docs(architecture): record GameWindow automated closeout

Reconcile the architecture, roadmap, milestones, issue ledger, session instructions, and durable memory with the 1,622-line native shell, canonical soak evidence, connected framebuffer comparison, and sole remaining user visual gate.

Co-authored-by: Codex <codex@openai.com>

memory/project_gamewindow_decomposition.md

@@ -2,11 +2,12 @@
 
 ## Current state
 
-The behavior-preserving App ownership campaign is complete through Slice 7 and
-Slice 8 checkpoints A–H. `GameWindow.cs` moved from the 2026-07-21 baseline of
-15,723 lines / 278 fields / 205 methods to 3,689 lines / 162 fields / 37
-methods: 12,034 lines (76.5%) were removed without changing accepted gameplay
-or rendering behavior.
+All eight behavior-preserving App ownership slices and their automated closeout
+are complete. `GameWindow.cs` moved from the 2026-07-21 baseline of 15,723
+lines / 278 fields / 205 methods to a 1,622-line native
+composition/callback shell: 14,101 lines (89.7%) were removed without changing
+accepted gameplay or rendering behavior. Only the user's final connected
+visual matrix remains before the campaign is marked complete.
 
 | Slice | Ownership moved out | Closeout size |
 |---|---|---:|
@@ -21,6 +22,9 @@ or rendering behavior.
 | 8 F | sole gameplay action route, typed commands, retained-root callback lifetime | 4,057 / 198 / 54 |
 | 8 G | pre-window/runtime settings ownership, one startup snapshot, inert late target binding | 3,663 / 162 / 37 |
 | 8 H | render/UI/frame/portal resource roots plus checked retryable GL leaves | 3,689 / 162 / 37 |
+| 8 I | executable ordered startup composition and terminal session start | 1,945 / startup shell |
+| 8 J | retryable typed lifetime/shutdown owner | 1,625 / lifetime shell |
+| 8 K–L | acknowledged canonical soak capture and automated closeout | 1,622 / native shell |
 
 Slice 8 is an ordered checkpoint campaign rather than another feature-body
 move. A froze/pruned the shell boundary; B gave all nine native window callback
@@ -141,6 +145,23 @@ review rule live in `docs/architecture/code-structure.md`.
 
 ## Accepted verification
 
+- Final automated closeout: 293 focused Slice 8 tests pass; App Release passes
+  3,451 / 3 intentional skips; the complete Release suite passes 7,823 / 5.
+  The Release build retains only the 17 pre-existing test-project warnings
+  tracked by #228. Behavior/retail, architecture/ownership, and adversarial
+  failure reviews are clean.
+- Checkpoint K (`bca41487`, `6c5e0604`): two fresh-process 403-second
+  nine-stop soaks produce the exact ordered same-frame canonical checkpoints,
+  zero pending teardown/retirement/staging/warmup work, no unconfounded owner
+  growth, unchanged process/performance gates, and graceful exits. #232 is
+  closed.
+- Checkpoint L connected lifecycle/reconnect passes in 314.4 seconds with six
+  valid PNGs and graceful capped/uncapped exits. The uncapped final checkpoint
+  is 174.74 FPS / 5.72 ms. Frames reconstructed from exact Slice 7 commit
+  `9d7df1bf` preserve deterministic geometry, UI/paperdoll layering, private
+  viewports, depth/alpha, and world reveal; remaining differences are live
+  entities, particles, vitals, and sub-frame camera timing.
+
 - Slice 8 Checkpoint H: 61 focused ownership/lifetime tests pass; App Release
   passes 3,236 tests / 3 intentional skips. Four corrected-diff cycles closed
   every architecture, retail, and adversarial finding; all three final reviews
@@ -177,11 +198,10 @@ review rule live in `docs/architecture/code-structure.md`.
   all movement/jump/combat exercises, and graceful exit passed; update p95 was
   at or below 0.8 ms. Caul return → plateau working/private delta was
   +85.7/+59.3 MiB, inside the unchanged gate.
-- Two immediately preceding identical-binary runs exceeded the coarse process-
-  residency threshold while deterministic ownership/lifecycle/timing checks
-  passed. #232 adds canonical resource-owner snapshots; do not solve this by
-  loosening the process-memory threshold or treating residency alone as an
-  identified owner leak.
+- Two Slice-7 identical-binary runs exceeded the coarse process-residency
+  threshold while deterministic ownership/lifecycle/timing checks passed.
+  Checkpoint K closed #232 by adding canonical owner snapshots; it did not
+  loosen the process threshold or treat residency alone as an identified leak.
 - The 18–33 missing VFX table/emitter diagnostics are known DAT-driven records,
   not a Slice 7 regression.
 - TS-50, TS-51, and TS-53 remain registered; Slice 7 introduced no new retail
@@ -189,13 +209,12 @@ review rule live in `docs/architecture/code-structure.md`.
 
 ## Next work
 
-Slice 8 checkpoints I–L group the remaining ordered startup composition,
-shutdown wiring, canonical soak snapshots, and closeout and reduce Silk callbacks to
-narrow calls into input, update, render, resize, focus, and shutdown owners.
-It removes leftover feature state only where Slices 1–7 already established a
-canonical owner; it does not redesign gameplay, rendering, or resource lifetime.
-The reviewed A–L execution ledger is
-`docs/plans/2026-07-22-gamewindow-slice-8-composition-lifecycle.md`.
+Run the final connected visual matrix: first login/radar; movement, mouse,
+resize, focus, and combat; shared inventory/skills/spellbook panel; outdoor,
+building, dungeon, portal, paperdoll, particle, and alpha presentation; then
+graceful close and fresh reconnect. After user confirmation, mark Checkpoint L
+and the campaign complete and resume M4. Full `GameEntity` aggregation remains
+a separate migration and must not be folded into the visual handoff.
 
 ## Do not retry
 
@@ -207,7 +226,9 @@ The reviewed A–L execution ledger is
 - Do not add a second GUID, session, spatial, or resource owner.
 - Do not fold full `GameEntity` aggregation into this campaign; evaluate it
   after Slice 8.
-- Do not loosen the soak process-memory threshold to hide #232. Add canonical
-  owner snapshots and preserve process residency as a secondary guard.
+- Do not revert Checkpoint K to exact cache equality: normal retirement may
+  shrink owners, and visible/server workload changes may change populations.
+  Hard-fail owner growth only when workload is stable, name workload-confounded
+  growth explicitly, and preserve the unchanged process-residency limits.
 - Small value/policy callbacks and diagnostics are acceptable only when they do
   not call a substantial window body or hide mutable ownership.