Skip to main content
Orbb has not examined every kind of evidence for every person. When you ask about one it has not examined, it tells you so — it does not answer “no”. This is the most important thing on this site. A pair we never looked at and a pair we looked at and found nothing for are different answers, and code that treats them the same will quietly drop exactly the people you were looking for.

Reading coverage

Every circle, edge and paths answer carries coverage, one list per outcome:
limits names what bounds the answer regardless: NO_LIVE_SEARCH (comments are read from those already known, never searched for live), NEVER_ALONE (an education match is shown only beside another type), SOURCE_LIMITED, EXCLUDED_BY_SCAN, STALE. Two more fields summarise it:
  • hit is RESEARCHED only when every requested type has completed and none is out of date; otherwise UNKNOWN.
  • research_state gives, per type, its coverage, whether it is stale, and when it completed.

Rules for code that filters people

1

Never read a missing strength as zero

strength is absent when no type was found. Sorting it as 0 ranks a person we never examined below one we checked and found weak.
2

Only NO_EVIDENCE_FOUND means no

An edge verdict of UNKNOWN means we could not check. Show it as “not checked”, not as “don’t know each other”.
3

Say what was not examined

If you show a circle to a person, show not_examined with it. “No investor relationships” and “investor relationships not checked” lead to different next steps.

Asking Orbb to research: on_miss

By default (on_miss: UNKNOWN) a call answers from what is already held, reports the gaps in coverage, and costs its flat price. on_miss: RESEARCH researches the requested types first, then answers:
  • It takes longer, and is more often answered 202.
  • Research is charged on top of the call’s price, up to max_tokens (default 2, at most 600). The call reserves that much before it starts and returns what it did not use. See Billing.
  • A type researched recently is mostly not researched again. On circle and paths, force: true researches it anyway.
  • Only the listed types are researched — narrow them to what you need.
  • At most 6 research or resolve jobs run per account at once; past that, 429 research_busy. See Rate limits.
Start with the default. Look at not_examined, and research only the people and types where the gap changes your decision.