All 3 contract terms met. Meta’s SAM 3.1 boxed every “drink can” in all 326 frames; Meta’s Muse Spark confirmed the LaCroix product and rejected 1 look-alike.
Pay or hold · a recommendation, you decide
Nothing here moves money; you pay as you normally do.
Every contract term was met, each with the moment that proves it below.
Muse Spark with tools · you approve anything sent
Reads this report and @Aldo 2’s other videos, drafts the note to them and makes the pay/hold call. About $0.10, at most $0.16.
No password needed on this example report. It reads the stored results; nothing is sent to the creator. Approving a re-cut link needs the demo password.
Anyone with the link can view this report and its proof clip.
Select a term to play its proof.
Boxes on the video come from SAM 3.1 tracking “drink can”. Lime is counted. Dashed red was rejected by Muse Spark as a look-alike.
SAM 3.1 tracked them; Muse Spark checked each against the brief. Look-alikes it rejected and objects it could not make out are shown for transparency and never count.

Heard by Muse Voice Transcribe. Select one to hear it.
Whether the video says it is an ad: heard by Muse Voice Transcribe, or read on screen by Muse Spark in up to 10 sampled frames (#ad, “Sponsored”, #partner).
Analyzed 326 frames at 15 fps · Every Meta API call, $0.155 · How accurate is this?
SAM 3.1 is billed per frame ($0.20 per 1,000); Muse Spark and Voice Transcribe count the fixed amount the budget reserves per call ($0.02 / $0.01), not the billed amount.
Measured frame by frame on 11 test Reels as delivered, against labels a person checked: precision 98.4%, recall 91.1%. Precision is the share of the seconds Screen-Time counted that really showed the product; recall is the share of the seconds the product really showed that it counted.
A test campaign run by Screen-Time with friends, who agreed to be shown. Not affiliated with, sponsored or endorsed by LaCroix or Poppi; their names and cans belong to their owners.