Confidence

Kernel scores its own work. Every step in the cleaning and enrichment process carries a confidence level — a quick way to see how sure the system is about its decision.

These levels indicate how certain Kernel is about the accuracy of the data or recommendation:

• High – everything lines up cleanly; Kernel’s confident in this result.

• Medium – there’s a mix of good signals and a few weak ones.

• Low – data conflicts or uncertainty make this less reliable.

Kernel also rolls those checks into one overall confidence rating for each account.

If an account ends up with Low confidence, it’s automatically flagged for Review. That way, you’ll know exactly which records need a human look before you act on them.

Kernel considers varying elements in the confidence scoring process:

• Strength of Source Evidence: how clearly the original content supports the extracted information.

• Complexity of the Extraction: how difficult the underlying interpretation or reasoning is.

• Structure of the Input: how organized or unstructured the source material is.

• Cross-Source Consistency: whether multiple references align on the same datapoint.

• Internal Consistency Checks: how well the datapoint fits established patterns or prior values.

• User Feedback Signals: how past confirmations or corrections influence confidence.