feat(training): serve the AdCP 3.2 RC.1 surface

[bokelley]

Summary

• serve the AdCP 3.2 RC.1 surface from the training agent while retaining RC.0 compatibility
• expose the Reliable Reporting RC.1 lifecycle and required revision-digest semantics
• align the banner, release notes, migration guidance, training, and SDK setup across TypeScript, Python, and Go
• add the reviewed 3.1 → 3.2 comparison: schema/tool/compliance growth, feature vs. hardening mix, contributors, and the 76–95% reduction in comparable MCP validation-schema artifacts

SDK releases

• TypeScript: @adcp/sdk@14.0.0-rc.32 (release PR)
• Python: adcp==8.0.0b14 (release PR)
• Go: github.com/adcontextprotocol/adcp-go/adcp/v3@v3.2.1 (release PR)

Verification

• docs, metrics, and product expert reviews: pass
• pre-commit repository suite: pass
• server unit suite: 563 files / 8,190 passed / 32 skipped
• TypeScript typecheck: pass
• RC.1 reporting/controller/integration tests: 121 passed
• tenant/certification tests: 77 passed
• targeting discovery tests: 22 passed
• docs navigation: 55 passed
• schema link lint and git diff --check: pass

Official RC: https://github.com/adcontextprotocol/adcp/releases/tag/v3.2.0-rc.1

---

Open workspace in Conductor

[aao-secretariat]

Ladon verdict: Approve

Approve — RC.1 go-live alignment with no blocking or medium findings.

The reviewer verified the full chain of this release-alignment PR: version-constant→advertisement propagation, intentional NODE_ENV gate removals with per-request enforcement intact, the behavior-preserving reporting-reliability type refactor, migration 581 (idempotent, S1-guarded follow-up to 580, judged as production-facing), and an in-scope patch changeset with no static/schemas/source changes. No Critical/High/Medium findings were reported.

Gate check: high_risk is true, but all docs/reference/** and server/src/addie/** matches are (modified) with no medium-or-higher findings, so the modifications are presumed safe (no row 3 deletion, no row 4/5 medium trigger). gated_paths is false, so row 2 does not apply despite review_decision: REVIEW_REQUIRED. No no-auto-approve team match. No prior decision. With zero findings, none of rows 1–8 fire — falls through to row 9: approve.

[aao-secretariat]

Ladon verdict: Approve

Approve — subsequent pass, clean delta.

The delta since the prior approve (head c80587e) is generated/lockfile-only: package-lock.json dependency bumps (@xmldom/xmldom 0.8.15, fast-uri 3.1.7, libc field removals) plus two regenerated Addie tool-surface artifacts (budget contract-hash rotation with status within_budget and baseline id unchanged; wire_schema_sha256 rotations in tool-surface-inventory.generated.json). No hand-authored source, schema, docs, changeset, or migration in the delta.

No critical/high/medium findings surfaced. high_risk is true only from (modified)/(added) reference-doc and Addie-path files, but with no medium-or-higher concern the modifications are presumed safe (no deletions). gated_paths is false; no no-auto-approve team match. Prior decision was approve, so sticky-escalation (row 6) does not apply.

Decision table: rows 1–8 do not fire → row 9 = approve.