Skip to content

Commit ba7a608

Browse files
reid-spencerclaude
andcommitted
Record rc.25 as cut, published and verified
Every channel was checked against its own registry rather than the publish log: Maven coordinates, the npm dist-tag, which Homebrew formula moved, and that notify-blog was skipped. The released macOS binary was downloaded and run rather than inferred from a green workflow. The durable part is the ORDER. Both corpora migrated against a staged binary BEFORE the tag, so rc.25 shipped green where every previous RC with corpus cost shipped red. A corpus needs the rule, not a published release. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
1 parent 76cb9ea commit ba7a608

1 file changed

Lines changed: 25 additions & 30 deletions

File tree

NOTEBOOK.md

Lines changed: 25 additions & 30 deletions
Original file line numberDiff line numberDiff line change
@@ -33,36 +33,31 @@ unowned** — type-checking `put`/`return`/`require`, `find -type <unknown>` mat
3333
nothing silently, and whether `StatsPass.numPromptStatements` gets renamed. Each entry
3434
carries its `file:line` evidence so it need not be re-derived.
3535

36-
### Build state — verified 2026-08-25 by running the commands
37-
38-
**`../bin/riddlc` is `2.0.0-rc.24-33-f4076e2c`, staged 2026-08-25, and the local ivy
39-
artifacts are from THE SAME BUILD** (`scripts/publish-and-stage.sh` runs `publishLocal` and
40-
`nativeLink` in one sbt invocation, so both halves move together or neither does). It carries
41-
every one of the 17 completed tasks: rule ids, `validate --json/--fix/--no-msg-ids`, corpus
42-
mode, and multi-line `do`/`prompt`.
43-
44-
**No RC was cut for it, deliberately (Reid's call).** The corpora have not migrated to the
45-
rules shipped since rc.24, so cutting now would publish a knowingly-red RC and measure the
46-
certification floors on a red run. The order is: stage -> migrate the corpora -> cut a GREEN
47-
rc.25. Task files with exact, rule-id-derived censuses are dropped in
48-
`../riddl-models/task/` (51 errors, 38 of 190 models) and `../riddl-examples/task/`
49-
(30 sites, 4 models).
50-
51-
**Latest tag is `2.0.0-rc.24`; the branch is well ahead of it.** Run
52-
`git log 2.0.0-rc.24..HEAD` before attributing anything to the RC.
53-
54-
**BAST `FORMAT_REVISION` is 22** (`language/.../bast/package.scala:150`). The corpus's
55-
190 checked-in `.bast` files are a revision behind; riddl-models knows.
56-
57-
**Prettify's one-space-after-`is` fix has the widest byte-level reach of anything on
58-
this branch** — 188 of 188 corpus models had drifted. Whitespace is load-bearing
59-
because riddl-models diffs checked-in `.bast` against source byte for byte.
60-
61-
**Test floors are 2903 / 917 / 2864** (JVM / JS / Native), set at rc.24, where
62-
deliberate failures were ONE (riddl-examples' `dokn`, 5 morph errors). **That count
63-
goes stale in hours** — the corpora are live sibling checkouts, and riddl-models has
64-
committed *during* a certification run and turned a suite red with nothing in riddl
65-
having changed. **Confirm any red by its MESSAGE, never by comparing counts.**
36+
### Build state — verified 2026-08-25
37+
38+
**`2.0.0-rc.25` IS CUT, PUBLISHED AND VERIFIED.** Certified from a genuinely cold cache
39+
(`/tmp/sbt-verify-rc25`, 140K -> 283M): **3036 / 979 / 2994**, zero failures, zero `No
40+
tests to run`, 7/5/7 module legs. Every row landed on the number predicted BEFORE the run.
41+
TatSu clean; riddl-examples 9/9 and riddl-models 190/190 through the external validator.
42+
43+
**Every channel verified against its own registry, not the log**: 11 Maven coordinates at
44+
`2.0.0-rc.25` (the four retired modules correctly absent), npm published `with tag rc`,
45+
Homebrew touched ONLY `Formula/riddlc-rc.rb` with stable `riddlc.rb` still on 1.31.0, and
46+
**`notify-blog` skipped**. The released macOS binary was downloaded and run: it reports
47+
`2.0.0-rc.25` and carries the rule ids, the `--json` shape and multi-line `do`.
48+
49+
**Deliberate failures are ZERO, and the ORDER is why.** Both corpora migrated against a
50+
STAGED binary before the tag rather than after it, so rc.25 shipped green instead of red.
51+
That reverses the previous practice and is now recorded in the `/rc` skill as preferred: a
52+
corpus needs the RULE, not a published RELEASE. Do not go back to tagging red without a
53+
reason.
54+
55+
**`../bin/riddlc` is `2.0.0-rc.25-1-76cb9eab`** -- one commit past the tag, and that commit
56+
touches only `.claude/skills/rc/SKILL.md`, so the binary is rc.25 in everything that
57+
compiles. ivy and the binary are from the same build. Anyone wanting the exact tag takes
58+
`brew upgrade ossuminc/tap/riddlc-rc`.
59+
60+
**BAST `FORMAT_REVISION` is 23**, and riddl-models has regenerated all 190 `.bast`.
6661

6762
### `task/` — 2 open, everything else closed
6863

0 commit comments

Comments
 (0)