acdream/docs/plans/2026-07-22-gamewindow-slice-8-checkpoint-l-closeout.md

# GameWindow Slice 8 Checkpoint L — campaign closeout

**Status:** Automated closeout complete 2026-07-22; final user visual matrix
pending.
**Parent:**
[`2026-07-22-gamewindow-slice-8-composition-lifecycle.md`](2026-07-22-gamewindow-slice-8-composition-lifecycle.md),
Checkpoint L.
**Integrated baseline:** `6c5e0604`; Checkpoints A–K are implemented.
`GameWindow.cs` is 1,622 raw lines, App Release passes 3,451 tests / 3
intentional skips, and the complete Release suite passes 7,823 tests / 5
intentional skips. Two fresh-process canonical nine-stop soaks pass with
graceful exits and the unchanged residency/performance limits.
**Behavior rule:** L proves and records the shipped ownership state. It may
correct documentation, test/gate coverage, or a defect demonstrated by the
audit, but it does not add gameplay, alter retail behavior, loosen a gate, or
perform the separately deferred `GameEntity` aggregation.

## 1. Outcome

Close the eight-slice thin-`GameWindow` campaign only after every exit
criterion has current evidence:

- the host is a native construction/callback shell with no AC gameplay
  algorithm, entity scan, packet builder, DAT landblock builder,
  animation-part composer, or draw-graph body;
- update, render, input, session, framebuffer, and shutdown callbacks are
  narrow typed handoffs;
- startup and shutdown have one ordered, transactional, retryable ownership
  graph without a service locator or callback facade into the host;
- extracted owners have focused tests and symmetric teardown;
- the Release build/test, connected lifecycle/reconnect, canonical soak, and
  framebuffer gates preserve the accepted behavior;
- architecture, roadmap, milestones, issues, divergence bookkeeping, durable
  memory, and both session instruction files agree with the code.

Line count is reported as a structural signal, not substituted for those
ownership requirements.

## 2. Completion audit

### 2.1 Host boundary and dependency direction

Measure raw lines, fields, methods, and the remaining callback bodies from the
actual source. Run the Slice 8 boundary/source-shape suites and inspect every
remaining `GameWindow` method. Search for prohibited network packet assembly,
entity/world scans, DAT build loops, part-transform composition, render graph
bodies, direct environment reads, and anonymous device subscriptions. Any
substantial body or callback into the host reopens the responsible checkpoint;
it is not waived because the file is below the target size.

Verify the composed owners and their tests cover:

- ordered partial-load rollback and terminal session start;
- exact frame-pair publication and one update/render handoff;
- logical callback cutoff before session retirement and physical detach;
- hard session/GPU shutdown barriers, soft physical-cleanup reporting,
  retry/no-replay progress, and native-window-last release;
- deferred checkpoint acknowledgement, same-frame post-diagnostics capture,
  FIFO order, cancellation, and shutdown.

### 2.2 Three corrected-diff review passes

Review the complete Slice 8 diff and final source in three independent passes:

1. **Behavior/retail boundary:** no input, session, update/render, UI, camera,
   settings, teleport, or gameplay behavior was silently redesigned; every
   retained adaptation remains registered.
2. **Architecture/ownership:** no duplicate canonical state, broad runtime bag,
   window back-reference, leaked subscription, conflicting disposer, or layer
   inversion; all ownership transfer and teardown paths are symmetric.
3. **Adversarial failure:** partial acquisition at every phase, callback
   concurrency, repeated/reentrant close, failed detach, session/GPU barrier
   failure, deferred render failure, checkpoint cancellation, and process/GUID
   reuse cannot strand or replay work.

Confirmed findings are fixed at their owner and all affected gates are rerun.

