Understanding OmniQuant

How accuracy is measured

• OmniQuant scores the calls it publishes from its served model — not challenger, shadow or on-demand runs. Each call is scored in the direction it was published.
• A call is a hit when the price moved in the called direction by more than the noise band by the target date, and a miss when it moved the other way. A move inside the noise band is neither: it leaves both the hits and the misses.
• Accuracy = hits ÷ (hits + misses), over calls made in the last 90 days at most, counted by the day each call was made and stated separately for each market and horizon. Pending calls are not counted until their outcome is known.
• Each call is compared with how often its own direction happened in the same market and horizon over the same days (the base rate): an up call with how often prices rose, a down call with how often they fell. Lift = accuracy − base rate.
• Every figure states how many calls it rests on, how many moves fell inside the noise band and how many calls are still pending.
• No edge is claimed. Accuracy and lift describe past calls; they are not a return and not a promise about future calls. The model score shown with a call ranks calls — it is not the chance that the call is right.