Human validation measures how people respond to the same study structure used by the APM arm.
What the process separates
- Collection: Were participant responses gathered?
- Review: Were responses scored independently and adjudicated where needed?
- Sufficiency: Is the reviewed evidence adequate for the intended claim?
These stages are separate. A completed import does not mean review is complete, and completed review does not guarantee sufficient evidence.
Import responses
CiCi accepts the study response file associated with the sprint reference. Valid rows become human tasks. Malformed or unresolvable rows are quarantined or rejected with a reason rather than silently entering denominators.
Re-importing an identical file is a no-op. It does not create duplicate responses.
Review responses
The reviewer workspace separates blind independent review from adjudication. During the independent pass, comparison targets are withheld by the server—not merely hidden in the interface.
After the independent pass is committed, CiCi reveals the comparison needed to determine agreement or adjudicate a disagreement. The original model score, independent human score, and authoritative result remain distinct.
Close collection
Close collection only when fieldwork is genuinely over. Before closure, failing gates mean not yet. After closure, those same gates can support an insufficient evidence outcome.
You can reopen collection with a reason. Reopening clears a previous acceptance because new evidence can change what the study supports.
Do not treat responses awaiting review as provisional customer-facing findings. Half a review is not half an answer.
