[bokelley]

A reporting service currently gives every installed handler a class name derived from the service's address. Installing two sellers with different tool sets on the same service fails registration; a later service can also reuse a previous address. Allocate a fresh UUID for each new installation while preserving same-platform reinstall identity and strict conflicting-registration refusal.

The public regression installs sellers with distinct advertised tools, checks their delegation and reinstall behavior, and verifies explicit registry conflicts still fail without changing the original surface. This fixes the latent collision exposed by beta.10 Python 3.13 CI without changing protocol schemas, registry rules, dependencies or released artifacts. A proper cherry-pick -x forward-port to next will follow the stable source merge.

Validation:

• Original public same-service case reproduces the failure; repaired reporting/registry suites pass 176 tests. Independent protocol/security review passed 33 related cases; independent Claude Opus 5.5 code/security/DX review approved.
• make lint typecheck-all validate-generated and all applicable normal commit hooks pass.
• Full required UV_NO_CACHE=1 make test passed: 18,865 passed, 2,819 skipped, 10 deselected and one existing xfail; 92.16% coverage. All tests and normal coverage gates ran unchanged.
• Final-head required CI is awaiting only Python 3.12 coverage; this PR remains Draft until all required checks pass and fresh review completes. The separate non-required central-ledger checkout timed out before its policy script; the required IPR Policy / Signature check passed.

Fixes #1846

[aao-secretariat]

Ladon verdict: Approve

Approve — clean, well-scoped fix.

This PR swaps the installed reporting-handler class name from id(self):x to uuid4().hex (service.py:1433), fixing the beta.10 Py3.13 collision (#1846) where two sellers of the same original class with distinct per-instance tool sets produced an identical installed_name, tripping the "called twice with different tool sets" guard in __init_subclass__ → register_handler_tools. uuid4 guarantees a unique install name; the same-platform reinstall early-return (service.py:1324) still bounds registry growth to one entry per distinct install, and method qualnames preserve the stable original.__name__, so nothing that parses the class name regresses.

The new test (tests/test_reporting_install_identity.py) exercises same/different-service installs, distinct tool surfaces, delegation, and confirms the conflict guard still fires. Correctly labeled fix(reporting): — no public-surface, wire-shape, schema, layering, forward-compat, or credential changes; no semver or version-pathway obligation triggered.

No reviewer findings. Not gated (gated_paths: false), not high-risk, no author team gate. review_decision is REVIEW_REQUIRED but that only blocks when gated_paths is true, which it is not here. Falls through to row 9.