The causal model inside Ceptron, with open weights. Which variables drive which, at what lag, from a plain time-series table, in one pass. Nothing is fitted on your data and there is no statistical fallback.
| Series | Files | AUROC | F1 |
|---|---|---|---|
| Controlled process series | 72 | 94.3 | 79.1 |
| Retail daily series | 26 | 88.4 | 69.6 |
| Business-style series, answered by every compared checkpoint | 103 | 83.6 | 47.2 |
| Business-style series, all sent | 118 | 79.3 | 41.2 |
It is a general discovery model for regular or irregular time series with mixed column types. It is not the whole of Ceptron: it reports which variables drive which and at what lag, not effect sizes, root causes or the sign of an effect. Its scores are the network's readouts, not calibrated probabilities. Its weak spots, from the card: coarsely sampled or irregular series, delays of 20 to 60 steps, slow hidden trends, and same-step links between count or binary columns. Control loops, sensor read-through and industrial process mechanisms were not part of its training; the trap results elsewhere on this page are for the causal model inside Ceptron, not for this download.
Twenty ways data misleads, each one a test file we built with the answer planted first, so the result could be marked rather than judged. The causal model inside Ceptron passed all twenty. Click a card for the trap, what the model did, and the test files.
The open model runs on any time-series table: one column per variable, one row per time step. Install it, load it, call discover.
These are the twenty traps that passed, not the whole picture. Our testing also covers cases that are partly handled and cases that are not handled yet. The test files, the planted answers and the raw outputs are available on request. Nothing here is a result on customer data.
Tell us what you want to model and we will be in touch.