Phase 1 has 2 plans in 1 wave (parallel):
- PLAN-1.1: Delete dead ObjectPool code (3 files, zero references)
- PLAN-1.2: Add in-memory ActivityListener to SharedSetup for CI trace coverage

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…rage

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Refactor RunWithTraceVerification to delegate to shared helper
- Make activity collection opt-in (listener still always-on)
- Revert SharedSetup visibility to internal

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Phase 1 results: ObjectPool dead code removed, in-memory ActivityListener
added (always-on listener for coverage cascade, opt-in activity collection),
shared trace verification helper, 2 new CLAUDE.md lessons.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

3 plans in 1 wave:
- PLAN-1.1: New CreateJobTablesCommandHandler tests
- PLAN-1.2: New GetJobId + GetJobLastKnownEvent query handler tests
- PLAN-1.3: Expand existing GetDashboardJobs/ErrorRetries tests (sync+async)

Rescoped via research: SetJobLastKnownEvent and SendJobToQueue are
transport-specific only; deferred to Phase 3.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Introduces TestHelpers/AdoNetMockFixture (sync, interface-based) and AdoNetAsyncMockFixture (async, DbConnection/DbCommand/DbDataReader base classes) to centralize NSubstitute scaffolding previously duplicated across Phase 2 unit tests. Refactors CreateJobTablesCommandHandlerTests, GetJobIdQueryHandlerTests, and GetJobLastKnownEventQueryHandlerTests to use the shared fixture. No behavioral changes; 216 tests still pass.

Dashboard sync tests now use AdoNetMockFixture with SetupReaderRows(); async tests use AdoNetAsyncMockFixture. Eliminates duplicated NSubstitute scaffolding and multi-row reader ladder previously copy-pasted across the four files. All 216 tests still pass.

Phase 2 results: 39 new unit tests for shared job handlers in
Transport.RelationalDatabase. Extracted AdoNetMockFixture +
AdoNetAsyncMockFixture helpers (-81 net lines). Added 3 CLAUDE.md
lessons (sync/async DbConnection mocking, MSTest 3.x ThrowsExactly,
no CancellationToken on dashboard async handlers).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…factors

10 plans across 2 waves:
- Wave 1: Refactor SqlServer/PostgreSQL SetJobLastKnownEvent to inject
  IDbConnectionFactory (production code change to enable unit testing)
- Wave 2: New test files for SqlServer/PostgreSQL/SQLite JobSchema,
  SendJobToQueue, and the refactored SetJobLastKnownEvent handlers

Rescoped via research: LiteDb and Redis transport-specific handlers
deferred to a future phase due to LiteDb concrete-type and Redis
sealed-multiplexer testability issues.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…dler to inject IDbConnectionFactory

…ler to inject IDbConnectionFactory

Replaces the hardcoded new SqlConnection(_connectionInformation.ConnectionString)
with _dbConnectionFactory.Create() so the handler can be unit tested without a
live SQL Server. Parameter setup switched to generic IDbDataParameter creation
to match the IDbConnection abstraction.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ests

…tests

…p NpgsqlConnection cast, expand tests

The Wave 1 refactor cast IDbConnection to NpgsqlConnection, which made
the handler untestable (NpgsqlConnection is sealed). Re-refactored to
operate on IDbConnection directly using generic DbType (Int64 for time
fields, AnsiString for JobName) and IDbCommand.CreateParameter().
Added 3 missing Handle() lifecycle tests.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Phase 3 results: 41 new unit tests across 8 new test files for relational
transport job handlers (SqlServer/PostgreSQL/SQLite). Refactored 2 production
handlers (SqlServer + PostgreSQL SetJobLastKnownEvent) to inject
IDbConnectionFactory. Mid-flight re-refactor of PostgreSQL handler to drop
sealed NpgsqlConnection cast.

Added 2 CLAUDE.md lessons:
- Casting IDbConnection to sealed transport types breaks NSubstitute
- IDbConnectionFactory injection is the correct test seam

LiteDb + Redis deferred to a future phase due to concrete LiteDB types
and Redis BaseLua sealed multiplexer issues.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…LiteDb+Redis

- Mark Phases 1-3 as COMPLETE with accurate scope and outcomes
- Insert NEW Phase 4 for the deferred LiteDb + Redis transport handlers
- Renumber old Phase 4 (Dashboard.Api) to Phase 5
- Add cumulative outcomes section

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Test seam for Lua handler subclasses to override the Redis interaction
without mocking sealed StackExchange.Redis types.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…erface

Changes constructor to take IQueueCreation (already implemented by
RedisQueueCreation) instead of concrete sealed RedisQueueCreation.
Adds Guard.NotNull. Enables wrapper unit tests via NSubstitute.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…dler test coverage

