Reverse Engineering: Gundam VS. Series - Gundam vs. Zeta Gundamre-gvs-gvzg
RESEARCH NOTEBOOK Local edition

Findings and lessons

Chapter 093 min readResearch record

At a glance

  • Question: What did the experiments establish, and what did the workflow teach us?
  • Result: Recompilation, reverse engineering and runtime tests solved different parts of the problem. Rendering and multiplayer prototypes now have evidence from limited tests.
  • Still open: Current-phone throughput, complete game flows and end-to-end multiplayer remain unverified.

Recompilation made the original game executable in a new environment. Reverse engineering made specific behavior understandable enough to change. The rendering and multiplayer prototypes have evidence from limited tests.

The tools answer different questions

ps2-web-recomp translates guest instructions into C++ and supplies the runtime behavior needed to execute them as native code or WebAssembly. The generated code is not the developer's original source, and translation does not automatically identify physics, animation or network semantics.

Ghidra with the PS2 extension helped recover those meanings from the original executable. The AI-assisted analysis workflow also supported browser experiments, although its decompilation path did not accept the R5900 target. The runnable translation, static analysis and runtime observations remain complementary. Keeping all three avoids treating plausible pseudocode as a behavioral proof or a passing frame counter as an explanation.

What the evidence supports

Swipe or scroll horizontally to read all columns.

Area Supported by retained evidence Further scope
🟡 Browser port Executable game paths, ordinary controls and tested save persistence Every mode and a portable clean setup for every reader
✅ Above-60 rendering PoC Distinct native intermediate images. About 120 renders and OS display updates/s, with 60 gameplay updates/s, in bounded PC combat. More game modes and sustained current-build phone use; further pacing refinement is optional
🟡 Phone Earlier saved-battle Auto capture: about 78 computed and 72 browser-observed updates/s at normal speed Latest full-game high-refresh throughput or a fixed main-menu slowdown
🟡 Streaming Implemented two-controller/media transport and a user's two-device observation A retained full match with separate host and two clients, measured latency and recovery
🟡 Independent clients Actor ownership, bounded deterministic trials and 2,640-tick browser agreement with delayed disc delivery Full selection-to-rematch integration, adaptive cross-rate agreement or restored original online service

The chapter links above explain each result. Evidence conventions define the labels; the evidence index lists records and runnable exports.

Lessons from the workflow

The useful loop was to reproduce a symptom, trace its instructions, make a narrow change, compare controlled behavior, then test ordinary interaction. The failed Start press showed why the final step matters. A successful throughput benchmark still missed a real input defect.

  • Separate the metrics. Computed renders, gameplay ticks and browser-shown updates measure different things. See the rendering goal.
  • Keep results with their conditions. An older phone capture and a bounded PC test cannot establish current-phone performance. See adaptive results and limits.
  • Keep the multiplayer routes distinct. Streaming shares one runtime. Independent clients must agree on gameplay state. Their integration and validation needs differ.

See Further work for the proposed checks and integrations.