Solve (or route around) the FL Studio v25 scalar keystream

[sepehrsadri]

FL Studio projects from v10 through v24 parse cleanly. v25 does not, and this blocks support for the current release.

What we know (see docs/FORMATS.md)

v25-era files appear to add an additive, file-offset-dependent obfuscation keystream over scalar (fixed-size, ID < 192) events:

• Two real autosaves of one project, same size: only 17 of 53,448 bytes differ (0.032%) — FL's writer is otherwise deterministic and positionally stable.
• But when file size changes by 48 bytes, 957 of 973 fixed-size events change, and 88.6% of the changed bytes differ by exactly +48 mod 256 — the file-size delta. All deltas are multiples of 16.
• Event count was identical, so the key is keyed on byte offset, not event index.
• A per-file constant K was recovered (K=135 for one project, K=136 for another), and byte_real = byte_stored + byte_offset + K (mod 256) correctly decodes IsEnabled/Type/PingPongLoop — but not tempo or time signature, so the key function is not fully solved.
• v25 also breaks the event-width rule at a fixed spot: ID 172 (0xAC) carries a 3-byte payload, so the stream must resync at 0x34.

The task

Solve the keystream, or prove it cannot be solved and implement the fallback (diff only variable-length events).

The experiment that would settle it — requires FL Studio 2025:

1. Open a project, save as A.
2. Change one note velocity (no length change) → save as B.
3. Change one mixer knob → save as C.
4. Add one note (length change) → save as D.
5. Rename a channel → save as E.

Then byte-diff A/B/C/D/E. B vs A isolates whether a same-size edit is local under v25 obfuscation. D vs A confirms whether any size change globally rewrites scalars. C vs A measures MixerID.Params blob locality. Repeat on FL 20.9 to confirm the regression.

Start from experiments/flp_parse.py. Please report exact FL version numbers.

[sepehrsadri]

Scope note (M0.5 truth-up): confirmed out of the 0.0 pilot — FL Studio is phase-4 in the write-path roadmap, not part of the Logic/GarageBand + Ableton 0.0 scope.

[sepehrsadri]

Environment note, so nobody plans around a capability that isn't here: the FL Studio install available for this project is FL Studio 20, not v25. That is squarely in the ≤v24 band where event locality is already proven (17 of 53,448 bytes changed between real autosaves, docs/EXPERIMENTS.md), so it is useful material for the Phase 4 event-stream parser — but it cannot be used to attack the v25 offset-dependent scalar keystream this issue is about.

This issue still needs someone with FL Studio 25 and its saves. Leaving it open and unassigned.