Added 6 Handle() branch tests (message not found, empty queue, happy path,
multi-row isolation, null payloads, format exception). Lifts coverage from
40.9% baseline to ~100% of executable Handle() lines.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Phase 4 (LiteDb + Redis transport job handler tests) -- post-build pipeline:
- Verification: PASS (full-solution Debug build clean; LiteDb.Tests 166/166;
  Redis.Tests 190/190; deferred scopes in 2.2/2.4 validated against existing
  integration test coverage)
- Security audit: CLEAN (no secrets, no new deps, seam refactors benign)
- Simplification: one minor finding on DashboardUpdateMessageBodyLuaTests
  disposed-connection test -- resolved via documentary comment acknowledging
  the test is a named-for-intent duplicate of the null-return path
- Documentation: three Lessons Learned added to CLAUDE.md (in-memory LiteDB
  pattern, BaseLua virtual TryExecute seam, LiteDbConnectionManager no-seam
  limitation)

State: phase 4 complete. Phase 5 (Dashboard.Api DashboardExtensions) is the
only remaining phase in the code-coverage milestone.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Phase 5 (Dashboard.Api DashboardExtensions coverage) -- planning:
- CONTEXT-5.md captured (balanced 60-70% target, researcher-decides test
  layer, feature branch workflow for PR merge)
- RESEARCH.md authored directly by orchestrator after the dispatched
  researcher agent stalled mid-work (documented subagent lockup pattern)
- Key finding: Decision 4 (delete dead overloads) is void -- all 3 public
  methods on DashboardExtensions have real callers. The IConfiguration
  overload has 0% coverage despite being the production Dashboard.Ui
  entrypoint; this is the highest-ROI test target
- Four plans produced across two waves:
  - Wave 1 (parallel unit tests): PLAN-1.1 IConfiguration overload + 5
    transport switch arms (~50 lines), PLAN-1.2 Swagger + ApiKey security
    (~38 lines), PLAN-1.3 CORS + AuthorizationPolicy (~17 lines)
  - Wave 2 (integration): PLAN-2.1 Swagger + CORS integration tests via
    DashboardTestServer (~12 lines)
- Self-critique fixed a wave-1 file conflict (1.2 and 1.3 both appended to
  same test file) and a wrong DashboardOptions property name in 1.1
- Stale Schyntax cleanup PLAN-1.1.md archived under
  plans/archived-schyntax-cleanup/ (git mv, history preserved)
- Projected coverage: 33.3% -> ~64% (within balanced band)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…tch tests

Adds DashboardExtensionsFromConfigurationTests.cs with 5 new test methods:
- Happy-path SQLite :memory: registration via IConfiguration overload
- Parameterized test over all 5 transport switch arms (SqlServer, PostgreSql,
  SQLite, LiteDb, Redis) using throwaway connection strings
- Unknown-transport negative test (ArgumentException)
- Missing-Transport and missing-ConnectionString defensive-guard tests

Uses Assert.ThrowsExactly<T> per MSTest 3.x convention. Asserts against
DashboardOptions.ConnectionRegistrations (internal, InternalsVisibleTo already
grants access from the test project).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ionConvention tests

Adds DashboardExtensionsCorsAndAuthTests.cs with 4 test methods:
- CORS policy registers when EnableCors=true with non-empty origins
- CORS policy does NOT register when CorsOrigins is empty (branch guard)
- DashboardAuthorizationConvention.Apply adds AuthorizeFilter to a dashboard
  controller (positive branch)
- DashboardAuthorizationConvention.Apply ignores controllers from other
  assemblies (negative branch)

Note: the DashboardExtensions-level branch "AuthorizationPolicy set -> convention
added" is deliberately not asserted at unit-test scope. Debugging showed the MVC
options pipeline in a bare ServiceCollection doesn't reliably surface the
Dashboard's conventions via IOptions<MvcOptions> — the filters are added but
the conventions are dropped somewhere in the pipeline. The convention class
itself is exercised directly via Apply() instead; end-to-end coverage of the
DashboardExtensions branch belongs in the Dashboard.Api.Integration.Tests
project (PLAN-2.1 or a follow-up).

PLAN-1.2 note: the PLAN-1.2 builder wrote DashboardExtensionsSwaggerTests.cs
before hitting a sandbox block on git commit, and the file was carried along
in the preceding PLAN-1.1 commit due to pre-staging. Both files are Wave 1
and both pass tests; per-plan atomicity is not strict here.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…integration tests

Adds SwaggerEndpointTests.cs with 3 test classes / 3 test methods exercising
the UseDotNetWorkQueueDashboard pipeline branches that unit tests cannot cover:

- SwaggerEndpointTests.SwaggerJson_ReturnsOk_WithValidOpenApiShape — spins up
  a dashboard with EnableSwagger=true and verifies /swagger/v1/swagger.json
  returns 200 with valid OpenAPI content, exercising the UseSwagger() branch
  on line 204 of DashboardExtensions.cs.

