BugSorter Bench Twin
Configure the sample-prep line, then read what each station does to the sample, which stations earn their place (leave-one-module-out), what the vision model sends to the entomologist, and whether the specimen can still go to whole-body DNA or RNA sequencing.
Pipeline
Each station is a Merkle leaf; the root at the top commits to the configuration and every station's output.
Where the sample went
Break points and checks
Leave one module out
The same sample run again with one station removed. A station that changes nothing when removed is not earning its place for this sample.
Entomologist queue
Specimens the vision model could not place. Confirming a class makes it known; retraining updates the model version (simulated) and re-hashes the run.
Sorting hierarchy
Nested partitions: station → size class → class label (an abstraction of biological hierarchy sorting).
Run record
Configuration, attestations and hashes for replay. Copy it into the repo or a notebook.