Run the experiments on your own sessions and report the numbers

[sepehrsadri]

Every number in docs/EXPERIMENTS.md was measured on a handful of projects on a single machine. This is the single biggest weakness of the research, and it is the easiest one to fix — no Rust, no reverse engineering.

The task

Run the prototypes against your own sessions and report what you get.

python3 experiments/als_semantic_diff.py --chain '/path/to/YourProject/Backup/*.als'
./experiments/storage_bench.sh '/path/to/YourProject/Backup'
python3 experiments/flp_parse.py '/path/to/project.flp'

Ableton keeps timestamped autosaves in your project's Backup/ folder, so you already have a version chain.

Claims most in need of independent checking

• Consecutive Ableton saves differ by 0.07–0.25% of XML lines
• Track IDs are stable across the save chain
• A median of 3 of 28 tracks change per save
• Delta chains cost ~11 KB per save

Please include the material context — genre, track count, project size, how many versions, DAW version. No file names needed. Context matters enormously: sparse stems and dense masters behave completely differently, and a number without material is close to meaningless.

Results that contradict ours are the most valuable outcome. Use the "Measurement report" issue template.

Everything in experiments/ is read-only and never modifies your projects.

[hemanth-sai6792]

Hi, I'd like to work on this. Could you assign it to me?

[sepehrsadri]

@hemanth-sai6792 — sorry for the slow reply, and thanks for offering. Assigned.

Quick orientation since this issue predates a plan change:

• The steps in the issue description still work exactly as written — nothing here
  needs the Rust code that's landing separately. python3 experiments/als_semantic_diff.py --chain '/path/to/YourProject/Backup/*.als' etc. are still the whole task.
• Most useful material: Ableton .als project backups if you (or anyone you know) uses
  Live — that's what most of the claims-to-verify list is about. FL Studio .flp is the
  other prototype with real numbers behind it if you have that instead.
• Please read the "Please include the material context" note in the issue body before
  you post results — genre/track-count/project-size context is what makes a number useful
  here, not just the number itself.
• Use the "Measurement report" issue template for what you find (link at the top of the
  new-issue picker). A result that contradicts what's documented is the most valuable
  outcome, not an awkward one — say so plainly if you get one.

No deadline, no expectation you have Ableton/FL Studio lying around — if you don't, no
worries, just let us know here rather than sitting on it.