Evidence-backed stats

See the evidence behind every supported ApexLens result.

A result is more useful when it includes status, detector-local confidence, provenance, source time, and limits—enough context to check it against the footage when something looks wrong.

Join the private beta See how it works

Detected, estimated, or unavailable
Detector and timestamp provenance
Correctable review output

Illustrated product concept. Beta output varies by recording and detector confidence.

The field wrapper

A number should arrive with enough context to challenge it.

Most structured ApexLens fields use the same contract. The value is only one part of the result; status tells you how it was obtained, confidence expresses detector-local quality, and provenance records the method and source time.

Confidence is not a statistical probability, and a field can stay unavailable when the recording does not expose enough evidence.

team_killsDetected

Value: 6
Confidence: 0.82 detector-local
Detector: summary_stats
Method: summary frame
Source time: 11:02

Source moment · beta UI example

Representative schema example, not a claim that every recording returns this value or confidence.

Three honest states

Unavailable is better than a made-up zero.

Detected

The available evidence supports a direct result from the configured detector.

Estimated

The output is useful as an estimate, with confidence and provenance still visible.

Unavailable

The footage does not support the field well enough, so the value remains null.

Field-level boundaries

The label matters as much as the value.

Weapons used: Visible POV weapon labels, not teammate loadouts.
Accuracy evidence: Sampled hit-marker frame rate, not true shots-fired weapon accuracy.
Damage taken: Observed damage-indicator windows, not a numeric damage total.
Activity mix: A low-confidence baseline for coarse timing, not authoritative behavior scoring.