A fact is more than its value
A useful fact includes its source, supporting quote or evidence, retrieval time, and confidence. This makes it possible to understand why the app believes something and decide whether that evidence is sufficient for your task.
Source type matters
The contact intelligence pipeline distinguishes user records, conversational evidence, and researched sources. Your CRM records participate in reconciliation and outrank scraped sources so a web page does not silently overwrite information you deliberately supplied.
Corroboration needs independent evidence
Two copies of the same claim are not necessarily two independent confirmations. Read the provenance behind corroboration. Multiple websites can repeat a single outdated biography or confuse two people with the same name.
Confidence is not truth
A confidence value reflects the system’s assessment, not a guarantee of accuracy or an objective score of a person’s honesty. A reputable publisher can carry an outdated detail. A direct statement can be tentative, misheard, or about someone else.
Keep inference visibly separate
Model-generated conclusions and suggested actions should remain distinguishable from quoted evidence. Review identity matches, conflicts, and partial research before using a profile. Personal notes and sourced research serve different purposes.
Try it with a sample
- Open Research & Facts for a fictional or permitted sample.
- Inspect the quote, source, confidence, and retrieval context.
- Resolve identity ambiguity before accepting the detail.
About this explanation
Reviewed against Modules/Assistant/Shared/Services/ContactIntelCoordinator.swift. Examples use fictional people and content. Availability depends on the build, edition, permissions, and selected services. Source changes trigger a review notice and removal from website AI retrieval until reviewed.