Widescreen / high-res opportunities per app

Status 2026-08-31. This is a backlog document: no app settings, config seeds, or byte pokes have been applied. The one emulator-side change that landed is the mode-enumeration widening (see baseline below). Everything else here is a verified recipe waiting for a decision to wire it up.

Every mechanism below was checked against our binaries (demo/shareware builds), not taken from the web as-is — community patches are written against retail and several claims did not transfer. Corrections are listed so the next session does not re-trust the same sources.

Emulator baseline (what a guest sees today)

The guest desktop is already arbitrary-resolution: get_screen_size packs the live canvas, nothing rounds to 4:3.
IDirectDraw::EnumDisplayModes (src/09a8-handlers-directx.wat) advertises 10 resolution slots — 640x480, 800x600, 1024x768, 1152x864, 1280x1024, the host canvas (rounded down to a multiple of 8, clamped 640x480..1920x1080), 320x200 (8bpp only, raw index 18 — Jazz2 cinematics), 1280x720, 1600x900, 1920x1080 — at 8/16/32bpp where applicable, 28 modes total.
EnumDisplaySettingsA/W (src/09a3-handlers-audio.wat) walk the same table densely (28 rows, FALSE past the end); ENUM_CURRENT_SETTINGS returns the canvas. Regression: test/test-display-mode-enumeration.js.
SetDisplayMode accepts any WxH and never refuses.
Exclusive presentation letterboxes, never stretches (lib/renderer.js _computeExclusiveTransform).

Open emulator issues that gate some recipes:

RCT crash above 800x600 — marker 0xCA002E20, guest RETs at 0x407165 into string data with ESP inside the threaded-code region. Our bug; it is why the RCT launchPrefs poke was removed 2026-08-26 (apps/rct.md:346-356, test/test-launch-prefs-resolution.js:66-80). Gates every RCT recipe.
GetDeviceCaps(HORZRES/VERTRES) ignores an active DirectDraw mode while GetSystemMetrics(SM_CXSCREEN) honors it (src/09a4-handlers-gdi.wat:1601 vs src/09a-handlers.wat:3475) — a game mixing the two sees two screens.
ChangeDisplaySettingsA returns success for a mode it never applies; an app that sets a mode and re-reads it through GDI sees the canvas.
launchPrefs is plumbed in both hosts (lib/app-profiles.js, host.js, lib/browser-shell.js:615, test/run.js:3624) but LAUNCH_PREFS = {} — there is currently no live poke to copy from.
DLL pokes are an implementation gap: launchPrefs patches the EXE after load_pe; poking a DLL (Diablo II's d2gfx.dll) needs the DLL's load base.

Corrections: retail claims vs our binaries

Web claim (retail)	Our demo/trial binary
TA registry `…\Total Annihilation`	`…\Total Annihilation Demo` (traced live)
TA `-screenwidth`/`-screenhight` args	absent — registry only
MW3 `…\MechWarrior 3\1.0`	`HKCU\SOFTWARE\MicroProse\MechWarrior 3 Demo\0.183`
JJ2 `VideoSize` REG_BINARY	subkey with DWORDs `Width`/`Height`/`BPP`
GTA2 keys under HKCU	`HKLM\SOFTWARE\DMA Design Ltd\GTA2\Screen`
AoE2 `Screen Width`/`Screen Height` regvals	not present in the trial (retail-AoC only)
AoE argv `NoStartup NormalMouse`	uppercase tokens; AoE1 trial has `NORMALMOUSE` but no `NOSTARTUP`

Per-game verdicts

NATIVE_CONFIG = engine takes a resolution from registry/ini/args and renders it. PORTABLE_POKE = documented hex technique re-derived against our binary. NOT_PORTABLE = community solution is an injected DLL / replacement engine / build-gated patch that cannot apply to our build.

Game	Native mechanism	Community patch	Verdict	Confidence
total_annihilation_demo	registry `DisplaymodeWidth`/`Height`, arbitrary WxH	v3.9.02 replacement exe (unneeded)	NATIVE_CONFIG	high (traced)
deus_ex_demo	`System/DeusEx.ini` `[WinDrv.WindowsClient]` viewport keys (lines 68-72; SoftDrv present)	—	NATIVE_CONFIG	high
quake2_demo	`sw_mode`/`gl_mode` index into fixed 10-entry table	—	PORTABLE_POKE (8 bytes)	high
rct	none; 3 hardcoded slots + `Game.cfg[0x17]` selector	ComputerLife22 slot swap; jeFF0Falltrades clamp patch	PORTABLE_POKE, gated on our crash	high
mw3	registry `InGameVMode` DWORD (+`*_SW` renderer pin)	Teleguy HUD assets	NATIVE_CONFIG	high key / med values
gta2_demo	`window_width`/`full_width`/`start_mode` (strings confirmed, HKLM)	GTAMP (offsets unpublished)	NATIVE_CONFIG	high
halflife_uplink	`HKCU\Software\Valve\…\Settings` `ScreenWidth`/`Height`; `-w`/`-h`	WON res patch	NATIVE_CONFIG	med-high
diablo2_demo	`-w`, `-res640`/`-res800`	5-site hex recipe (retail offsets)	PORTABLE_POKE — sites located in our build, but in `d2gfx.dll`	high
aoe1 / aoe2	bare-number argv token (`EMPIRES2.EXE NOSTARTUP 1280`); AoE1 ≤1024, AoE2 ≤1280	aoe2wide (MD5-gated, rebuilds `interfac.drs`)	NATIVE_CONFIG to stock modes	med (trial untested)
jazz2_demo	`VideoSize` subkey; fixed list, 640x480 max	JJ2+ = exe patch + guest `plus.dll` → 800x600	NATIVE_CONFIG (capped)	high
mcm	in-game menu, 800x600 ceiling	—	NATIVE_CONFIG (narrow)	med
captain_claw_demo	typed cheats `MPINCVID`/`MPDECVID` → up to 1280x1024	—	PORTABLE_POKE (keystrokes)	med
caesar3_demo	3 fixed modes; binary `c3.inf`, no registry	c3resPatcher (SHA-gated to retail 1.0.1.0)	NOT_PORTABLE to demo; re-derivable	high
heroes2_demo / heroes3_demo	none (F4 windowed = scaling only)	HD Mod = per-version patched exe + hook DLLs; refuses this lineage	NONE_EXISTS / NOT_PORTABLE	high
diablo_demo / diablo_shareware	none	every "D1 HD" replaces the engine (Belzebub = SDL2/GL)	NONE_EXISTS	high
starcraft_shareware	none	Resolution Expander = injected DLL, 1.16.1-only	NOT_PORTABLE	high
worms2_demo	none	wkReSolution = WormKit injected DLL, offsets unpublished	NOT_PORTABLE	high
fallout_demo	none	Mash Hi-Res patch rejects `FALLDEMO.EXE` by name	NOT_PORTABLE	high
abedemo	none — always 640x480	DDhack only upscales output	NOT_PORTABLE	high

