below.

**What the page now specifies** (enumerated from the merged section, not the wave description — that was the point): entity escape set on BOTH attributes and bodies; `<br>` for body newlines only; binding decode order (`<br>` first, amp LAST, over an extracted substring only, never the whole envelope line); CR normalization to `\n`, CRs unrepresentable; never-split guarantee (`<EVENT-PART>` never appears on shell surfaces).

**⭐ TEMPLATE — naming a prior note's error (doyle: "now the template").** Quote the prior wording → state what was absent → state what closes it. No self-flagellation, no quiet supersession. v0.39.2 claimed the page published each message type "with its attributes, body format, and per-message authentication stamp"; body/attribute encoding was absent; a decoder built strictly from it broke on the first `&`/`<`/`>`/`"`/newline. A reader who built against the old page is owed that statement, not just new text.

**⭐ VEHICLE FRAMING (doyle, corrected my bias).** I weighed "is another cut worth the overhead"; the real question is **"what does the wrong artifact cost while it stands"**. Overhead is bounded and ours; exposure is unbounded and theirs. A page shell authors clean-room from, proven to break a from-scratch decoder, is a live trap for every new author until fixed. Use this framing for vehicle questions.

**⭐ CLAIM CLASS ≠ EVIDENCE (doyle's sharpest distinction this arc).** A consumer's clean-room read is **input to the gate, not a verdict on the merge**, and a pre-ship diff changes the section's CONTENT, never the CLAIM CLASS. "Specified" is earned by what the artifact states; a clean read does not upgrade it to "complete". Consumer-verifies-core-decides ([[consumer-verifies-core-decides]]) applied to flynn exactly as to perri. My own contribution here was the mechanism that produced it: **enumerate the line from the published artifact at cut time, never from the wave description or the message thread** — v0.39.2's overstatement happened because I wrote what the page was SUPPOSED to cover without checking the artifact.

**Branch-first consumer read earned its keep:** flynn diffed off the branch pre-cut and DID find a chunking gap, closed on-branch (one edit away instead of one release away). It did not, and could not, upgrade the sentence.

**⭐ TWO-SAMPLE RULE PAID OFF SAME DAY.** Linux `n1-gate` sat ~25min alone AFTER its sibling Linux test had already passed on kitsubito — the wedge signature again. Two samples 45s apart: worker count 0→1, `busy:true` throughout, no competing in-repo run ⇒ genuine occupancy from another repo. It dispatched and passed in 1m24s. **The single sample I'd have trusted that morning was the 0, and it would have produced the wrong verdict a second time.** See the retraction in [[v0391-published]] and the incident in [[v0392-published]].

**Also merged this cut:** PR #49 — RELEASE-RUNBOOK step 1 retired the OR-form ("feature PR's final commit"), which was correct as operator-directed 2026-06-23 but **superseded, not mistranscribed**, by wave-branch/counter-drift rulings and v0.38.0–v0.39.2 practice. Written standard now matches practice. See [[release-standard-bump-in-pr]].

**Ceremony:** counter 71 verified from published metadata at authoring time (v0.39.2=70). Residual check with ESCAPED dots ([[v0392-published]] lesson). `xtask gen` no drift — confirmed by the CI drift gate passing, not assumed. Ladder contiguous, 0.39.2 intact. Full green both platforms. Signed rel-primary-2026, verified vs SHA256SUMS, update-set v71. Latest verified: counter 71, product 0.39.3, isDraft:false, isPrerelease:false.

**Open:** flynn's post-publish clean-room re-read of the PUBLISHED page (framed as build-from-this-page-alone; surprises route to doyle). Next: v0.40.0 NOTIF-TRUTH. Still unverified from earlier: v0.39.0's three TEARDOWN-AUTHORITY claims ([[v0390-published]]).