Every run tests a prediction.
We plan to build an open pilot hub in Lombardy where companies and researchers scale processes from 1 L to 1,000 L. Before each run the engine's prediction is written down. After the run it is scored, and the engine is corrected where it was wrong.
Three levels, one geometry family.
Each vessel is characterised before it runs a single batch: kLa and mixing time are measured on the vessel itself.
Labs
Bench reactors.
Partner processes are reproduced where they already work, and the inputs the engine is most sensitive to are measured: oxygen uptake, residual substrate, uptake rates.
Pilot
First step up.
The first scale-up prediction of each process is tested here: which limit, at what operating point, under which scale-up rule.
Scale
Where most failures show.
Predictions made from bench and 150 L data are scored at the volume where gradients, oxygen and heat start to bite.
The loop each run follows.
A run that was not predicted teaches less. A prediction that is never scored teaches nothing.
The engine's verdict, first limit and ratios are recorded before the run, with every input and its confidence level.
The process runs on a characterised vessel, with dissolved oxygen, off-gas and temperature logged.
Predicted and observed limits are compared. Misses stay on the record.
Correlations are recalibrated on measured data, and the change is dated in the method note.
What exists and what does not.
Filled nodes are done. Dashed nodes are planned and have no date we can stand behind yet.
- 2026Prototype engineFive checks, published correlations, sources labelled by status. Internal only.
- plannedDesign partners3 processes with existing 1-20 L data, checked before any hub exists.
- planned1-20 L labsCharacterised bench reactors.
- planned150 L pilotFirst scored scale-up predictions.
- planned1,000 LOpen to companies and researchers.
Have 1-20 L data on a process that must scale? That is where the hub starts.
Bring us a process