fix(mcp): render parked refusals and graph-tool errors as typed fields

[ScriptedAlchemy]

Fixes #2280

What was wrong

• Parked refusals (fix(mcp): code-index corruption refusals format park facts into a route detail #2280). Since 21552e3 the admin-sync (tracedecay sync / tracedecay init) and background freshness-probe refusals carried the park as typed Parked detail inside the route error, but every consumer still flattened it. The CLI printed Display: one sentence, The code index for this worktree is parked; remedy: …; cause: …, and the 512-byte bound on that sentence cut the cause off. The daemon log did the same (error=project route error (…): …). The MCP JSON-RPC problem data had the detail but no kind. Branch publication still raised the same park as a format!("{reason}; {remediation}") string in two places.
• Graph-tool handler errors. The owner mapped a handler error with no kind of its own (database, io, json, …) to unavailable/authority, which claims a missing authority. It is an internal failure.

Change

• CodeIndexConvergenceParkedV1::publication_authority_corrupt_error (contracts) is now the one constructor for this refusal. Admin sync, the freshness probe, and both branch-publication sites use it, and the server-local copy is deleted.
• MCP: project_route_problem takes the problem kind for a typed detail from ApplicationProblem::from_detail, so a parked refusal is {kind: unavailable, code, reason_code, retryable: false, detail: {kind: parked, cause, remedy, retries_on_wake}}.
• CLI: the process boundary prints a route refusal with typed detail as its reason code plus one label: value line per detail field (ApplicationProblemDetailV1::labelled_fields). Nothing is truncated.
• Daemon log: the startup catch-up and read-refresh refusals log reason_code, cause, remedy, retries_on_wake as tracing fields.
• Graph-tool owner: every handler error now has a kind. Argument errors are invalid_request, route refusals are unavailable under their reason code, typed details/lock deadlines keep their detail, and any other handler failure is execution_failed (graph_tool.failed, retry: never, legal_actions: [contact_administrator]), the contract's internal-failure kind. The test(mcp): pin owner refusals as typed isError problems #2319 isError problem-record shape is unchanged. The four refactor(mcp): answer info and runtime reads through their owners #2335 pins for a broken registry table now assert execution_failed.

Fail before / pass after

With the behavioral pieces reverted (tests kept), these fail with the literals below and pass on this branch:

• mcp::tools::handlers::dispatch_tests::admin_sync_reports_terminal_publication_corruption_without_queueing: wire data lacked "kind": "unavailable", "code": ….
• tool_errors::tests::project_route_parked_detail_stays_structured_on_the_wire: same.
• commands::index::init_bootstrap_tests::brokered_init_prints_a_parked_refusal_as_fields: fixture daemon answers with the real tool_error_response. Left was project route error (code_index_publication_authority_corrupt): The code index for this worktree is parked; remedy: …; cause: … xxxx (cut); right is project route error (…)\nParked cause: <full 600+ byte cause>\nParked remedy: run \tracedecay daemon restart`\nRetries on wake: false`.
• mcp::server::background_refresh_writer_tests::parked_freshness_probe_logs_cause_and_remedy_as_fields: left error=project route error (…): … parked; remedy: …; cause: … xxx (cut); right reason_code="code_index_publication_authority_corrupt" cause="<full>" remedy="run \tracedecay daemon restart`" retries_on_wake=false`.
• code_index_scheduler::tests::branch_publication_tests::mid_wait_branch_publication_surfaces_terminal_publication_park: typed detail was None.
• mcp::tools::handlers::application_surface::tests::every_graph_tool_handler_error_renders_as_a_kinded_problem: io error kind was unavailable, expected execution_failed.

Runtime proof (debug CLI, isolated HOME, one daemon under systemd-run --scope -p MemoryMax=6G)

Corpus: a scratch git repo. Its code-index publication was corrupted (garbage generation manifests) and the scope store made read-only, so the automatic reset fails and the worktree parks terminally.

Master binary, tracedecay sync:

Error: project route error (code_index_publication_authority_corrupt): The code index for this worktree is parked; remedy: the derived code-index publication is corrupt and could not be deleted; fix the named filesystem fault on the project's code-index-v1 scope store, then run `tracedecay daemon restart` to rebuild it from source; cause: code-index production owner failed: the publication authority is corrupt and requires an index reset: parent generation manifest identity is corrupt; reset failed: code-index production owner failed: the publication authority is unavailable:

(the cause is cut before code-generation retention storage failure: Permission denied (os error 13))

Branch binary, same profile, tracedecay sync (exit 1):

Error: project route error (code_index_publication_authority_corrupt)
Parked cause: code-index production owner failed: the publication authority is corrupt and requires an index reset: active code-generation byte size does not match its durable entry; reset failed: code-index production owner failed: the publication authority is unavailable: code-generation retention storage failure: Permission denied (os error 13)
Parked remedy: the derived code-index publication is corrupt and could not be deleted; fix the named filesystem fault on the project's code-index-v1 scope store, then run `tracedecay daemon restart` to rebuild it from source
Retries on wake: false

Daemon JSON-RPC error data: {"code": "code_index_publication_authority_corrupt", "detail": {"cause": "…Permission denied (os error 13)", "kind": "parked", "remedy": "…", "retries_on_wake": false}, "kind": "unavailable", "reason_code": "code_index_publication_authority_corrupt", "retryable": false, "tool": "tracedecay_admin_sync"}
Daemon log after a read triggered the freshness probe:

WARN tracedecay::mcp::server::lifecycle: background read reconciliation was not admitted reason_code="code_index_publication_authority_corrupt" cause="…Permission denied (os error 13)" remedy="the derived code-index publication is corrupt …" retries_on_wake=false

Graph tools already rendered the whole record on master (tracedecay tool tracedecay_rank --args '{"direction":"bogus"}' --json: isError: true, problem.kind: invalid_request); that is unchanged.

Suites

• Focused (filters parked publication admin_sync graph_tool brokered_init project_route freshness_probe, after merging current master): tracedecay lib 38 passed; tracedecay-cli bin 5 passed; tracedecay-code-index-runtime lib 87 passed; tracedecay-mcp lib 9 passed.
• Full mcp_suite (rebased tip, CLI built first, run outside any memory scope): test result: ok. 580 passed; 0 failed. Master itself now has 580 tests: since this lane started, master removed 6 and added 1, and this PR adds or removes none. Two earlier runs under heavy shared load each hit unrelated readiness timeouts (code index did not publish … after 20000 ms; composition gate … waiting=88, and one diff_context truncation test that passes alone). The clean run above is the evidence.
• cargo clippy -p tracedecay-contracts -p tracedecay-code-index-runtime -p tracedecay-mcp -p tracedecay -p tracedecay-cli --all-targets -- -D warnings: clean with and without tracedecay/test-transport,tracedecay/test-helpers,tracedecay-cli/test-transport.
• cargo fmt --all -- --check: clean. dashboard contracts:check: contracts up to date.
• cargo check --workspace --all-targets --target x86_64-pc-windows-gnu --features tracedecay/test-transport,tracedecay/test-helpers,tracedecay-cli/test-transport: exit 0. No cfg(windows) code touched.

Left as is

tracedecay status already reports the park as code_index_freshness.parked.{reason, remediation} fields in --json. Its human warning line still joins them in one sentence, and this PR does not change it.

[changeset-bot]

⚠️ No Changeset found

Latest commit: 7a392cd

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-27T15:25:23.894588Z	7a392cd	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.