From 84574e6e115c359ba0707cbd3c4cc655bb3a6c58 Mon Sep 17 00:00:00 2001 From: Ayla Croft Date: Mon, 7 Sep 2026 21:13:27 -0400 Subject: [PATCH] Hold the 0.3.1 tag and publish until the CI flake is fixed The hold goes at the top of "To release 0.3.1" because that is the section a releaser reads, and a hold recorded anywhere else does not stop a tag. Not a defect in 0.3.1. Measured: PR #15's pull_request run passed and its push run failed on the SAME commit, and that commit contains no code -- only a PLAN file and an archived specification page. run 34174978505 event=pull_request success run 34174975336 event=push failure test FAIL (exit 2), max_cases: 8 HTTPBanditTest "a refusal over the adapter's drain cap now announces the close it always did" While that is true, "gate green, every step line reading pass" is a sample rather than a measurement -- and that sentence is the entire evidence that 0.3.1 is fit to ship. Tagging now would publish on a green nobody can distinguish from a lucky one. SCR-289, slice 006. Lift the line when that is on main, then re-run the gate on the release commit and read the step lines. Gate OK. GATE_EXIT=0 Signed-off-by: Ayla Croft --- HANDOFF.md | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/HANDOFF.md b/HANDOFF.md index 6758f8f..443db47 100644 --- a/HANDOFF.md +++ b/HANDOFF.md @@ -116,6 +116,19 @@ Every one of these is in `slices/003-release-0-3-1/FINDINGS.md` with its measure ## To release 0.3.1 +> **HELD. Do not tag and do not publish.** Owner instruction, 2026-09-08: the release waits until +> the CI test flake (SCR-289, slice 006) is fixed and merged. +> +> The reason is not the defect in `0.3.1` — there isn't one. It is that a test in this repository +> answers differently on two runs of the same commit, **measured**: PR #15's `pull_request` run +> passed and its `push` run failed on a commit containing no code. While that is true, *"gate +> green, every step line reading pass"* is a sample rather than a measurement, and that sentence +> is the whole evidence that `0.3.1` is fit to ship. Tagging now would publish on a green that +> nobody can currently distinguish from a lucky one. +> +> Lift this line when slice 006 is on `main`, then re-run the gate on the release commit and read +> the step lines before step 2 below. + 1. Merge the PR to `main` (the ruleset requires two green checks). 2. Date the `[0.3.1]` heading. 3. `mix hex.publish`, then tag `v0.3.1` signed — or tag first and publish immediately after.