Spec 017 phase 5: tranche 2b and the recorded falsehoods — 295 blocks build, both targets met

[iancooper]

Summary

Phase 5 of spec 017 repairs the 16 tranche 2b pages, which are the remaining pages with one hard block each, and fixes the recorded falsehoods (P0-7): ITimerProvider, the unshown Order, and the sync attributes on async handlers. Every tranche page is now whole: each C# block on it either builds against Brighter 10.7.0 / Darker 4.1.1, is skipped with an accepted reason, or is listed with its reason in tasks.md § Blocks that stay FAILED.

This PR changes the published site, so it needs your sign-off to merge. After the merge, may I delete the head ref spec/017-phase5?

git diff --name-only origin/master..HEAD -- contents lists 53 pages: 15 of the 16 tranche pages (AggregationQueryPatterns.md was made whole by its scaffold unit alone), and 38 pages outside the tranche, changed by P0-7, by repairing recurrences, and on your rulings.

Defects found by running the blocks, repaired everywhere they occur

Commands, pipelines and Darker

• BuildingAPipeline.md used .Successor =, but the method is SetSuccessor(). The page also said a manual chain runs from the registry alone; in fact a factory must return the wired instance.
• [UseResiliencePipeline] was stacked on one method (CS0579) on 5 pages. The attribute is not repeatable, so the strategies are now layered in one pipeline. PolicyRetryAndCircuitBreaker.md described Polly's strategy order backwards: the first strategy added is the outermost, so the example's 10 s timeout covered every retry together.
• Handlers declared without public never ran, because AutoFromAssemblies() scans only public types. They are now public on 7 pages (your ruling).
• Darker's query processor was said to default to Transient. It is Singleton, and resolves handlers from the root provider. Also corrected: Darker's handler scan rules, and Activator.CreateInstance as a factory.
• "Cannot use AutoFromAssemblies" with an Agreement Dispatcher was wrong on three pages. The scan works; this was run with a control.
• Monitoring.md is rewritten for V10 (your ruling). Two upstream defects found while running it are stated on the page.

Transports, external bus and observability

• PostgreSQLMessageBroker.md had 11 defects. For example:
  • A consumer whose channelName differed from the publication's Topic received nothing.
  • PublishAsync was said to send to the broker; it writes no row.
  • A delay was said to set visible_timeout.
  • [ClaimCheck] was shown with parameters it does not have.
  • A subscription without messagePumpType throws. This last defect recurred on non-generic subscriptions on 5 more pages.
• Tracing never recorded a Brighter span. AddSource(…) alone is not enough; AddBrighterInstrumentation() is needed. Telemetry.md is rewritten, including its span names, attributes and propagation headers as captured on the wire. ConfiguringOpenTelemetry.md now points the OTLP exporter at Jaeger instead of using the deprecated Jaeger exporter.
• CloudEvents header names and the content mode are now as captured on the wire. Extension properties reach the message only through CloudEventJsonMessageMapper<>.
• S3LuggageStore.md said ACLs default to Private; they are null.
• The Control API page showed V9's status JSON and described subscription names as topics.

P0-7 and what stood beside it

• ITimerProvider does not exist: 4 → 0. The InMemory scheduler's time seam is InMemorySchedulerFactory.TimeProvider, and the fake-clock example is run with FakeTimeProvider. The page's Paramore.Brighter.InMemoryScheduler package does not exist either.
• Order is now shown in CQRSWithBrighterAndDarker.md. Showing it surfaced a CS0029 (Id assigned to a Guid key) and a CancellationToken passed where PublishAsync takes a RequestContext.
• Six sync attributes on HandleAsync → the async form. These pipelines throw ConfigurationException when built. attr_mismatch.py now finds exactly one hit, the deliberate Before (error) example at PipelineValidation.md:250, whose prose now says it is wrong. The published After (fixed) example on that page itself reported an error. It now reports nothing, run with ValidatePipelines().
• PublishAsync swept wherever it meant the bus (your ruling). It dispatches to handlers in this process only.
• The Avro mapper in DefaultMessageMappers.md is rewritten against Confluent's API (your ruling). This is the one pin change besides FakeTimeProvider: 98 → 100 packages, each measured alone, and no verdict moved.
• Sync and async take different mappers. ReactorAndProactor.md said mappers have no async variants. MessageMappers.md now shows IAmAMessageMapperAsync<T> and which call uses which. A pump or a producer call given only the other kind of mapper silently uses the default mapper instead; all four producer calls and both pumps were run.
• V10MigrationGuide.md § 4 built a registry Brighter never reads, and § 5 listed IRequestContext members that do not exist. InstrumentationOptions was said to be "added in 10.7.0"; it ships after 10.7.0 and is now marked as not yet released. Replay validation is marked the same way on three pages.
• UsingTheContextBag.md's examples now post with the context they fill. Bag keys act only on the context passed to the call that makes the message (run, with a control).

Every block that asserts behaviour was run against the released packages, with a control. The tables are in tasks.md § Phase 5 as executed.

Upstream issues, filed in this phase on your word: BrighterCommand/Brighter#4453 and #4454 (monitoring), #4458 (AWS CloudEvents souce), and #4465 (Control API wrong-case 500). Each is stated on its page.

Gates

#	Gate	Result
1	linkcheck	165 files, 0 broken
2	pagelint	0 errors; warnings 616 → 537. 35 fewer are on the tranche pages, inside the predicted band; 19 on the P0-7 pages; 25 on 13 other pages, from recurrences and rulings
3/4/7	shape / redirects / --verify	unchanged: 161 / 77 / 161
5	versioncheck	0 stale of 18
6	optioncheck	0 mismatches, 59 tables, 519 rows
8	symbolcheck	0 findings, 22 entries, 3 silenced
9	blockcheck	990 blocks: 295 BUILT (234 before), 678 FAILED, 17 SKIPPED; baseline 295; 45 units, 0 violations; 545 reference assemblies

• pagelint --changed origin/master exits 0.
• The AC2 diff against c7329bb prints 207 lines. 194 are FAILED -> BUILT; the other 13 are the accepted skip, or insertions and splits recorded in § Splits. The diff was checked with controls in both directions.
• Both spec targets are met. BUILT is 295 against a target of ≥ 250. Pages with nothing BUILT are 41 against a target of ≤ 60, and 57 before this phase.
• BUILT is 1 above 5.1's predicted ceiling of 294, and the pin grew against a prediction of no change. Both are reconciled in § Task 5.6.

🤖 Generated with Claude Code

https://claude.ai/code/session_016Am29wrQUS81NVh8dUyNuy