docs: list the kitchen's balance follow-ups

The kitchen ships before its balance pass, and the design doc did not say
what that pass still owes.

Follow-ups from the 2026-10-29 nightly (200 runs a bot):
- Normal default wins 14.5% (target 15 to 25%)
- Normal SWAP=1 wins 48.5% (target 30 to 45%)
- Hard wins 3.5% (target 5 to 12%)
- Nightmare wins 0% (target 1 to 4%)
- Portrait Normal wins 9% and portrait Easy 40% (target as landscape;
  the gap predates the kitchen)
- Normal close calls 65% of wins (target 30 to 50%)

Author: Tanner-Scadden

GAME_DESIGN.md

@@ -380,7 +380,7 @@ Built on 2026-10-29, the first room of "Exploring the manor". Proposed, waiting
 - **Slippery soup.** A Groaner, Hotfoot, Kaboomer, or Big Leaguer that steps onto soup from dry floor skids on and lies dazed for 0.9 s, with a splash and a thud. One that gets up in the soup walks out of it, so nothing slips forever. Big Lugs, Ankle Biters, and Uncle Chunk keep their feet. The player slides a little: moving on soup builds a skid that carries on for a moment.
 - **No off-screen hints.** The door preview shows a marker only for a door whose spot is on screen. A kitchen door out of view counts at the kitchen doorway instead, when the doorway is on screen, since that is where its zombies come into the parlour. Nothing points at a door above or below the window. The portrait sideways pins for parlour doors stay as before.
 - **The bot and the model.** The balance bot buys the kitchen once it holds the Chatterbox at its second upgrade, both Plank Piles, and a Grumpy Gnome, while it keeps its reserve; it walks to the boards and bumps them. `KITCHEN=0` is the bot that never opens it. The analytical model buys it at the same point in its list and counts it in a full kit. The bot still holds the parlour, so it uses the ovens and the soup only when the kitchen's zombies come to it.
-- **Balance, measured 2026-10-29** in the nightly after the polish pass (200 campaign runs a bot, with wave 6 eased to a health scale of 1 and a smaller horde): Normal default 14.5% wins (target 15 to 25%), SWAP=1 48.5% (30 to 45%), Easy 59% (55 to 70%), Hard 3.5%, Nightmare 0%, portrait Normal 9% and portrait Easy 40%. Normal deaths by wave: 0, 0, 0, 2, 4.5, 20, 19.5, 5.5, 6, and 27.5% on the finale, the deadliest again. The median win takes 10.3 minutes, so the win target moved from 7 to 9 minutes to 9 to 12. The baseline was rewritten from this run. Still to do: close calls (65%, target 30 to 50%), Hard, Nightmare, and portrait, whose gap to landscape predates the kitchen.
+- **Balance, measured 2026-10-29** in the nightly after the polish pass (200 campaign runs a bot, with wave 6 eased to a health scale of 1 and a smaller horde): Normal default 14.5% wins (target 15 to 25%), SWAP=1 48.5% (30 to 45%), Easy 59% (55 to 70%), Hard 3.5%, Nightmare 0%, portrait Normal 9% and portrait Easy 40%. Normal deaths by wave: 0, 0, 0, 2, 4.5, 20, 19.5, 5.5, 6, and 27.5% on the finale, the deadliest again. The median win takes 10.3 minutes, so the win target moved from 7 to 9 minutes to 9 to 12. The baseline was rewritten from this run. Shipped as it stands on 2026-10-29 at the owner's request, with the balance pass left for later. Follow-ups for the kitchen balance pass: Normal default wins 14.5% (target 15 to 25%), SWAP=1 48.5% (30 to 45%), Hard 3.5% (5 to 12%), Nightmare 0% (1 to 4%), portrait Normal 9% and portrait Easy 40% (the gap to landscape predates the kitchen), and close calls 65% (30 to 50%).
 - **Cost.** Zombies off the camera's view, and any outside the house, are not drawn. The room is still one baked image per state, both baked as the game loads, and the ovens add a few rectangles a frame and one flame image while one belches. `src/perf.scene.test.ts` has a kitchen scenario: 696 shapes built and 554 steady paint operations, under the 740 and 590 budgets.
 
 ### Portrait play