[electricessence]

States the NULL policy and the async transaction token behaviour correctly, and fixes a wrong sentence about a struct's own parameterless constructor, so newcomers read facts that match the code.

What changed

• NULL policy: the full rules live in mapping.md "NULL in a non-nullable column"; the table row, the matching and required bullets, the single-value bullet and the Dapper comparison now link to it instead of restating it. It gains the single-value case: Map<int>().Single() over a NULL is 0, Map<int?>() and Map<string>() give null. coming-from-dapper.md says Dapper's QuerySingle<int> over a NULL throws. That page's link also pointed to "Types and NULL" (a section about value conversion that says nothing about NULL); it now points to "NULL in a non-nullable column".
• Async transaction token: foundation.md "Transactions" says the token is checked after the action returns (a cancelled token throws and nothing commits) and the commit is sent without the token. New async sample transaction_async.
• Struct constructor: mapping.md said a struct's own public parameterless constructor is not called; it is (new T()). The sentence is fixed, and the test AStructWithAPublicParameterlessConstructor_IsBuiltWithIt now asserts a side effect only the constructor can produce, not only Id == 7.
• Constructor rule, two lines: a record with a second public constructor is in the "two or more public constructors" row (unless the second is parameterless); an all-optional constructor does not count as parameterless.
• Excluding a member: one line saying new(nameof(X.Member), null) needs no null!.
• Not changed: whats-new-11.md (the release PR writes the notes), the three pages on reading output values back, anything under Source/.

Verification

• Samples (new checks in Program.cs; check-only models in ConstructorRules.cs; a commit-token recorder in Transactions.cs): dotnet run --project docsite/samples -c Release ends with All 125 sample checks passed.
• Checks added: single-value NULL (int 0, int? and string null); Dapper QuerySingle<int> over NULL throws; the exclusion pair leaves the member unmapped while a column of that name exists and the others map; a record with a second constructor and an all-optional constructor are rejected; a record with a parameterless second constructor uses it; the async transaction commits; a token cancelled in the action throws OperationCanceledException and rolls back; the commit receives a token that cannot be cancelled although a live token was passed.
• The struct claim is asserted by the Core test, run with dotnet test Tests/Open.Database.Extensions.Core.Tests -c Release --filter UnmappableTypeTests: 213 passed on net10.0 and on net11.0.
• DocFX build: Build succeeded, 0 warnings, 0 errors. quick-check.ps1: Release build 0 warnings, 0 errors, format style and analyzers clean.

Not done / decisions

• The "A missing column and a NULL" table in mapping.md and the existing NULL text in whats-new-11.md still restate parts of the NULL rule; left alone on purpose (a different cut of the rule, and the release PR owns the notes).
• The "Dapper throws" claim is checked against Dapper 2.1.89 on SQLite only.
• The commit-token claim is shown with a recording transaction wrapper, since SQLite cannot observe the token itself.

🤖 Generated with Claude Code