Spec 017 phase 6: acceptance — 42 of 42, AC8 and AC11 repaired on your ruling, 299 blocks build

[iancooper]

Summary

Phase 6 of spec 017 is acceptance. Every criterion has been walked by its command or by the reader it names. The changed pages have been checked for widening, and both ledgers are complete. The spec closes at 42 of 42, with the residual sentence that 018 starts from.

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-phase6?

Phase 6 was planned to touch no page. At 6.1, the named readers found AC8's second half and AC11 unmet. You ruled on 2026-09-28 that both be repaired in this PR. As a result, git diff --name-only origin/master..HEAD -- contents lists 24 pages.

For your review: AC8 and AC11 re-read

Both walk rows in tasks.md § The acceptance walk still carry 6.1's verdict of unmet. The repairs are in § AC8 and AC11, repaired under the ruling. Re-reading them is this PR's review.

AC8. Three blocks were marked // ... where the using could have been written. They now carry their usings: QueryPipeline.md #8, PolicyFallback.md #3 and HowConfiguringTheCommandProcessorWorks.md #6 (plus #5 beside it). Once PolicyFallback.md #3's attribute resolved, it exposed a CS7036: [FallbackPolicy(backstop: true, step: 1)] omits the required circuitBreaker. All four lines on the page with that defect are repaired.

AC11. All thirty blocks that lacked a run with a control have now been run against Brighter 10.7.0 and Darker 4.1.1. Postgres, MySQL, DynamoDB-local and LocalStack ran in Docker. The runs found these claims false or partly false:

• Darker step order. The lowest step is outermost. Logging at step 1 does not see retries, and QueryPipeline.md Fix typo in title of faq.rst #16's "won't log individual retries" says the reverse of the truth. The prose beside sphinx-ify "Usage Documentation" #1 also recommended the wrong fallback/retry layout.
• PolicyFallback.md / HandlerFailure.md. Fallback receives the original exception. BrokenCircuitException appears only on later calls. PolicyFallback.md Update documentation #10's Polly fallback did not compile (CS0029, CS1929) and has been rewritten so that the service call runs through GetPipeline<Product>.
• FeatureSwitches.md. With no registry, Config is On. A registry that is missing the handler throws ConfigurationException. On the in-memory transport, redelivery happens at once, not after a visibility timeout.
• InMemoryOutbox.md Update ImplementingAHandler.rst #2. PostAsync deposits the message and dispatches it at once.
• V10MigrationGuide.md Update BuildingAPipeline.rst #12. The page's CreateCopy omitted Policies.
• Scheduler docs. "Falls back to immediate requeue if no scheduler configured" is false, because DI always registers InMemorySchedulerFactory.
• Kafka/MessageMappers schema-registry header. The header encoding changes only .Value. The damage comes from a text-column outbox, and binaryMessagePayload: true is the repair.
• Telemetry.md. If tracing is left unset, it behaves as None. With no AddProducers, there is no Command Processor span at all.

Outside the runs: 15 blocks on 10 pages registered a typeof(...) transaction provider without the configuration it is built from. Each block now registers it. Several MongoDB and EF Core V9 leftovers were repaired too. As a result, 4 MongoDB blocks now build.

Open, yours

• Whether to file the CommandProcessorBuilder finding upstream (not yet ruled). In 10.7.0, Build() gives the Command Processor no tracer when there is no external bus (:304). This was run both ways and is stated on both pages.
• No .accepted file yet. That is yours to add.

Gates

#	Gate	Result
1	linkcheck	165 files, 0 broken
2	pagelint	0 errors; warnings 537 → 524
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: 299 BUILT (295 before), 674 FAILED, 17 SKIPPED; baseline 299; 45 units, 0 violations

pagelint --changed origin/master exits 0. attr_mismatch.py reads 1, the deliberate PipelineValidation.md:250.

What 017 shipped

	016's close (c7329bb)	Now
BUILT	101	299
Pages with a FAILED block / with nothing BUILT	140 / 97	88 / 39
pagelint using debt	743 blocks, 115 pages	524 blocks, 66 pages
Scaffold units / pages mapped	14 / 15	45 / 67

Over the whole spec, 116 pages changed. The defect ledger has 116 rows, each at 0 after repair. Friction entries 67–80 were added. For D4, you ruled that attr_mismatch becomes pagelint rule 8, to be built in 018.

The residual, the line 018 starts from: 674 of 990 C# blocks still do not compile against the released packages. Of those, 630 are on the 64 pages that 017's tranches never reached: 314 name a pinned type they never import, 145 do not parse, and 113 need a type their page never shows. So 018 first builds pagelint rule 8 and gives --classify its second compile, then tranches those 64 pages by what that instrument finds.

🤖 Generated with Claude Code