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

Evidence conventions

Project reference

Each result names its test scope. A fixture, a native test, a browser session and a physical-phone capture answer different questions.

Result labels

Swipe or scroll horizontally to read all columns.

Label Meaning
✅ Verified The stated, bounded check has supporting code or measurements. Read the scope beside the label.
🟡 Partially verified Some parts work or have evidence; the described user journey still has missing checks.
⬜ Untested A proposal or acceptance target with no retained test result.

These labels apply to a claim, not to the entire game. The evidence index also identifies the support type: measured, reported, inspected or proposed.

The historical game records were reviewed on 2026-10-25. The initial 2026-10-26 revision checked source code, exported visuals and reran the portable input regression. A later session on October 26 UTC, ending October 27 local time, added new PC game tests, desktop video and aligned OS presentation traces. It did not rerun phone or multiplayer tests. Chapter-specific conditions remain beside each result.

What the records mean

The JSON files in docs/evidence/ contain selected metrics and provenance. Historical source paths are relative to the original ps2-web-recomp working tree. The new PC display record names its separate private test archive. A provenance path identifies an archive file; it is not a download link. The reproduction inventory lists exported material and remaining gaps.

Each curated record states its extraction or review method and limitations. Where a number is copied from an earlier written analysis rather than recalculated from raw samples, it is identified as a reported historical measurement. A source hash refers to the source bytes; a build ID, WASM hash and report commit are different identities.

Raw captures can include screenshots, browser state, device characteristics, connection details and personal paths. Only selected measurements and explanatory material belong here. Compression does not remove personal information. Complete logs, memory cards, disc images, home-directory paths and operational credentials stay outside this repository.

How to interpret frame rates

Swipe or scroll horizontally to read all columns.

Metric What it can show What it cannot establish alone
Gameplay updates per second Progress of the game's canonical simulation Smooth display presentation
Render updates per second Completed guest render iterations Distinct frames scanned out by the screen
Shown updates per second Browser callbacks observing new game output A photometric measurement of physical refresh
OS display updates per second PresentMon display events for the selected game swapchain Unique image content or the panel's pixel response
Browser callback frequency Current scheduling cadence A permanent panel capability
Game speed Guest-time progress per wall-clock time Low input or streaming latency
PCM frames per second Audio production rate Audible equivalence, absence of every glitch

An early saved-fight phone capture, a later full-game phone capture and the latest PC build are separate experiments. Their rates are not a matched device comparison.

The historical PC capture calibrated browser scheduling near 60 Hz; it did not measure the panel's maximum capability. The new session confirmed three active 180-Hz displays and approximately 180 idle browser callbacks/s. Its matched trace supports about 120 OS display updates/s, with normal-speed gameplay. Neither session tested 144-Hz rendering.

The accepted PresentMon 2.6.0 runs used QPC timestamps and one selected GPU-process swapchain per run. Display rates use the first-to-last displayed-event interval. Counter rates use the browser's before/after interval. A boundary-frame difference between these counts is not automatically a lost frame. The selected rows and endpoint counters are exported, while process IDs, swapchain addresses and absolute timestamps remain private.

The desktop video is a separate showcase. Its variable-rate capture did not maintain the requested 180 FPS, and encoding corrected some timestamps. The 60- and 120-fps viewing copies can duplicate or drop source frames. Use the OS trace for timing and the native repeat/interpolate comparison for distinct intermediate-image evidence.

Repeatable game-code tests

A useful acceptance record identifies the matched build, requested rate, effective mode, internal/display size, physical device and refresh mode, duration, exact measurement window, game speed, frame counters, input sequence, audio production/consumption and failures. Record menu, loading, pause, resume, results and re-entry, not only steady combat.

For performance tests, use zero output gain while keeping the AudioContext and ring-buffer consumption active. Suspending audio changes the workload. Do not alter system-wide volume. For multiplayer, keep both endpoint records, the input sequence, guest tick/checkpoint comparisons and any intentionally injected transport or disc delay.

AI assistance

AI agents assisted code investigation, test preparation and writing. Claims in this report rest on retained code, byte checks, measurements and observed behavior; an agent's explanation alone is not experimental evidence. Agent, model and provider names are intentionally omitted.

Checks on this report

During assembly, 69 unique original-source SHA-256 fingerprints matched the retained files.

Reader validation records the Wiki build and interface checks. Those checks do not add evidence about game performance.