diff --git a/docs/mutation-testing.md b/docs/mutation-testing.md
index 8db67be5..a8671a40 100644
--- a/docs/mutation-testing.md
+++ b/docs/mutation-testing.md
@@ -3,7 +3,7 @@
 CultLib measures its C# test suites with Stryker.NET, pinned as a repo-local
 tool in `dotnet-tools.json`. Coverage is added project by project; a project
 with no `stryker-config.json` next to its test project has not been scoped
-yet (see "Follow-ups" below for what's outstanding).
+yet (see "Scope and follow-ups" below for what's outstanding).
 
 ## Running it
 
@@ -126,14 +126,17 @@ project as precedent for the `vstest` caveat itself - that caveat is
 `mtp`-runner-specific and documented above for `static readonly` field
 initializers, and GameCult.Geometry has never actually hit it.
 
-## Follow-ups
+## Scope and follow-ups
 
-CultMath.Tests and GameCult.Geometry.Tests are the only projects currently
-scoped. GameCult.Geometry's `stryker-config.json` has no `test-runner`
-entry: the project is NUnit 4 with `NUnit3TestAdapter` and
+CultMath.Tests, GameCult.Geometry.Tests and GameCult.Caching.Tests are the
+projects currently scoped. GameCult.Geometry's `stryker-config.json` has no
+`test-runner` entry: the project is NUnit 4 with `NUnit3TestAdapter` and
 `Microsoft.NET.Test.Sdk`, so Stryker's default `vstest` runner applies
 directly (no `mtp` workaround needed, unlike CultMath.Tests's xunit.v3
-suite).
+suite). GameCult.Caching.Tests is the same shape (NUnit, default `vstest`)
+and its `stryker-config.json` is driven the same way, by the repo-local
+`dotnet-stryker` tool: `dotnet tool restore` once, then `dotnet tool run
+dotnet-stryker` from `tests/GameCult.Caching.Tests`.
 
 GameCult.Geometry has only ever been run `--since:`-scoped to the cut that
 added surface nets; a full-project baseline (running mutation against all of
@@ -151,7 +154,76 @@ file, anything under `Fixtures/`) loses `--since` line-level scoping for that
 run; a future cut whose diff is C#-only, against a base commit that already
 has `stryker-config.json`, should scope correctly.
 
