docs(audit): align validation claims with current evidence

The reconstruction ledger still described an obsolete function count, claimed
local Windows/macOS cross-target checks despite those standard libraries being
absent, and attributed lock, claw, drain, and repeated-launch coverage to the
current autoplay trace. Those statements were stronger than the current
artifacts prove after the corrected trajectory.

Record the final C ledger counts, distinguish configured native CI from local
Linux production validation, add both stopped-Wine slingshot probes, and route
claw/panel/lock/drain evidence to the deterministic and focused tests that
actually cover it.

Test Plan:
- `rumdl check --flavor commonmark RECONSTRUCTION.md` -- passed
- `git diff --cached --check` -- passed

tdkpin-rs/RECONSTRUCTION.md

@@ -10,10 +10,10 @@ a9022f1894e3e6e21fc42e8f6c932f7c549ca77f63aaa0c488bb9d55d9d0174c  TDKPIN.EXE
 f1d9ac980c7bfba5dc53eaa9e7cb2c3cd9b82f879ee8962ad40bf863d641bb49  MMTIMER.DLL
 ```
 
-The Ghidra recovery covers all 301 named NE exports plus 62 internal function
-entries. Its raw output and coverage tables remain beside the binary, so later
-work can be checked against the same evidence rather than against this Rust
-implementation.
+The reconstruction ledger verifies all 391 ordinary in-image functions, six
+shared/overlapping entries, and all 301 NE entry points. Its raw output and
+coverage tables remain beside the binary, so later work can be checked against
+the same evidence rather than against this Rust implementation.
 
 ## Coverage by subsystem
 
@@ -50,10 +50,10 @@ decoded, build-ready subset; it does not replace that evidence archive.
 
 - Static coverage: all intended decoded visual/audio resources are preserved;
   the recovered static collision layout is represented in Rust.
-- Build coverage: `cargo check`, unit tests, and strict Clippy complete on the
-  host. Cross-target checks pass for `x86_64-pc-windows-gnu` and
-  `x86_64-apple-darwin`; CI is configured to build and test natively on Linux,
-  macOS, and Windows.
+- Build coverage: the production build, unit tests, and strict Clippy complete
+  on the Linux host. This checkout does not have the Windows or macOS Rust
+  standard-library targets installed; the repository CI is configured to run
+  check, tests, and Clippy natively on Linux, macOS, and Windows.
 - Runtime coverage: the Linux executable was launched through the real window
   backend, the attract screen was inspected, a game was started, a ball was
   launched, collision scoring was observed, and a rendered frame was captured.
@@ -63,16 +63,21 @@ decoded, build-ready subset; it does not replace that evidence archive.
   controlled terminal-18 claw capture and release. A current one-substep
   record-2 probe proves special-respawn publication `(17000,23000)/(0,3040)`
   followed by the same response's `(16698,21191)/(-302,-1809)` continuation.
-  Rust tests assert these exact live states.
+  Stopped one-substep probes cover both relative slingshot chains: record 57
+  produces `(77369,356597)/(1839,1832)` with spin `-5.28`, and record 72
+  produces `(233253,358494)/(-2095,1669)` with spin `+5.46`. Rust and C tests
+  assert these exact live states.
 - Deterministic mechanics coverage: named launcher, flipper, and all four claw
   terminal scenarios are driven by a 120 Hz validation clock while production
   physics accumulates the recovered 100 Hz substep. Seeded runs use the original
   Borland random stream. Traces include the last collision-object id and can
   export the logical 640x460 render target.
-- End-to-end gameplay coverage: seeded autoplay charges each launcher ball and
-  operates both flippers from ball position. The two-minute acceptance run
-  covers repeated launches, both flippers, targets, bumpers, lock holes, a claw
-  capture/release pair, and drains while checking every state for finite values.
+- End-to-end gameplay coverage: seeded autoplay charges and launches a ball,
+  operates both flippers from ball position, and covers target and bumper hits
+  for two minutes while checking every state for finite values. Separate named
+  deterministic scenarios cover all claw terminals, launcher/flipper edges,
+  target rotation, the full panel, and effect-seven multiball; focused game
+  tests cover drains and lock-hole completion.
 - Semantic boundary: collision, random, spin, fixed-point state, detail callback
   batching, and flipper publication now use the reconstructed Borland/Win16
   timing model. Native texture drawing replaces Win16 GDI calls while preserving