Provenance & drift
A model is only as defensible as the data behind it. Provenance traces training data as lineage, from source to transformation to training set to model version, so you can answer where a model learned what it knows. Drift watching runs alongside, telling you when the data the model sees stops matching the data it learned from.
Training data is traced as lineage on the knowledge graph, from source system through transformation to training set to model version.
Provenance claims are checked against evidence, and a claimed training source with no lineage behind it shows as unverified.
Drift is watched over time on inputs and upstream quality, with tolerances set by the model's risk tier.
Impact questions are answered through lineage: when a source system changes or degrades, every model trained on it or fed from it is identified.
Personal-data flags mark training sources, so GDPR and EU AI Act exposure is visible model by model.
Drift findings close only through verified closure. Either evidence shows the drift resolved, or the issue stays open.
Trace
Training sets connect to their source lineage on the graph. Access is federated, and nothing is copied.
Watch
Input distributions and upstream data quality are monitored continuously for drift.
Flag
A breach opens an issue tied to the affected models, and a feed that goes quiet is itself treated as a failure.
Close
Fixes pass verified closure. The issue closes on evidence, and nobody closes it by hand.
“What data was this model trained on, and where did that data come from?”
Training data is traced as lineage on the knowledge graph, from source system through transformation to training set to model version, so you follow the trace for any model. A claimed training source with no lineage behind it shows as unverified rather than being accepted.
“If we change this source system, which models are affected?”
Lineage answers that directly. When a source system changes or degrades, every model trained on it or fed from it is identified on the graph, so the impact analysis is a query rather than an investigation.
“How would you know this model is drifting before the numbers go wrong?”
Input distributions and upstream data quality are monitored continuously, with tolerances set by the model's risk tier. A breach opens an issue tied to the affected models, and that issue closes only when evidence shows the drift is resolved.
Start with proof, not a promise.
There are three ways in, and each one earns the next. You approve the exit criteria before we spend any more of your time or budget.
A 60-minute working session
Bring one report that is expensive to prove. We walk the mechanism against it. Nothing is required from your side.
A two-week guided sandbox
Your team works hands-on against a synthetic sample bank. It requires zero access to your systems and zero procurement.
A six-week proof on your systems
We use read-only metadata across five systems and one priority report, scored against criteria both sides sign first. The decision it produces is yours.