Every value carries its proof.
When Lirovo returns a value, it returns where that value came from with it. The exact second in the video, and whether it was said out loud or shown on screen. Evidence is not a feature you turn on. It is built into the harness.
What an evidence span is
A span is the smallest unit of proof Lirovo records. It pairs a timestamp with a modality and points at the source moment behind a value. A single value can carry several spans.
Timestamp
The exact moment in the video the value came from, down to the second. Not a paragraph of transcript, the precise point you can jump straight to.
Modality
Whether the value was said out loud (audio) or shown on screen (vision). The same value can carry several spans, so a fact heard and shown lands as two.
Source moment
The transcript line that was spoken, or the on-screen frame that was read. The span points back to the real moment, not a summary of it.
Read together, a span says: this value came from this second, in this way, and here is the moment that proves it. An answer you cannot trace is an answer you cannot trust.
Verify it yourself
You never have to take a value on faith. Click any field and jump to the exact moment it came from in the video. Read the transcript line that was spoken. See the on-screen frame that was read. The proof is one click away from the answer.
Because evidence is mandatory and architectural, there are no unsourced answers. Every extracted value points back to a source moment and a modality. If a value made it into your output, the moment behind it is right there to check.
A value backed by two modalities, both heard and shown, is stronger than one backed by a single mention. Confidence reflects that, so the values you can trust most are the ones the recording confirms more than once.
When sources disagree
Sometimes the audio and the on-screen text say different things. Lirovo does not silently pick one. It flags the contradiction and shows both sides with their spans, so you decide which one is right for your use.
The spoken valuation (above $13B) disagrees with the lower-third graphic (about $14B). Caught by cross-checking the transcript against what was read off the frames.
the contradictions view: where audio and on-screen text disagree, both sides shown with their spans
See the proof on your own video
Run a live extraction and click through to the moment behind each value, then pull the evidence straight into your own systems, all on your machine and your own inference.