docs: 013 phase 1 — repair an API surface the corpus never had

[iancooper]

P0-2 and P0-4. Eleven pages, ten code blocks, and every gate where design §9 said it would be.

Gate	Predicted	Measured
linkcheck.py	unmoved	160 ✓
pagelint.py pages	unmoved	158 ✓
pagelint.py warnings	must fall from 779	773 ✓
--changed scope	8 strict	10 strict — see finding E
shape · redirects · versioncheck · optioncheck	unmoved	unmoved ✓

What is now zero across contents/: .ResiliencePipelines(, .ConfigureResiliencePipelines(, InputChannelFactory, DispatchBuilder.With(, .Subscribers(, and all eight dead relational type names. .Resilience(, DefaultResilience and AddBrighterDefault were on 0 pages and are now on 5 — the tell that opened this phase, inverted.

Every repaired block was compiled before it was written to a page: one project per block, <ImplicitUsings>disable</ImplicitUsings>, optioncheck's pinned packages, 0 errors and 0 warnings each.

Seven findings the approved task list did not predict

A. Two of the seventeen sites are correct, and design §5's table would have falsified them. MigratingToPollyV8.md:99 and :101 sit inside a block the page labels V9. At tag 9.33, git grep -c declares With() 1 and Policies( 2; both are 0 at 10.7.0. They are genuine V9 API, correctly shown as history on a migration page — rewriting them would have told readers V9 had a method it never had. A repair inventing a history, the same shape as design §11 Q6's invented type name. The block now carries CLAUDE.md's ❌/✅ markers, so the next dead-API sweep meets an explicit label rather than a bare .With().

B. Every fluent example was missing three mandatory chain steps. CommandProcessorBuilder's interfaces force ExternalBus|NoExternalBus, ConfigureInstrumentation|NoInstrumentation and RequestSchedulerFactory between Resilience and Build. Four of five builder blocks skipped two or three. Repairing only the names design §5 lists would have produced blocks that still do not compile. No grep could see it — the compiler reads a chain, a sweep reads a token.

C. AddBrighterDefault lives in Paramore.Brighter.Extensions — a namespace on 0 of 158 pages. The method the spec set out to add is one the corpus cannot import.

D. UseResiliencePipelineAsync is in Paramore.Brighter.Policies.Attributes, and the two pages already using it carry no using lines at all.

E. Two of P0-4's ten sites are inside fences, not prose — so ten blocks turned strict where eight were budgeted. --changed said so immediately, which is the argument for reading the scope line rather than the verdict.

F. Renaming a type inside a block is not finished until its namespace is imported. BrighterBasicConfiguration.md had Paramore.Brighter.MySql but not ...MySql.EntityFrameworkCore. It passed rule 6 either way — directives, but not the right ones.

G. BrighterOptions.PolicyRegistry raises CS0618 and is NOT a defect — under AC10's three states it is live, merely deprecated. Recorded, not swept.

Checked and deliberately left alone

• Five Darker calls on CQRSWithBrighterAndDarker.md, verified real in ../Darker against a control of 0. Only the Brighter half of that block was touched — obligation 10.
• DynamoDbUnitOfWork and MongoDbUnitOfWork survive on five pages — the NoSQL stores genuinely have a unit of work, and they are the control proving the P0-4 sweep discriminated rather than pattern-matching on UnitOfWork.

Full ledger: spec/013-howto_guides/tasks.md § Phase 1 as executed.

🤖 Generated with Claude Code

https://claude.ai/code/session_0146UueHL6H3zGBGTYwz7GtL