Signals identify meaningful departures from established patterns, then show the observations and relationships that contributed to the result.
A Probbl is assembled from weighted observations, inferred relationships, and the agency’s configured relevance rules.
A linked device appeared outside its recurring 90-day activity corridor during an agency event.
The device cluster was observed within a 180-meter confidence area at 01:38. The location falls outside 98.6% of the cluster’s prior overnight observations.
The cluster-to-person link is probabilistic. Presence indicates an associated device, not necessarily the person.A vehicle associated with the same household was observed three minutes after the device presence event.
Plate 7LXE229 was read 0.4 miles east of the location confidence area. Direction and time were consistent with departure from the event corridor.
The vehicle is registered to Jordan Lee and associated with Avery Bell through shared-address and repeat-location evidence.A second-degree connection links the subject’s household to a person with an active agency watch.
Jordan Lee and Riley West have 11 recurring co-presence events and one shared commercial account. Riley West appears in an active Bellwether agency watch.
Avery Bell has no recorded direct contact with Riley West. Relevance is inherited through the graph.Independent commercial, partner, and agency sources align closely enough to increase confidence in the combined interpretation.
Temporal and geographic agreement among four records exceeded the agency’s coherence threshold. No single record independently supports the resulting narrative.
Coherence increases confidence in correlation. It does not independently establish causation or personal presence.Probbl separates observation confidence from operational relevance, allowing agencies to decide which combinations warrant attention.
Ingest source records with native timestamps, identifiers, locations, and provenance.
Raw records are normalized into event objects. Probbl retains source-specific confidence, licensing restrictions, and correction status.
Associate each observation with likely entities in the Commonplace Graph.
Probbl uses stable identifiers, normalized contact data, device clusters, recurring location patterns, household proxies, and association history.
Measure the observation against an individual or cohort baseline.
When personal history is sparse, agencies may permit comparison with a behaviorally similar cohort until an individual baseline matures.
Combine confidence and agency relevance into a ranked signal.
Agency configuration determines thresholds and factor weights. Protected characteristics may be excluded as direct inputs while correlated proxy variables remain present.
A score becomes operational when it crosses a configured threshold, persists long enough to satisfy policy, and matches a route with people available to act.
Monitor a person, household, vehicle, device, address, relationship cluster, behavioral cohort, event corridor, or geographic area.
Monitoring can be assigned directly or inherited when a newly resolved entity enters an already monitored graph neighborhood. The inherited entity need not appear in the originating agency record.
Require agreement among signal confidence, operational relevance, source diversity, and a minimum persistence window.
confidence ≥ 80 · relevance = high · source classes ≥ 3 · persistence ≥ 4 min. Customer policy determines whether inherited associations may satisfy relevance.
Send a Priority Context brief to an investigator, supervisor, RTCC queue, dispatch workflow, or nearby authorized field unit.
A brief may be created, exposed, and routed without a named complainant, reported offense, direct agency observation, or individual user search. Agencies determine whether the result supports observation, contact, or another response.
Probbl preserves the history needed to explain what the system showed, what the user did, and what later information changed.
The signal becomes visible to users whose role and purpose match agency policy.
A user can confirm, escalate, dismiss, or leave the signal unresolved.
Inputs, outputs, user actions, and the original score are preserved for audit and model evaluation.
A dismissed signal may contribute to a future signal when new observations materially change its relevance.
A score of 87 means the configured model found strong agreement among the available observations and associations. It does not mean there is an 87% chance the subject committed, intends, or knows anything.
Probbl makes this distinction visible in the interface. Agencies remain responsible for how contextual signals affect encounters, inquiries, and investigative decisions.
Read the human judgment principle →