Methods

How each number is modelled

One page per predictor. Each lists the relations and constants in the code, what the model leaves out, which committed artifacts calibrate its intervals, and the gates that block a regression. The same ids appear on every evidence receipt.

Two MFC numbers, two models

The lab’s MFC tiles come from mfc-butler-volmer-monod-v2, a Butler-Volmer and Monod snapshot. /api/ml/predict, the MCP server and the chat assistant answer from mfc-mechanistic-closure-v1, the polarization curve of the stated reactor, when a request states a COD, and from Enhanced_MFC_Experimental, a calibrated material table, when it does not. For the same design the numbers can differ. Each number names its predictor on its evidence receipt.

Open decision: whether the lab tiles should move to mfc-mechanistic-closure-v1 is for the maintainers to decide. Until then all three are disclosed here.

mfc-butler-volmer-monod-v2mfc-mechanistic-closure-v1Enhanced_MFC_Experimental

How served intervals are built