What the literature reports about producing cells from iPSC — expansion, embryoid bodies and aggregates, differentiation, maturation, scale-up and harvest — read once and kept on this server. Every run starts its literature search here and looks only for what the library lacks. Each value keeps the sentence it came from; a value about a neighbouring quantity (EBs per well rather than cells per mL) is marked as a proxy with its conversion.

Registered data this server can see. Datasets someone registered privately are never listed or served here: they stay on the machine that holds them and are reachable only from the command line there.

A processed table (.csv or .tsv): one row per sample or population, a column that says which condition each row belongs to, and the readouts. It is registered privately on this server and the bioinformatics agent can analyse it when a run names it. It is never served over HTTP and never becomes a literature citation.

A tool that cannot run here is not offered. That is deliberate: an analysis refused while it is being planned is a decision you can act on, and one that fails half-way through a run is a wasted afternoon.

Each process exposes the knobs it actually has, with its own limits. A parameter marked not modelled is still a real design variable you can set in the lab — the model simply has no term for it, so it produces no prediction for it, rather than a number that would be invented.