## 3. Automated and connected gates

Run, in order:

1. focused Slice 8 composition, callback, settings, frame, lifetime,
   automation, and source-boundary tests;
2. App Release;
3. `dotnet build AcDream.slnx -c Release --no-restore`;
4. the complete Release suite;
5. `tools/run-connected-world-lifecycle-gate.ps1 -SkipBuild`, requiring the
   capped six-stop lifecycle, exact semantic checkpoints, five capped PNGs,
   code-zero authoritative graceful close, and immediate uncapped fresh-process
   reconnect with its sixth PNG;
6. consume the two Checkpoint-K fresh-process nine-stop reports, requiring
   exact nine-name/sequence/process identity, zero pending teardown/retirement,
   staging and warmup work, no unconfounded canonical owner growth, unchanged
   process/update/allocation limits, and graceful exit;
7. compare the six current lifecycle/reconnect PNGs with the accepted Slice 7
   `9d7df1bf` baseline. If the old generated artifacts are absent, reconstruct
   them from that exact commit in an isolated disposable worktree rather than
   inventing a new baseline. Accept only live weather/particle timing,
   authoritative vitals, and sub-frame camera-settling differences; geometry,
   UI/paperdoll layering, private viewports, alpha/depth, and world reveal must
   remain equivalent.

Generated logs and the isolated baseline worktree are evidence only and remain
untracked. The user's unrelated `TransitionTypes.cs` edit remains untouched.

## 4. Reconciliation and handoff

After all nonvisual evidence is green:

- mark K and the automated portion of L in the parent ledger;
- close #232 with both canonical soak reports and preserve #225, #153, #116,
  #228, TS-50/TS-51/TS-53, and other carried work at their true scope;
- update `code-structure.md`, the architecture, roadmap, milestones,
  `AGENTS.md`, `CLAUDE.md`, and `memory/project_gamewindow_decomposition.md`
  with the same counts, commits, gates, next-work boundary, and do-not-retry
  lessons;
- audit the retail divergence register and add/remove no row unless runtime
  behavior actually changed;
- commit the evidence/doc closeout as a bisectable unit and rerun final
  source/status checks.

The sole remaining pause is the user's connected visual matrix: first login
and radar; movement/mouse/resize/focus/combat; the shared inventory/skills/
spellbook panel; outdoor/building/dungeon/portal/paperdoll/particle/alpha
presentation; graceful close and fresh reconnect. Only that confirmation marks
L and the overall structural campaign complete.

## 5. Automated closeout evidence

- The final host is 1,622 raw lines. Source-boundary inspection finds no AC
  algorithm, packet builder, entity scan, render graph, animation-part
  composer, or stored `GameWindow` back-reference in an extracted owner.
- All 293 focused Slice 8 ownership, composition, callback, frame, lifetime,
  automation, and source-boundary tests pass. App Release passes 3,451 tests / 3
  intentional skips; the complete Release suite passes 7,823 / 5. The build's
  17 test-project warnings are the unchanged #228 set.
- The connected lifecycle/reconnect gate passed in 314.4 seconds with graceful
  capped and uncapped exits. Its uncapped reconnect reached 174.74 FPS / 5.72
  ms at the final checkpoint. All six PNG artifacts are valid; the only route
  warning is the expected 25 world-edge empty-landblock misses.
- Both Checkpoint-K nine-stop reports passed in about 403 seconds, produced the
  exact names/sequences/process identity, and found no unconfounded canonical
  owner growth under the unchanged residency/performance limits.
- The six current lifecycle PNGs were compared against artifacts rebuilt from
  exact accepted Slice 7 commit `9d7df1bf` in an isolated worktree. The five
  deterministic destinations preserve geometry, UI/paperdoll layering,
  private viewports, depth/alpha, and world reveal. Facility Hub was nearly
  pixel-identical (mean absolute RGB delta 0.91/255); the Aerlinthe captures
  were 3.42 and 3.87/255. Holtburg and reconnect differences are live
  entities, particles, vitals, and sub-frame camera timing. Capped login used a
  different server-saved location and was checked only for complete rendering
  and presentation layering.
- Three corrected-diff reviews are clean: behavior/retail boundary,
  architecture/ownership, and adversarial lifecycle/failure handling. The
  divergence audit found no new runtime behavior in K or L and requires no
  register change.

The generated reports, screenshots, and disposable baseline worktree are
untracked evidence. The user's unrelated `TransitionTypes.cs` edit was not
modified or staged.