docs(releases): v1.0.0-alpha.0 reads as an unexplained stray, and it is a live positive control

bosun commented

2026-09-07 01:17:37 +02:00

Owner

A visitor to the releases page sees a v1.0.0-alpha.0 pre-release from July sitting oddly among the v0.6x line, with nothing saying what it is -- and it must NOT be deleted, because it is the only real-world instance a shipped gate points at.

Why it cannot go

cmd/rt/build_ref_check_test.go   TestBuildRefCheck_RefusesTheRealIncident
  "not a fixture: v1.0.0-alpha.0 is published history. It shipped with
   BUILD_BAKED_TOOLKIT_REF: main -- a tree tagged without being baked for the tag."
also referenced by: readme_pin_check_test.go (x2), adr_supersession_check_test.go,
                    build-ref-check.yml, mirror-release.yml since_tag default

🔴 The defect is baked into published history and cannot be reconstructed. Deleting the tag deletes build-ref-check's only positive control -- crew-doctrine#171 exactly: a repair deletes the live instance that was its own control.

What a reader would notice

"The oldest mirrored release is v1.0.0-alpha.0, published 2026-09-03, sitting below v0.34.0. Unexplained."

(The date is the reader's error -- it was created 2026-07-24. The unexplained-stray impression is fair.)

AC

The release notes on v1.0.0-alpha.0 say what it is, that it is deliberately retained, and that a gate depends on it
It is marked so a reader browsing releases does not read it as a current or intended version
The tag is NOT deleted, and the reason is recorded where someone tidying up would look

Anchor

Round-3 cold read raised it as unexplained; @bosun measured the dependency before recommending anything. Operator asked whether to drop it -- the answer is no, and this tracker is why.

A visitor to the releases page sees a v1.0.0-alpha.0 pre-release from July sitting oddly among the v0.6x line, with nothing saying what it is -- and it must NOT be deleted, because it is the only real-world instance a shipped gate points at. ## Why it cannot go ``` cmd/rt/build_ref_check_test.go TestBuildRefCheck_RefusesTheRealIncident "not a fixture: v1.0.0-alpha.0 is published history. It shipped with BUILD_BAKED_TOOLKIT_REF: main -- a tree tagged without being baked for the tag." also referenced by: readme_pin_check_test.go (x2), adr_supersession_check_test.go, build-ref-check.yml, mirror-release.yml since_tag default ``` 🔴 **The defect is baked into published history and cannot be reconstructed.** Deleting the tag deletes `build-ref-check`'s only positive control -- `crew-doctrine#171` exactly: *a repair deletes the live instance that was its own control.* ## What a reader would notice > *"The oldest mirrored release is v1.0.0-alpha.0, published 2026-09-03, sitting below v0.34.0. Unexplained."* *(The date is the reader's error -- it was created 2026-07-24. The unexplained-stray impression is fair.)* ## AC - [x] The release notes on `v1.0.0-alpha.0` say what it is, that it is deliberately retained, and that a gate depends on it - [x] It is marked so a reader browsing releases does not read it as a current or intended version - [x] The tag is NOT deleted, and the reason is recorded where someone tidying up would look ## Anchor Round-3 cold read raised it as unexplained; @bosun measured the dependency before recommending anything. Operator asked whether to drop it -- the answer is no, and this tracker is why.

bosun added the

priority/low

size/S

kind/docs

labels

2026-09-07 01:17:37 +02:00

bosun commented

2026-09-07 08:24:49 +02:00

Author

Owner

🔴 The note said "retained test artefact" and that was WRONG. Corrected on the release object.

The operator supplied the history, and the repo corroborates every part of it.

v1.0.0-alpha.0 is not a test artefact. It is the only cut of an abandoned version line: the port from bash to Go was planned as a PARALLEL line, with v1.0.0-alpha.N for the Go work while the bash trunk kept releasing v0.x. The two-line model was then dropped, Go moved to trunk, and v0.x numbering resumed.

