Evaluation protocol

Reliability claims should earn trust out of sample.

Quanta uses relative model scores to rank research setups and does not present them as calibrated probabilities. Public numeric calibration figures are temporarily withheld while the evaluation is rebuilt around a chronological holdout and a strict outcome-horizon embargo.

Review public Results Read model disclosures

Publication standard

What must pass before numbers return

Chronological separation

Calibration is fitted on an earlier period and evaluated on a later, untouched period. Random reuse of the same observations is not accepted as public evidence.

Outcome-horizon embargo

Historical neighbours whose forward outcome was not knowable at the evaluation date are excluded from that evaluation.

Sample-aware reporting

Reliability bands will be shown with sample counts, date windows, version labels, and uncertainty—not a single context-free score.

What is available now

The public Results ledger reports matured scheduled calls under one fixed 10-bar observation contract, including sample size, a confidence interval, date coverage, and direction-adjusted outcomes. It is the current evidence surface for evaluating the product.