Skip to content

release: v3.3.0 — bump workspace + inter-crate pins, CHANGELOG, README - #206

Merged
avrabe merged 1 commit into
mainfrom
release-v3.3.0
Sep 3, 2026
Merged

release: v3.3.0 — bump workspace + inter-crate pins, CHANGELOG, README#206
avrabe merged 1 commit into
mainfrom
release-v3.3.0

Conversation

@avrabe

@avrabe avrabe commented Sep 3, 2026

Copy link
Copy Markdown
Contributor

Release prep for v3.3.0 — "agent-consumable scry". rivet release status v3.3.0 = cuttable, 7 accepted / 0 proposed.

The bump moves 19 sites as one commit

cargo publish resolves each leaf from the crates.io index, so a stale version = "3.2.7" pin would publish a 3.3.0 core depending on the previous release.

count site
1 workspace version
13 pinned inter-crate deps (analyze-core ×10, viz, segment, mcp)
1 Bazel CARGO_PKG_VERSION pin — #152, the one that shipped scry 0.0.0 when missed
1 README shipped-claim
3 claims.yaml (2 texts + 2 patterns, minus overlap)

The claim-check gate did its job again, unprompted

My bump script failed partway through claims.yaml on a regex-escaping bug, leaving Cargo.toml and README moved but claims.yaml stale — and claim-check immediately went 7/7 → 5/7. That is exactly the drift it exists for, caught within one command of being introduced. Back to 7/7 after.

The historical "what v3.2.7 did" note in claims.yaml is deliberately not bumped (it refers to that release's shipped-0.0.0 bug); a guard in the script asserts it survives.

Two stale things the bump exposed, fixed rather than shipped

What the CHANGELOG says this release does NOT give you

Both are things a consumer would otherwise assume:

Falsification statement

Four criteria in the CHANGELOG, with the weakest flagged as such: the before/after comparison is not corpus-wide, and #126's reporter has offered theirs.

Note for the release runner

scry-sai-mcp is a first-time publish — v3.2.7's tree contained no crates/scry-mcp. Already wired into scripts/publish.rs in leaf-after-core position.

And per FEAT-066's release-machinery note: cargo package is not a valid pre-release check mid-cycle. It fails for a correct tree, because crates.io's scry-sai-core predates the Query and VerifyReport APIs. A red there means nothing; the publish path handles ordering.

Verified locally

cargo test 160 + 12 pass · fmt clean · rivet validate PASS · claim-check 7/7 · release-ordering PASS · required-checks PASS

Refs: FEAT-066

🤖 Generated with Claude Code

https://claude.ai/code/session_01KkNzkNYzPh7366DkNijeNc

Version bump moves the workspace version AND every pinned inter-crate
dependency together — 19 sites. They have to move as one: `cargo publish`
resolves each leaf from the crates.io index, so a stale `version =
"3.2.7"` pin would publish a 3.3.0 core depending on the previous
release.

  1   workspace version (Cargo.toml)
  13  pinned inter-crate deps (analyze-core x10, viz, segment, mcp)
  1   Bazel CARGO_PKG_VERSION pin (scry#152 — the shipped component
      announced `scry 0.0.0` when this was missed)
  1   README shipped-claim
  3   claims.yaml (2 claim texts + 2 count-min patterns, minus overlap)

THE CLAIM-CHECK GATE DID ITS JOB AGAIN, UNPROMPTED. The bump script
failed partway through claims.yaml on a regex-escaping bug, leaving
Cargo.toml and README moved but claims.yaml stale — and claim-check
immediately went 7/7 -> 5/7. That is exactly the drift it exists for,
caught within one command of being introduced. Back to 7/7 after.

The HISTORICAL "what v3.2.7 did" note in claims.yaml is deliberately NOT
bumped — it refers to the shipped-0.0.0 bug of that release, and a guard
in the bump script asserts it survives.

Also fixed two things the bump exposed rather than shipping them:
  * README said "the full v0.1 -> v3.2 arc is done" while announcing
    v3.3.0 — the same tense-of-a-claim drift repaired in #198.
  * crates/scry-mcdc/Cargo.lock still pinned scry-sai-core 3.2.7; the
    MC/DC gate builds scry_mcdc.wasm from it.

CHANGELOG carries the release's falsification statement per methodology,
with the weakest of the four criteria flagged as such: the before/after
is not corpus-wide, and #126's reporter has offered theirs.

It also states what this release does NOT give you, because both are
things a consumer would otherwise assume:
  * obligation identity is NOT stable across builds — FEAT-073 measured
    43-45% churn per build (#123). Anchors and the feed schema ship; the
    guarantee behind them does not. Treat an obligation_id as stable
    WITHIN a build only.
  * the MCP `verify` tool is structurally absent — FEAT-065 was refuted
    pre-ship with six wrong-verdict paths (#122).

NOTE: `scry-sai-mcp` is a FIRST-TIME publish. v3.2.7's tree contained no
crates/scry-mcp. It is already wired into scripts/publish.rs in
leaf-after-core position.

Verified: cargo test 160 + 12 pass, fmt clean, rivet validate PASS,
claim-check 7/7, release-ordering PASS, required-checks PASS,
`rivet release status v3.3.0` = cuttable (7 accepted / 0 proposed).

Refs: FEAT-066

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KkNzkNYzPh7366DkNijeNc
@avrabe
avrabe merged commit 0b315fd into main Sep 3, 2026
12 checks passed
@avrabe
avrabe deleted the release-v3.3.0 branch September 3, 2026 06:53
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant