How reviewing works

A quick guide for anyone new to reviewing on RoboVerify — the concept that trips people up most is claims vs. evidence, so start there.

1. A claim is one fact. Evidence is proof for it.

A claim is a single statement about a robot — "Payload: 14 kg." A claim has exactly one value.

Evidence is a quote from a preserved artifact that speaks to that value. A claim can — and often should — have more than one evidence link.

Common mistake: if the manufacturer says 35 kg and an independent review says 34 kg, don't create two separate "Weight" claims. Create one claim (use the manufacturer's stated value) and link two pieces of evidence to it — one Supported, one Contradicted. That's what lets the system detect the conflict automatically.
2. What each relation means, when you link evidence
SupportedThis evidence states the same value as the claim.
ContradictedThis evidence states a clearly different value.
UnclearThe evidence touches on this but doesn't clearly confirm or deny it.
Context differentBoth numbers may be correct, but for different configurations, models, or conditions.
Exact mismatchA precise, unambiguous conflict — same metric, same conditions, different number.
Partial mismatchClose but not identical — worth flagging, less severe than an exact mismatch.
Source conflictTwo evidence pieces disagree with each other, independent of the claim's own value.
Insufficient evidenceNot enough here to confirm or deny — flagging that more digging is needed.
3. What happens automatically

Any relation other than "Supported" immediately creates an open Mismatch — you don't need to do anything extra. It shows up in the Mismatch queue right away.

4. Resolving a mismatch — the four decisions
  • Approve mismatch — it's real, keep it flagged as a genuine conflict on the public profile.
  • Mark contextual difference — both sources are right, they're just talking about different things (e.g. base vs. arm-equipped weight).
  • Request more evidence — not enough to decide yet; leave it open and go find a third source.
  • Reject false positive — on closer look, it isn't actually a conflict (e.g. rounding, unit confusion).
5. What the confidence badge means

It's a plain rule, not a hidden score: Disputed if there's an open mismatch, High if 2+ evidence links support it and at least one is from an independent source, Medium if it has any supporting evidence at all, Low if it has none yet.

Quick checklist for reviewing a robot end to end
  1. Approve the robot and its sources in their queues
  2. Capture at least one artifact per source
  3. Add each claim once, with its most authoritative source's value
  4. Link every piece of evidence you find to that same claim, with the right relation
  5. Resolve any mismatches that appear
  6. Publish the claims that are ready, then publish the robot profile