Observation
What happened, where and when.
Methodology
Kaupr records commercial reality first. Everything else—identity, memory, change and signal—must remain traceable to what was actually observed.
What happened, where and when.
The underlying record stays attached.
Domains resolve to economic entities.
Observed facts become Commercial State.
States are compared across time.
Sentinelle ranks what deserves attention.
What was observed. When it was observed. Which domain it came from. Which economic entity the evidence belongs to. How that state differs from prior observations.
Missing history is not treated as stable history. No recorded event is not automatically treated as “nothing happened.” Semantic similarity alone does not establish ownership. A prediction is not rewritten after reality arrives.
These results support a forward-looking ranking of change regimes. They do not establish exact-next-action prediction or prospectively calibrated event probabilities.
Historical examples without enough later observation are censored instead of mislabeled as negatives.
A challenger must improve untouched shadow ranking without unacceptable calibration regression before replacing the frozen champion.