feat: prove MVP capacity and make the Go/No-Go decision

PLANS.md

@@ -268,7 +268,7 @@ If all criteria pass, set this milestone to `DONE`, append its execution record,
 
 ## Milestone 004 — Prove MVP capacity and make the Go/No-Go decision
 
-**Status:** `READY`
+**Status:** `IN PROGRESS`
 **Depends on:** Milestone 003
 
 ### Objective
@@ -298,6 +298,14 @@ Before running the capacity test, define numerical limits based on the board spe
 
 Do not adjust a threshold after seeing the result unless the execution record contains an explicit justification.
 
+### Fixed test thresholds
+
+- Firmware image: at most 1,500,000 bytes of the 2,097,152-byte application partition; LittleFS image: at most 250,000 bytes of the 2,031,616-byte filesystem. These reserve room for the game engine and complete offline interface.
+- Minimum free heap: at least 96,000 bytes throughout the run. Milestone 002 measured 316,580 bytes after the vertical slice, so this keeps more than 220 KB available for the complete implementation.
+- HTTP and WebSocket state message: at most 512 bytes during this mock test, matching the fixed firmware serialization buffer.
+- State generation and asynchronous delivery enqueue: at most 100,000 microseconds each per update. This leaves substantial margin below the 2-second polling interval.
+- Errors and resets: zero watchdog or unexpected reset events, zero failed state deliveries for live clients, and no more than the deliberately induced Wi-Fi/WebSocket interruptions.
+
 ### Acceptance criteria
 
 - Two players and eight spectators simultaneously receive the state intended for their roles.