MOV is where the two engines diverge most, and the page should lead with that

[dhobi]

You asked whether v1 and v4 MOV differ more than the docs let on. Checked it, and the answer is in two parts.

The page itself was right

mov.md's DreamFactory 1 section is the most thorough v1 account in the doc set — seven differences, each with the DF.EXE read behind it. That is why the alignment pass in #279 never touched it, unlike SET and audio, which needed their v1 sections written from scratch.

I verified it rather than assuming. Every claim checks out against mov-v1.ts, including the two gaps it honestly admits to: +0x06 is not wait flags and the movie loop never reads it, and +0x00 is the hotspot count while the reader walks past it instead (disagreeing on 422 of 525 frames, always by finding more).

The census figures are better than accurate — they are guarded. dust/tests/movies.ts pins 152 frames of 6717, 69 segments of 185, 50 films of 160 against the disc, so that paragraph cannot drift without a red test.

The framing was wrong, and it was mine

Dust's movies are the same machine ... with everything v4 factors into per-frame logic containers kept inline

That reads as a packaging difference, and the one-line summaries I wrote in architecture.md and Dust's index repeated it. It is not a packaging difference.

The model survives — chain of segments, frame state machine, hotspots, chained films. Almost every mechanic under it differs, and more of it is behaviour than layout:

	v4	v1
advance	implicit	an authored action 2, target goto; a loop is a backward target
timing	frame rate	a frame can block until the sound it started finishes — max(hold, what's left)
hotspots	a region table	typed, variable-stride records walked from a per-frame offset, first hit wins
chains	—	type-3 exit posts a name and there is no return
exiting	ESC clears the next-segment pointer	any mid-film exit drops the rest of the chain
blit	—	indices 0 and 255 keyed out through a monochrome mask
action frames	names	1-based positions

Of the formats the two engines share, this is the one that diverges most — and all three places now say that first. Dust's "what is different, in one paragraph" left films out entirely, which was the same mistake where a reader meets it earliest.

One rule was genuinely missing

The segment teardown follows the next-segment pointer only when playback stopped on the last frame (0x404b9c), so a mid-film exit drops the rest of the chain — v4 reaches the same place by the narrower route of ESC clearing the pointer. Behaviour rather than layout, and exactly the kind a port gets wrong by inventing a return, which mov-v1.ts records having done once already (the wardrobe closing between the diary and the page-reading flat).

Scope

Docs only, no code. VitePress build clean, every anchor and link resolving.

🤖 Generated with Claude Code