Status & Open Work

This page mirrors the repo’s single status surface, docs/legend/TODO.md. Every genuinely open item gets one row here; the full evidence and paths-tried tables live in the raw notes.

Legend has the most open work of the five games, for one structural reason: wofl.exe has no symbol table and a Borland-style overlay structure, and remains untraced. That is why restart.dat — not the executable — is this site’s source for item and character data.

Open items

ID	Status	Question (one line)
`legend-necs-fields`	open	Residual `NECS` scene-object record semantics: what the 543 Class-C marker records (ids 1504, 1512, 1514 — exit/entry-zone shaped, whose `u16` at `+0x18` resolves to no known id space) actually are; whether Class-B `1999` quads are drawn or are floor/walkable metadata; the meaning of `u16` at `+0x12` (a depth-ramp “shade” hypothesis was refuted) and at `+0x14` (a 13001–13066 per-instance tag); and 8 mostly-zero sprite slots.
`legend-exe-data`	open	The character/party table beyond its 5 base attributes (which are confirmed against a published walkthrough): a name-relative `01 64 00` triple run, a duplicate attribute block’s purpose, and a mostly-zero footer are all unidentified. The item-placement region after the item table (`0x172A`+) is only partially characterised — a leading 29-byte-stride empty-slot array, heterogeneous beyond that — and unstarted.
`legend-tapm`	open	What the 84 `TAPM` tile-patch resources mean. Structure is known: 37 of 84 are LZSS-compressed (the same convention as `PAMM`/`GAMI`/`LMRF`/`NECS`), 47 are raw, and decompressing collapses the record sizes to 7 consistent logical shapes (20 B×4, 24 B×16, 32 B×8, 48 B×29, 72 B×6, 84 B×20, 96 B×1). The semantics of any of them are open — a `(tileByte, flagByte)` pair hypothesis for the 48-byte family was tested against all 5 map atlases and refuted.
`legend-frml-animations`	open	Which frames of each `LMRF` resource form a named animation. 22 of 162 segments are render-confirmed and 8 of 162 structurally confirmed (an exact-prefix boundary shared across resources, name unverified); the remaining 146 of 162 are hypothesis-level, with no boundary signal firing. There is no executable-derived ground truth for which creature owns which resource id, and the id numbering does not divide cleanly into walk-cycle versus combat ranges.

Negative results worth recording

Item	Result
The 6 `.med` files	Not valid OctaMED modules — no `MMD` magic. The docs previously said otherwise and have been corrected. Beyond low-entropy byte-histogram observations they remain uncharacterised.
`TAPM` 48-byte family	The `(tileByte, flagByte)` pair reading was tested against all 5 map atlases and refuted. Recorded so it isn’t re-derived.

What this means for the galleries

The Creature Animations page shows segmentation at three different confidence levels, and labels which is which. It is not a rendering problem — all 162 LMRF resources decode cleanly, after three decoder bugs this corpus exposed (see FRML / LMRF). It is that nothing in the shipped data says where one animation ends and the next begins.

Where to read the full evidence

docs/legend/TODO.md — the single status surface.
docs/legend/plan.md — the format-by-format status, including the “NECS — still open”, “Character/party table”, “Item-placement region”, “TAPM” and “LMRF animation segments” sections each row above points at.
docs/legend/dosvga/engine.md — the byte-level spec.

The raw notes carry the full evidence, paths-tried tables, and corrections that this page summarises.