[iancooper]

Summary

Spec 018 phase 4, S2: Commands, Handlers and Pipelines, and Get Started (ten pages). It also includes the two sweeps you ruled on today, D19 and D20. This changes the published site, so it needs your sign-off.

• 106 of S2's 107 FAILED blocks now build, and so does one fence that was split, so the ten pages hold 115 C# blocks, 109 of them BUILT. The 2 that stay FAILED are same-page and listed; 4 are SKIPPED V9 forms. Ten scaffold units are new or extended.
• Rulings:
  • D13: Polly.RateLimiting is pinned, so each pin has 553 assemblies.
  • D14: FeatureSwitches.md now reads IConfiguration.
  • D15: ShowMeTheCode.md follows the WebAPI sample.
  • D17: MongoDBOutbox.md's handler was rewritten and run against MongoDB.
  • D16 and D18 were filed upstream as Brighter#4609 and #4616.
• Running the blocks overturned three claims (ledger 37–39). AgreementDispatcher.md routed a Command to several handlers, which Send refuses. Two [UsePolicy] attributes on one method are CS0579. new PolicyRegistry() fails at startup without the RETRYPOLICY and CIRCUITBREAKER policies.
• D19, the prose sweep:
  • Every defect the docs dated "at 10.7.0" was run at both versions, with 10.7.0 as the control: #4414, #4335, #4437, Replay, #4432, #4510, #4453/#4454, #4443/#4444 and messaging.system.
  • Every one held at 10.7.0 and is fixed at 10.8.0. 21 pages (#4414 among them) now state 10.8.0's behaviour and keep the 10.7.0 workaround, labelled as such.
  • The runs also found Monitoring.md's setup never registered IAmAControlBusSenderAsync, so [MonitorAsync] failed at both versions, and Telemetry.md's flag table lacked CircuitBreaker.
• D20, the relational DDL: six blocks are now 10.8.0's GetDDL output, and each was created on its own database: CausationId is absent at 10.7.0 and present at 10.8.0. Three of them (MSSQL, MySQL, SQLite) had never been builder output at all, and the SQLite one was not valid SQL.
• Probes: every run is a spec/018-compile_residual/probe/<group>/<name>/ project against the released packages, with its case and its control (tasks.md § Behaviour (4.4), Second pass (4.5), Sweep (4.5a), The relational DDL (4.5b)).

Gates

Gate	Result
blockcheck --report	994 blocks: 536 BUILT, 441 FAILED, 17 SKIPPED — 0 findings, exit 0. 553 assemblies on each pin, 60 units, 0 violations
pagelint	0 errors / 346 warnings (was 410); --changed strict, 0 errors; --plant 5 / 5
symbolcheck	0 — 24 entries, 14 silenced
linkcheck · shape · redirects · --verify · versioncheck · optioncheck	unmoved: 168 / 0 · 161 · 77 · 161 = 161 · 0 of 18 · 0 / 59 / 529

Every gate is inside the band task 4.1 predicted except pagelint, which is 1 above at 346. Eleven BUILT blocks build on their page's scaffold alone, so no repair touched them and none gained its usings (tasks.md § Phase 4 closed).

For you

1. Sign-off: this changes the published site.

   • S2's ten pages.
   • The D19 pages: the nine inbox pages, BrighterInboxSupport.md, SchedulingAMessage.md, InMemoryScheduler.md, FAQ.md, DispatcherConfigurationReference.md, Compression.md, Telemetry.md, ConfiguringOpenTelemetry.md, Monitoring.md, SweeperCircuitBreaking.md, PostgreSQLMessageBroker.md and RelationalDatabaseConfigurationReference.md.
   • The D20 pages: five Outbox pages and SpannerInbox.md.
   • Off-group repairs: AgreementDispatcherRouting.md, CloudEventsSupport.md, HowConfiguringTheCommandProcessorWorks.md, EFCoreOutbox.md, MongoDBOutbox.md and BrighterOutboxSupport.md.

   May I merge?

2. After the merge, may I delete the head ref spec/018-phase4?

🤖 Generated with Claude Code

https://claude.ai/code/session_01LjnH2pJ98aMabu1MyTpRxy