Refuted claims

Technical appendix · Home · Français

Claims that were believed and then disproved, kept so they aren't re-adopted. Section A comes from the full re-check on 2026-11-03.

Standing rules

A negative from a limited search pattern isn't evidence of absence. State the limit.
A text scan only rules out the encoding it tested. Every "no dialogue" scan below missed 5-bit coding.

A. Refuted by the re-check

Earlier claim	Result
INT 0FEh handler is missing or external	It is `maup.exe` EA `0x10628` (`0x53:0xF8`).
Segment `0x53` is a separate far API	`0x53:0` is `_main`, `0x53:6` is `Game_Main`, `0x53:0xF8` the handler.
`maup.exe` is overlaid	No overlay evidence; 24 ordinary relocated segments.
Text is fixed-pitch with `+0x66` unused	`Vid_DrawGlyph` adds `+0x66`; the `+0x6A` nudge is temporary.
The atlas reads as little-endian in the file	Wrong for odd byte offsets: 47 of 99 glyphs.
Index-to-character map unknown	Solved; see Text and fonts.
Dialogue is in no file	It is in `G3.CON` records `0x28`/`0x29`, 5-bit packed.
Output is in `0x400` chunks with input resync	The counter only increments; no effect on decoding.
`rec27_sub02` is a charset, 4,112 B when decoded	It's a raw image; its first u32 is palette and count bytes.
`Font_LoadStrings` loads strings	It decodes a picture for the intro script.
`.CON` records are vector geometry, no text	Many are tiled images; records `0x28`/`0x29` are text.
`wVidCanvas*` is the draw target	It's read as a blit source; draws go to `wPageSeg`.
`ES` writes and `DS` reads the canvas	Backwards for blits: `ES:DI` is the source.
`bGlyphDirectStore == 1` means direct store	1 selects `Vid_PutPixel`.
`word_1A4F4` is text origin; `wGlyphRowPitch` is pitch	Both hold the text colour.
`Vid_BlitTile`: `CH` rows, `CL` tiles	`CH` is width, `CL` is height.
`sub_10536` is a paragraph scan	It's a Sound Blaster port probe.
INT 0FEh site table	Four of six EAs were wrong; all six are file offset + `0xF800`.
Six service numbers	Six sites, five services.
Archive slot comes from the disk map	It comes from `wConRecordToArchive`.
Open uses `3D00h`	`3D80h`.
Sentinel list is 20 records	Six groups, with the record set of each archive.
Only nine files; nothing missing	More names exist, including conditional `G7.CON`.
Remap applies when the flag is set	It applies when `bSoundCard` is 0.
`0x15/0x16` both alias `0x17`	`0x15→0x17`, `0x16→0x14`.
`INSERT DISK` belongs to `maup.exe`	Only `INSTALLE.EXE` has that string.
`INSTALLE.EXE` is 37,376 B	12,684 B.
Last 4 KiB of `maup.exe` are zero	Only 1,458 of them are; there's also a 29-byte trailer.
`HNF.HUF` says `lanxhor`, `HEY`	`lankhor`, `URL`; ROT-13 was misapplied.
`HNF.HUF` tail is byte-identical to `SND.HUF`	Shared suffix is only 216 B.
Turbo C code here is BP-less	Standard `push bp / mov bp,sp` is common: 160 raw matches.
0 of 312 payloads pass `try_decompress`	Two `.CON` payloads pass, and all `.HUF` streams decode.
`unk_1CBE8`	The published offset DS:`27CB` is EA `0x1CBEB`.
Delphine Software or Loriciels made it	Lankhor: intro credits say so.
`sub_12722` is the idle worker, called from `0x1EFEB`	It's `Game_LoadAssets`, called once from `0x1F1EB`. `0x1EFEB` holds no call.
`SAUP.MAU` is a saved-game blob, out of scope	It's the starting game state, loaded into the block at DS:`166`.
Room records' sub 0 has an unknown format	Palettes, then a normal image at `+0x40`.
`MAUPITI1` only serves or parks as a TSR	It holds the game loop, resumed by `maup.exe` at `0x1EF4D`.
`MAUPITI1` "record/arena engine" with a separate resident role	It loads records through services; `maup.exe` does the file work.

B. Still valid

Claim	Evidence
`.HUF` containers aren't ROT-13	They decode from raw bytes.
No static Huffman table exists	The tree is adaptive.
The game doesn't print text through INT 10h	The character routines are used by the adapter/CLI path (unverified in full).
`Vid_ArenaAlloc` isn't a fatal-error routine	Verified.
`0x7D00` is a word count for a 64,000 B page copy	Verified at `0x16FBF`.
`Con_ResolveDestPtr` EA `0x11C48` and `sub_13C80` are separate code	Verified as separate call targets.

C. Invalid tests: don't reuse as evidence

These measured something real but don't support their old conclusions:

XOR, ROT-N and 5/6/7-bit scans for dialogue, including all "no French" results. Only byte-aligned or wrongly-tabled forms were tried.
"No glyph-index stream exists": correct for raw bytes, irrelevant to packed text.
Bitmap pitch, autocorrelation and bit-density tests on .CON data: pixels are nibble and run-length coded, in many scan orders.
MAUPITI1 "contains no text" from a printable-ASCII census: the game's text isn't ASCII, so the census can't show that. The text is in G3.CON; whether MAUPITI1 also embeds packed strings isn't tested.

D. Confirmed in `MAUPITI1.EXE`

Unpacks from .olz to an identical image; INT 0FEh handler isn't in it; 51 stubs; 21 INT 21h sites; one iret; no in, out, INT 10h, INT 33h; two rep movsb in the runtime; 7 printable runs; CD FE false hits are displacements; 0x1335A pushes constant 83, unrelated to segment 0x53. The instruction counts differ by tool (IDA 29,083 or 29,099; a linear sweep gives 29,145). IDA's runtime names, such as @Assign, weren't re-read, but the code behind them has no far callers.

E. Unverified

Conclusions from runtime dumps not in this repo: D-1F11.BIN, S8C76.BIN, S7832.BIN.

Previous: Naming · Next: Open questions · Technical appendix · Home

Hbtvl