Bagsack builder

Bagsack builder lets you create a custom game of Backsack and watch the AI play its best against it. AI tries to optimize item ordering and drop location policies to minimize bag packing costs (bag density, speed, out-of-bag-drops, item damage). Hopefully you'll see some cool emergent Tetris-like behavior, pick items wisely!
Planner trained on 6 episodes
Drop Oracle trained on 364 drop worlds
Selected 0 / 200 items Distinct products 0 / 0 $0.00 total item value left-click +1 · right-click −1
Item-to-item retrieval matrix
Constant = every live pair costs $0.01. Random reproduces the Virtual retrieval-matrix synthesis used by Planner data generation (same range and per-episode seed).
The seed fixes item placement and the random matrix — a pinned seed makes the Playback render reproducible. Names are stripped of quotes so the inline command stays valid.
Auto-generation is currently unavailable on this server, so a Generate click will only build the JSON and docker commands below (no video). You can still render the world manually with those commands.

Auto-generation (this server)

Submitting…

Planner world JSON

This is the raw Planner Bagsack world JSON (schema v1). It is the only encoder of this format — nothing is saved to disk or retained after refresh.

Save as a world file, then run Playback

Paste the JSON above into ./worlds/my_bag.json first (the playback service mounts ./worlds read-only). Run docker build -t dropsim-gen . once before the first run.

Inline world JSON literal

Passes the world as the --world-json literal string (single-quoted, so keep apostrophes out of the name). Both commands drive the canonical Planner entirely — no random play against real Genesis physics.