ADR-0009 phase 0a exit criterion:  "asset publish works to `v1.0.0-alpha.0` tag"
ADR-0009 §83:                      "v1.0.0-alpha.N (dogfood + iteration)"
ADR-0008:                          "retargets the port to v1.0.0 and collapses
                                    the two-line branch model"

v0.33.0          2026-07-23    bash trunk
v1.0.0-alpha.0   2026-07-25    the Go line's only cut
v0.34.0          2026-07-29    bash trunk, uninterrupted

No v1.0.0-alpha.1 was ever cut. That tag is the whole of the line.

⚠️ And the history was IN THE REPO the whole time

🔴 ADR-0009 names the tag explicitly, twice, as a phase exit criterion. Nobody read it. #1408's body — @bosun's — framed it as "a live positive control" and said nothing about where it came from; @shipwright wrote the note from that framing and neither of us opened the ADR that names the tag.

🔑 So this is not a case of provenance living only in the operator's head. It was written down, in the document series whose entire job is recording why decisions were made, and the artefact it explains sat unexplained for six weeks.

What the note says now

Name: v1.0.0-alpha.0 — the abandoned Go version line, retained for build-ref-check

Body: what it was (the Go line's first and only cut, with the ADR citations and the interleaved tag timeline), why it stays (build-ref-check's only real-world failing case — BUILD_BAKED_TOOLKIT_REF: 'main', unmanufacturable), and what it is not (not current, not pinnable, not evidence of a 1.0).

📌 #1426 still stands: the mirror will not carry this correction, because its idempotency is asset-granular. The audience the note is for still sees the one-byte version.

Correction applied by @bosun on the operator's account, verified against ADR-0008, ADR-0009 and the tag timeline before writing.

## 🔴 The note said "retained test artefact" and that was WRONG. Corrected on the release object. **The operator supplied the history, and the repo corroborates every part of it.** `v1.0.0-alpha.0` is not a test artefact. It is **the only cut of an abandoned version line**: the port from bash to Go was planned as a PARALLEL line, with `v1.0.0-alpha.N` for the Go work while the bash trunk kept releasing `v0.x`. The two-line model was then dropped, Go moved to trunk, and `v0.x` numbering resumed. ``` ADR-0009 phase 0a exit criterion: "asset publish works to `v1.0.0-alpha.0` tag" ADR-0009 §83: "v1.0.0-alpha.N (dogfood + iteration)" ADR-0008: "retargets the port to v1.0.0 and collapses the two-line branch model" v0.33.0 2026-07-23 bash trunk v1.0.0-alpha.0 2026-07-25 the Go line's only cut v0.34.0 2026-07-29 bash trunk, uninterrupted ``` **No `v1.0.0-alpha.1` was ever cut. That tag is the whole of the line.** ## ⚠️ And the history was IN THE REPO the whole time 🔴 **ADR-0009 names the tag explicitly, twice, as a phase exit criterion. Nobody read it.** *`#1408`'s body — @bosun's — framed it as "a live positive control" and said nothing about where it came from; @shipwright wrote the note from that framing and neither of us opened the ADR that names the tag.* 🔑 **So this is not a case of provenance living only in the operator's head. It was written down, in the document series whose entire job is recording why decisions were made, and the artefact it explains sat unexplained for six weeks.** ## What the note says now **Name:** `v1.0.0-alpha.0 — the abandoned Go version line, retained for build-ref-check` **Body:** what it was (the Go line's first and only cut, with the ADR citations and the interleaved tag timeline), why it stays (`build-ref-check`'s only real-world failing case — `BUILD_BAKED_TOOLKIT_REF: 'main'`, unmanufacturable), and what it is not (not current, not pinnable, not evidence of a 1.0). 📌 **`#1426` still stands: the mirror will not carry this correction, because its idempotency is asset-granular.** *The audience the note is for still sees the one-byte version.* *Correction applied by @bosun on the operator's account, verified against ADR-0008, ADR-0009 and the tag timeline before writing.*

bosun commented

2026-09-07 08:30:07 +02:00

Author

Owner

✅ Closed on the CANONICAL forge. The mirror half is `#1426` and is deliberately not part of this close.

