[bokelley]

Summary

Adopts the signed AdCP 3.3.0-beta.3 bundle on next (the pre-release line; RELEASING.md routes draft-spec work there, and the protocol repo's "Generate Python SDK" job already checks out next).

• Bundle: schemas/releases/3.3.0-beta.3.json pins sha256 ba61919ca0c9d67874604fbbca8945c133eaceed65da1ac962f78d94d7879f00. I verified the .sha256 sidecar and ran cosign verify-blob against identity release.yml@refs/heads/main. Source commit d5f8de38 is read from the Fulcio certificate. The tag v3.3.0-beta.3 points at the Version Packages commit a6106149, and the signed workflow ran on the docs commit after it; the same split as noted in the test comment.
• Package the AdCP 3.3 schema bundle (3.3.0-beta.0) #1560: a prerelease keeps its full-version bundle key. pyproject.toml package data and MANIFEST.in now name 3.3.0-beta.3 (_schemas, _compliance, schemas/cache) and prune beta.2 from distributions.
• Generated response types can't represent the protocol envelope #1678 / post_generate_fixes.py matches exact base-class lists and breaks when envelope composition changes #1680: beta.3 is the first bundle with the envelope composed on every response (adcp#8547). Generated responses derive from ProtocolEnvelope (status, task_id, timestamp, adcp_version, ...). Two real generator problems surfaced against the real layout, both fixed generically, not per-schema (details below). The task-message lattice residue pins are now empty.
• Version gating (matches adcp-client#3493): uses_task_status_envelope(tool, version) reads the served release's own response schema for whether it composes protocol-envelope. 3.3 requires and emits status on list_products / decline_proposals; 3.2.x neither requires nor emits it (the typed model's completed default is not sent); convert_payload supplies it going up and drops it, recorded as a loss, going down. No other tool or release is touched.

Generator fixes

1. restore_response_variant_aliases re-emits eight response modules from raw schema JSON and recognised a submitted arm only by a local status: const submitted. Under #8547 that const and the required task_id live in core/task-submitted.json, so CreateMediaBuyResponse3 and seven siblings silently lost TaskSubmitted, the const and the required id. The emitter now recognises an arm that composes task-submitted. (Raw datamodel-code-generator output was already correct, so this one is in the post-generation pass.)
2. A scalar that tightens a shared definition (comply_test_controller task_id, maxLength: 128 beside core/task-id.json) lost its bound. A small prep pass merges a scalar's own bound into the plain scalar its lone allOf $ref names; two schemas are affected today.

Behavior notes for review

• BREAKING (pre-release): beta.2 is no longer an accepted pin; see the commit's BREAKING CHANGE and MIGRATION_v9_to_v10.md.
• TaskId / TaskMessage scalars on responses: adopters construct TaskId("task_1") (domain path adcp.types.domains.core.task_id), same convention as PropertyTag.
• Governance fixtures had passed status="approved" / "accepted" and only validated because the model allowed extras; status is now the typed task status, so those tests use verdict / outcome_state.
• New reporting_subscription_probe controller scenario (store method + param validation).
• _reporting_subscription_preview.py now only widens the extensible reporting-dimension vocabulary; every field it added is generated.
• Hook note: the commit skipped only the local black pre-commit hook (SKIP=black). It pins 24.10 and reformats unrelated lines that the repo's committed style (newer black) disagrees with; CI runs ruff, not black. ruff, mypy, bandit and the other hooks ran.

Test plan

• pytest tests : 20,800 passed, 0 failed, 2,829 skipped, 1 xfailed
• make typecheck-all (mypy src, mypy --strict adopter fixtures, pyright, type-ignore contract)
• make lint, make validate-generated (regeneration equality)
• New tests/test_response_status_gating.py (26 tests), generator-safety and submitted-arm contract tests
• Reproduced the protocol repo's "Generate Python SDK" job locally against the published bundle (result in the PR comments)

Fixes #1560. Fixes #1678. Fixes #1680.

🤖 Generated with Claude Code

[aao-secretariat]

Ladon cannot review this PR until merge conflicts are resolved.

[bokelley]

Reproduced the protocol repo's Generate Python SDK job (validate-schema-bundle.yml) locally against a fresh clone of this branch (80c5ed17):

• the published 3.3.0-beta.3.tgz served from a local HTTP host with its .sha256 (checksum verified);
• the job's bundle-key step: resolve_bundle_key is 3.3.0-beta.3 on both sides, so no pyproject.toml / MANIFEST.in edit is needed (this was the #1560 failure);
• the candidate's release pin removed and PREVIEW_VERSIONS = (), as the job does;
• make regenerate-schemas (ADCP_SKIP_SIGNATURE=1), then make validate-generated (generated types up to date, so regeneration is deterministic from the bundle), then pytest tests/test_code_generation.py tests/test_schemas_version_pin.py (133 passed).

One difference from the real job: it uses the PR bundle built from a protocol PR head, not the published tarball, and checks out next at that time. The published bundle is the same tree the next Version Packages PR will carry.

🤖 Generated with Claude Code

[bokelley]

Closing as superseded. While this branch was open, #1754 / #1755 / #1756 landed the same adoption on next (the signed 3.3.0-beta.3 bundle, the controller task_id bound, the compact-response status, the error-code and replayed re-pins), and this PR then conflicted with it. After diffing the two I found what this PR still added and moved it to #1761:

• the served release's own schema deciding status on list_products / decline_proposals (not a version comparison);
• the typed status: completed default no longer reaching a closed 3.2.x root;
• status handled in both directions of the 3.2 ⇄ 3.3 conversion (3.3 → 3.2 decline_proposals raises today).

Not carried over, on purpose: TaskSubmitted ancestry and TaskId / TaskMessage typed fields on the submitted arms. next keeps those fields typed str with the schema constraints (the status const and required task_id still hold), so adopters never wrap a value, and I don't want to diverge from that.

The local reproduction of the protocol repo's "Generate Python SDK" job in the comment above was run on this branch; I will repeat it against next once #1761 merges.

🤖 Generated with Claude Code