Proof

The proof. Real moves, real baselines, real results.

Anyone can claim accuracy. Here is ours, scored against reality you can verify. Every example shows the baseline, the prediction MindLoop called, and what reality returned. Misses are left in.

Start here

It doesn’t guess. It checks.

When MindLoop says a markdown will clear 64% of your stock, it checks later whether it did.

MindLoop predicted64% would clear
Reality 64% cleared
How we label every claim
VerifiedA move in a real store, checked against actuals after the fact.
CalibratedScored on a public benchmark anyone can re-run. The six below.
ExploratoryA direction worth testing — not yet proven. Always labeled.

The six benchmarks below are all Calibrated — public benchmarks you can re-run yourself. In your store, the same checks run on your real actuals and come back Verified.

Proof

Scored against ground truth.

The same engine that answers your customers and prices your floor is graded on six public ML benchmarks. Our predicted band sits right next to where reality landed — and you can verify the source packs yourself.

01

Retail Demand · M5 / Walmart

M5 sales evaluation · 60,288 rows · 10,000 targets
Metric · WRMSSE (lower is better)
Predicted · 0.620.70Naive seasonal · 1.00Actual · 0.65
✓ Within bandM5 holdout
02

Advertising Uplift · Criteo

Criteo uplift treatment/control · 120,000 rows · 10,000 targets
Metric · Qini coefficient
Predicted · 0.0780.090Random treatment · 0.000Actual · 0.084
✓ On the noseCriteo holdout
03

RetailHero / X5 Uplift

RetailHero X5 treatment + purchase target · 80,000 rows · 5,000 targets
Metric · AUUC
Predicted · 0.5950.625Coin-flip uplift · 0.500Actual · 0.608
✓ Within bandX5 holdout
04

Hillstrom Email Uplift

MineThatData randomized email · 100,000 rows · 8,395 targets
Metric · Incremental conversion (pp over control)
Predicted · 5.1%5.9%No-mail control · 0%Actual · 5.7%
✓ Within bandHillstrom holdout
05

HIGGS Event Classification

UCI HIGGS event rows · 450,000 rows · 10,000 targets
Metric · AUROC
Predicted · 0.8660.882Logistic baseline · 0.733Actual · 0.879
✓ On the noseHIGGS holdout · single shot
06

FreshRetailNet · Stockout-aware demand

Weather · promo · holiday · inventory covariates
Metric · wMAPE (lower is better)
Predicted · 0.300.36Naive seasonal · 0.58Actual · 0.32
✓ On the noseStockout-corrected holdout

Public source packs. Independent holdout targets. Same metric the field uses. The predicted band and the actual landing point sit next to each other so you can see exactly how close we came.

Put your own store to the test.