Fixed-surface 2D titles not listed (Baldur's Gate demos, Icewind Dale, Civ2, Liquid War, Cave Story, …) get widescreen only as letterbox/crop on our side of the boundary. Windowed GDI apps are trivially fine on any desktop size.

Ranked recipes (when we decide to wire them)

Total Annihilation — zero patching. Traced end to end: queries DisplaymodeWidth/Height under HKCU\Software\Cavedog Entertainment\Total Annihilation Demo, gets not-found, writes back 640/480. Recipe: startupRegistry DWORDs. Menus stay 640x480 by design — validate on a gameplay frame, not a shell --png.
Deus Ex — two-line ini edit. WindowedViewportX/Y in the DeusEx.ini we already mount (lib/apps.js:716-728); mount a modified copy via vfsPath exactly as Quake II's config.cfg is mounted. Cost: UE1 software rasterizer at wide resolutions; watch vert- FOV.
Quake II — fully specified 8-byte poke. vid_modes[] in our exe at VA 0x0044d418 (raw 0x4b018), 10 × 16 bytes {char* desc, int w, int h, int mode}. Mode 9's dimensions live at 0x0044d4ac / 0x0044d4b0; poke them and add +set sw_mode 9 (or gl_mode) to args. ref_soft renders 8bpp indexed at any resolution.
RCT — derived, but sequence matters. All three published patterns occur exactly once in our unpacked RCT.exe:
- mode-3 slot: 0x40137c push 768, 0x401381 push 1024, 0x401395 mov [ebp-4],1024, 0x40139c mov [ebp-0xc],768
- width clamp: 0x401456 cmp 1280, 0x401463 mov 1280
- height clamp: 0x40146a cmp 1024, 0x401477 mov 1024 Patching alone does not help: this is the same fullscreen path that crashes us (0xCA002E20). The promising experiment is the windowed route — Game.cfg[0x17]=00 plus the clamp patch — which sidesteps SetDisplayMode and possibly the whole crash.
MechWarrior 3 — one DWORD. Seed InGameVMode under HKCU\SOFTWARE\MicroProse\MechWarrior 3 Demo\0.183 and re-trace; our 6000-batch run only observed the open→not-found, so read-back is unconfirmed. The *_SW values also pin the software renderer.

Cheapest experiments outside the top five: AoE1/AoE2 are an args change only (EMPIRES2.EXE NOSTARTUP 1280). Diablo II turned out easy to re-derive: d2gfx.dll (Shareware v1.04) has two clean if/else selectors at 0x100044ac and 0x10004a4c, four immediates each (width → [ebp-8], height → eax then global 0x1001c528); there is no d2client.dll in this build (client code is in the exe). Blocked on the launchPrefs DLL-base gap above.

Structural notes

We are the mode enumerator. GTA2 (start_mode), Half-Life fullscreen and DirectDraw surface creation resolve resolution through our EnumDisplayModes. For index/list-driven games, widening the advertised table (done) beats poking the guest.
Caesar III is genuinely not poke-able. Its three modes are ~9 independent inline .text constants per dimension and there is no mode table (.data has zero adjacent 640/480 or 800/600 dword pairs). That constant-folding is why the community wrote the Julius engine instead of a hex patch. By contrast HoMM2 and HoMM3 each have exactly one such pair (0x004e91d8, 0x005e937c) — a lead worth an hour, though the community verdict for both is that only scaling exists, no playfield widening.

Sources

id-Software/Quake-2 win32/vid_dll.c · SlashWork d2widescreenhack.md · SGD2FreeRes (mir-diablo-ii-tools) · jeFF0Falltrades/rct_patch · c3resPatcher (gitlab dr_afdch) · bvschaik/julius settings.c · withmorten/aoe2wide · MS KB Q175347 (AoE1) / Q242227 (AoE2) · PCGamingWiki + WSGF: Total Annihilation (WSGF via mirrors — 403 to fetchers) · DxWnd MW3 thread (sourceforge) · Jazz2Online JCF t-19223 · jj2.plus · StepS/wkReSolution · Resolution Expander (moddb) · beyondunrealwiki deusex-ini · GTAMP forum t=244 · Vault-Tec Labs Fallout 1 Hi-Res Patch · PCGamingWiki: Abe's Oddysee, Claw