- CorsIntegrationTests.CorsRequest_ReturnsAllowOriginHeader_WhenOriginMatches
  — hits /api/v1/dashboard/health with an Origin header and asserts the
  Access-Control-Allow-Origin response header, exercising the UseCors branch
  on line 199.

- AuthorizationPolicyIntegrationTests.DashboardController_Unauthenticated_
  Returns_Unauthorized_When_AuthorizationPolicy_Set — registers a minimal
  Test auth scheme and a DashboardAdmin policy requiring an authenticated
  user, sets options.AuthorizationPolicy = "DashboardAdmin", hits
  /api/v1/dashboard/connections (an MVC controller covered by the Dashboard
  authorization convention), and asserts 401. This recovers the end-to-end
  coverage gap left by PLAN-1.3's scope pivot — it proves the branch guard
  in DashboardExtensions.cs:101-105 DOES add the convention and that the
  convention DOES apply AuthorizeFilter to dashboard controllers in a real
  ASP.NET Core pipeline.

Extends DashboardTestServer with an optional configureServices and
configureApp hook to allow authentication/authorization registration without
touching existing call sites (the 1-arg overload remains the default path
for the other 40+ integration tests).

Verification: 226/226 Memory/Sqlite/LiteDb/Swagger/Cors/Authorization/Health
tests pass on both net8.0 and net10.0. Zero regressions from the
DashboardTestServer change.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…CLAUDE.md lesson

Post-review cleanups from the phase 5 simplification + documentation gates:

- Split SwaggerEndpointTests.cs into three per-class files
  (SwaggerEndpointTests, CorsIntegrationTests, AuthorizationPolicyIntegrationTests)
  to match the one-class-per-file convention used across all 28 existing
  integration test files. NoAuthHandler moved into AuthorizationPolicyIntegrationTests.cs.
  Mechanical split, zero logic change; 3/3 tests pass on both net8.0 and net10.0.

- Add WHY comment to DashboardTestServer.CreateAsync documenting that
  configureApp runs after Dashboard middleware and before MapControllers
  (noted as a structural fragility in REVIEW-2.1).

- Append CLAUDE.md lesson: AddControllers(action) in a bare ServiceCollection
  silently drops MvcOptions.Conventions while filters propagate. Cost 4
  debugging iterations in PLAN-1.3; use WebApplication integration tests
  instead of bare ServiceCollection for any test that must assert an
  IControllerModelConvention was registered.

Retroactive post-build pipeline after /shipyard:resume. All 4 plans
(PLAN-1.1, 1.2, 1.3, 2.1) had been committed in earlier sessions but
the reviewer gate and later gates were missed when the session was
interrupted.

Gates run in this session:
- Reviewers (4/4 PASS, 0 critical): REVIEW-{1.1,1.2,1.3,2.1}.md
- Verifier (PASS): VERIFICATION.md — build 0 errors, unit tests
  216/216, integration tests 226/226 on Memory|Sqlite|LiteDb filter,
  no regressions, projected ~70% DashboardExtensions line coverage
- Auditor (CLEAN): AUDIT-5.md — no secrets, no new deps, no IaC, no
  OWASP findings
- Simplifier (MINOR): SIMPLIFICATION-5.md — 1 medium applied
  (split SwaggerEndpointTests), 2 low pre-existing deferred
- Documenter (MINOR): DOCUMENTATION-5.md — 1 CLAUDE.md lesson added
  (MvcOptions.Conventions) + 1 inline comment on DashboardTestServer

Phase 5 COMPLETE. All 5 phases of the code-coverage milestone
complete. Ready for /shipyard:ship.

…tone

Appends a dated section to .shipyard/LESSONS.md covering:
- What went well: balanced-budget plan shape, 4/4 PASS on first reviewer
  pass, Wave-2 scope extension absorbing PLAN-1.3's gap
- Surprises: MvcOptions.Conventions propagation quirk in bare
  ServiceCollection, clean resume after 9 session interruptions,
  retroactive reviewer gate viability
- Pitfalls: stop-and-pivot rule for 4th-iteration DI unit tests,
  one-class-per-file convention, STATE.json staleness after interruptions
- Process: researcher should declare wave-1-to-wave-2 scope contracts,
  artifacts-over-STATE.json reconciliation on resume, retroactive
  reviewer pattern as supported resume path