refactor(arch): put module rules beside the code they describe

[mforce]

Refs #859

Follow-up to #1085. Owner decision (2026-10-05): a module rule lives next to the code it describes, and stays central only when it describes a relationship. Behaviour does not change. The analyzer and every architecture test give the same results as on main.

Change map

Read the commits in order.

Commit	What to check
refactor(arch): mark module contract types with [ModuleContract]	160 one-line [ModuleContract("<Owner>")] marks, placed by a codemod. RealModuleLedger collects them from Domain, Application and Infrastructure by reflection. The owner rows lose Contract. A mark naming no owner is a registry error. Inherited = false keeps nested subclasses of a marked record (FlockNameResolution.Found and its siblings) out of the contract. Global using of Cluckwork.Domain.Common.Architecture (Platform) in Domain and Application.
refactor(arch): one rules class per module beside its edges	ModuleOwners.cs and ModuleEdges.cs are deleted. Modules/<Owner>.cs holds internal static class <Owner>ModuleRules { } with that owner's [ModuleOwner] and its outgoing [ModuleEdge] cells (text moved verbatim). The analyzer walks Domain's types instead of reading assembly attributes. RealModuleLedger fixes the order: owners by design 3.4 (matrix layout), edges by from, then to (the old order). A cell on another owner's class is a registry error. The CW1001, CW1002 and CW1003 messages and the test's undeclared-edge message name Modules/<from>.cs and print [ModuleEdge(...)].
docs(arch): ...	New section in 859-typed-rule-registries.md. src/AGENTS.md paragraphs and one line in 514 updated.

Mechanical noise: the marks (about 100 files, one line each) and the moved row text.

Names. <Owner>ModuleRules avoids the existing facade classes AccessModule, FarmModule and the others. The classes live in Cluckwork.Domain.Common.Architecture, so no new namespace segment is added (#985). A repo-wide grep found no *ModuleRules type before this change.

Not done, by design. The analyzer reads no contract today, on main or here. It reads only namespaces, exact namespaces, claimed types and edges. A removed [ModuleContract] is therefore red only in the tests. An analyzer contract rule would be new behaviour, so it belongs in a separate decision. Contract lists are now ordinal-sorted. The hand order is gone, and no test reads it. No contract type sits in an assembly the tests cannot see.

Proofs

Check	Result
Parity: temporary JSON dump test, run on main and on each head	Owners, kinds, namespaces, exact namespaces, implementations, seams, claimed types and contract sets are identical. All 22 edges are identical, in order. Checked at commit 1 and at the final head. The dump test is not committed.
Analyzer edges vs ModuleLedgerRealTreeTests' LiveEdges (temporary CW1099 dump, dotnet build src/Cluckwork.Api --no-incremental)	73 and 73, identical
Coupling matrix, CLUCKWORK_REGENERATE_MATRIX=1	no content change
dotnet build Cluckwork.sln	clean, analyzers on
Application.Tests count (--list-tests)	base 725, head 725
Architecture classes, one at a time	22/22 green, analyzer tests included
Documentation 20/20, ImagePin 2/2	green

Mutations

Each mutation was applied, then restored.

Mutation	dotnet build src/Cluckwork.Api	Existing test, built with -p:RunAnalyzers=false
[ModuleContract] removed from IFlockLookup	green (the analyzer has no contract rule, as on main)	PeerContractRealTreeTests and AdapterReachRealTreeTests red: contract bypass
[ModuleContract("Farm")] on IFlockRepository (a Flocks type)	green, as on main	AdapterReachRealTreeTests and PeerContractRealTreeTests red: not in a namespace 'Farm' owns
[ModuleContract("Fram")] (no such owner)	green	ModuleLedgerRealTreeTests red: names owner 'Fram', which has no ModuleOwner row
FlockManagement -> Farm cell deleted	red, CW1001 naming Modules/FlockManagement.cs	ModuleLedgerRealTreeTests red: undeclared edge, same file
UpdateEggGradeHandler added to EggOperations -> Farm	red, CW1002 naming Modules/EggOperations.cs	ModuleLedgerRealTreeTests red: stale ledger row
FlockManagement -> Farm cell moved onto FarmModuleRules	green	ModuleLedgerRealTreeTests red: cell sits on the wrong class
Finance -> Farm cell on a class with no [ModuleOwner]	not run	ModuleLedgerRealTreeTests red: sits on OrphanRules

Not verified

The full test suite was not run (host rule: filtered runs only); CI is the authority. Live IDE squiggles were not observed.