name  "v1.0.0-alpha.0 — the abandoned Go version line, retained for build-ref-check"
body  1 byte -> 1939 bytes
tag   566288cc5d90, untouched

All three ACs met there: what it is, that it is deliberately retained and a gate depends on it, marked so a listing reader does not read it as current, and the reason recorded ON THE OBJECT — where someone about to press delete is standing.

🔴 And the first version of this note was WRONG

It said "retained test artefact, not a release line". It is a release line — the only cut of one. The bash→Go port was planned as a parallel version line: v1.0.0-alpha.N for the Go work while the trunk kept cutting v0.x. The two-line model was dropped, Go moved to trunk, v0.x resumed, and no alpha.1 was ever cut.

ADR-0009 phase 0a exit criterion   "asset publish works to `v1.0.0-alpha.0` tag"
ADR-0008                            "retargets the port to v1.0.0 and collapses
                                     the two-line branch model"
v0.33.0 2026-07-23 · v1.0.0-alpha.0 2026-07-24/25 · v0.34.0 2026-07-29

⚠️ The history was in the repo the whole time. ADR-0009 names the tag explicitly, twice, as a phase exit criterion — and neither the tracker author nor the note author opened it. This body called it "a live positive control" and said nothing about its origin; the note was written from that framing.

🔑 @shipwright's account of the failure is the one to keep: "The moment I was best placed to ask WHAT IS THIS was the moment I was writing the answer down. I asked instead what depends on it, measured sixteen files, and shipped a name asserting a category I had never checked. A dependency census is not an origin."

📌 Provenance supplied by the operator; verified against ADR-0008, ADR-0009 and the tag timeline before the note was rewritten.

## ✅ Closed on the CANONICAL forge. The mirror half is `#1426` and is deliberately not part of this close. ``` name "v1.0.0-alpha.0 — the abandoned Go version line, retained for build-ref-check" body 1 byte -> 1939 bytes tag 566288cc5d90, untouched ``` **All three ACs met there: what it is, that it is deliberately retained and a gate depends on it, marked so a listing reader does not read it as current, and the reason recorded ON THE OBJECT — where someone about to press delete is standing.** ## 🔴 And the first version of this note was WRONG **It said *"retained test artefact, not a release line"*. It is a release line — the only cut of one.** *The bash→Go port was planned as a parallel version line: `v1.0.0-alpha.N` for the Go work while the trunk kept cutting `v0.x`. The two-line model was dropped, Go moved to trunk, `v0.x` resumed, and no `alpha.1` was ever cut.* ``` ADR-0009 phase 0a exit criterion "asset publish works to `v1.0.0-alpha.0` tag" ADR-0008 "retargets the port to v1.0.0 and collapses the two-line branch model" v0.33.0 2026-07-23 · v1.0.0-alpha.0 2026-07-24/25 · v0.34.0 2026-07-29 ``` ⚠️ **The history was in the repo the whole time.** *`ADR-0009` names the tag explicitly, twice, as a phase exit criterion — and neither the tracker author nor the note author opened it. This body called it "a live positive control" and said nothing about its origin; the note was written from that framing.* 🔑 **@shipwright's account of the failure is the one to keep: *"The moment I was best placed to ask WHAT IS THIS was the moment I was writing the answer down. I asked instead what depends on it, measured sixteen files, and shipped a name asserting a category I had never checked. A dependency census is not an origin."*** 📌 **Provenance supplied by the operator; verified against `ADR-0008`, `ADR-0009` and the tag timeline before the note was rewritten.**

bosun closed this issue

2026-09-07 08:30:07 +02:00