-`GameCult.Caching`, `GameCult.Mesh` (+ `.Quic`, `.Quic.Native`),
+### GameCult.Caching.Tests
+
+`tests/GameCult.Caching.Tests/stryker-config.json` mutates all of
+`src/GameCult.Caching` (the project is `GameCult.Caching.csproj`; Standard
+mutation level, `break` 0, the same thresholds as Geometry). Caching was
+scoped before Networking because its suite is deterministic and quick (1415
+tests, about 40 s) over about 7.8k lines of source, where GameCult.Networking
+is about 20k lines and its tests lean on socket timing, so a mutant timing out
+there says little about the test. A full Caching run is about 2,200 mutants at
+roughly 1.6 s each, about an hour; run it detached and read the JSON report,
+never as one attached session.
+
+The first measurement was a narrow pass over the three files that carry the
+store-replace and element-id machinery, from `tests/GameCult.Caching.Tests`:
+
+```bash
+dotnet tool run dotnet-stryker \
+  -m '**/src/GameCult.Caching/StoreReplace.cs' \
+  -m '**/src/GameCult.Caching/CultStoreLinks.cs' \
+  -m '**/src/GameCult.Caching/CultElementIds.cs'
+```
+
+It tested 243 mutants and scored 84.91% in about 8.5 minutes. Two Stryker
+runs overlapped during that pass, and it showed: the same tests that kill a
+mutant by hand reported it `Survived` or `Timeout` (for example the `Holds`
+guards at `CultElementIds.cs` 311 and 316), which inflated that file's survivor
+list to 114. Never run two Stryker runs against one checkout at once, and check
+`ps` before starting one. Re-run on `CultElementIds.cs` alone, after the tests
+below, it tested 228 mutants: 208 `Killed`, 8 `Timeout`, 12 `Survived`, a score
+of 94.74%, and every one of the 12 is an equivalent mutant.
+
+Real gaps that run exposed, each now pinned by `ElementIdWalkTests` (and each
+failing when its mutant is applied by hand): the order of the id path (the
+codec refuses at the first type MessagePack cannot allow, so order is what an
+error names first; `Append` to `Prepend` at 163, 167 and 180); the
+union-cycle guard in `ConcreteTypes` (248; the mutant is a stack overflow, so
+the red is a crashed test host); a list of values reporting that it holds an
+id (`Holds`, 322); `[Key]` read through an override (`KeyOf`, 583); and the
+`IsWrittenWhole` and `IsKeyed` predicates (629 to 631 and 636: formatter,
+`[MessagePackObject]` and `KeyAsPropertyName` inherited from a base, and the
+`[DataContract]` without `[MessagePackObject]` and `||` arms).
+
+Equivalent survivors in `CultElementIds.cs`, each hand-checked against the full
+suite; none can be told apart by any test:
+
+- 142, 251, 636 (second arm), 506 (`IsDefined(UnionAttribute, false)` to
+  `true`): `UnionAttribute` is `Inherited = false`, so its `inherit` argument
+  changes nothing. Likewise 630's `DataContractAttribute` `inherit` argument
+  (`DataContractAttribute` is `Inherited = false`).
+- 264, 542: `group.First()` to `FirstOrDefault()` over `GroupBy` groups, which
+  are never empty.
+- 491: `new StringBuilder(bytes.Length * 2)` to `/ 2` changes only the initial
+  capacity.
+- 321: `!elements` to `elements` in `Holds`; `elements` implies the inner
+  type classifies as an owner, so `elements && Leaf` is never true and the
+  early return is a pure shortcut (the loop below returns the same `false`).
+- 578: removing the `Nullable.GetUnderlyingType` unwrap in `WalksAsOwner`;
+  `Classify` unwraps nullables itself, so the unwrap is dead. A cut
+  candidate, not a test gap.
+- 645, 657: `Append(type)` to `Prepend(type)` in `ElementTypeOf` and
+  `MapTypes`; `type` matches only when it is itself `IEnumerable<>`,
+  `IDictionary<,>` or `IReadOnlyDictionary<,>`, and none of those has a second
+  matching interface, so the position of `type` never decides the answer.
+
+The Windows-only P/Invoke group (the native-library and `RuntimeInformation`
+paths, such as `StoreReplace.cs` line 45) can only be exercised on Windows. It
+is a target for a Windows-leg mutation run, not a Linux coverage gap, and its
+`NoCoverage` mutants on Linux are not findings.
+
+`GameCult.Mesh` (+ `.Quic`, `.Quic.Native`),
 `GameCult.Networking` (+ `.WebSockets`), and the non-.NET runtimes
 (`cultcache-rs/py/ts`, `cultmesh-*`, `cultnet-*`) each still need their own
 `stryker-config.json` (or ecosystem-equivalent tool: `cargo-mutants` for
diff --git a/tests/GameCult.Caching.Tests/stryker-config.json b/tests/GameCult.Caching.Tests/stryker-config.json
new file mode 100644
index 00000000..ac366203
--- /dev/null
+++ b/tests/GameCult.Caching.Tests/stryker-config.json
@@ -0,0 +1,11 @@
+{
+  "stryker-config": {
+    "project": "GameCult.Caching.csproj",
+    "mutate": [
+      "**/src/GameCult.Caching/**/*.cs"
+    ],
+    "mutation-level": "Standard",
+    "reporters": [ "progress", "html", "json" ],
+    "thresholds": { "high": 80, "low": 60, "break": 0 }
+  }
+}