[mforce]

No P0–P3 findings at 11cd28a1be0704f51992a8231323ef492ef6dd3a, reviewed against main at ebf8200765c6f7e44b90ca083fb2b6798b68a9ee. The slice preserves behavior and the existing tests' authority.

I independently compiled the base commit's ModuleOwners.cs and ModuleEdges.cs rows in a temporary test and compared them with head's reflection output. This did not rely on the implementer's saved dumps.

Check	Result
Owner parity	All nine owners match, including kinds, namespaces, exact namespaces, implementations, seams, claimed types and contract sets.
Edge parity	All 22 cells match in order, including reasons and symbol order.
Contract marks	Exactly 160 explicit marks; every owner/type pair matches the base lists, with none missing or extra. Nested record subclasses do not inherit the mark.
Ordering	The nine owners have unique fixed ranks; valid edge pairs and contracts sort ordinally. Contract consumers use membership or their own sorting. Matrix regeneration produces zero diff.
Analyzer parity	A fresh diagnostic dump and the real-tree scanner produce identical sets of 73 edges.
Shadowing and behavior	No new namespace segment; all rule classes use the existing architecture namespace. Outside the architecture mechanism, changes are marks, global usings and comments. No executable product changes.

Reproduced mutations, each restored:

Mutation	Analyzer build	Existing tests with analyzers disabled
Remove IFlockLookup's mark	Passes	Peer and adapter contract guards fail with contract bypasses.
Delete FlockManagement → Farm	CW1001	Ledger guard fails with the undeclared edge.
Move that edge onto FarmModuleRules	Passes	Ledger guard reports the wrong rules class.
Move Finance → Farm onto ownerless OrphanRules	Not run	Ledger guard reports the ownerless class. This isolated probe introduces no duplicate cell.
Add UpdateEggGradeHandler to EggOperations → Farm	CW1002	Ledger guard reports the stale symbol.

CW1001 points to Modules/FlockManagement.cs and prints a syntactically pasteable [ModuleEdge(...)], with the reason left for the author. CW1002 correctly points to Modules/EggOperations.cs and asks to remove the stale symbol.

The analyzer never read contract lists on main and still does not. Test-only enforcement of removed marks matches #1085's stated scope: edge and namespace diagnostics, with architecture tests retaining authority. The decision docs explain this and introduce no copied issue-state claims. Two wording clarifications fall below P3: RealModuleLedger's introductory comment should limit “the analyzer reads too” to owner/edge rows; the decision record's diagnostic sentence should distinguish CW1001's insertion attribute from CW1002's deletion guidance.

Applied pstack:thermo-nuclear-code-quality-review and ponytail-review, slice-scoped through P3. No structural regression or warranted simplification within the accepted design. The collector is 79 lines; module files are 17–55 lines; no changed file crosses 1,000 lines. Ponytail verdict: Lean already. Ship.

Verification: solution build passes with zero warnings; all 377 architecture tests pass across 22 separately run classes; all 20 documentation tests pass. Every test command used DOTNET_USE_POLLING_FILE_WATCHER=1. After restoring the mutations, the solution build and ledger/analyzer classes pass again, and the checkout is clean. Full integration tests and live IDE diagnostics were not run.

Reproduction scripts, independent dumps, individual logs and detailed review notes are retained locally in /home/mforce/.cluckwork-slices/reviews/1086-11cd28a1-astra-evidence/. No product changes, commits or pushes.