Every claim validated.
We publish validation studies with methods, numbers, and limitations for every claim. We'll run the same validation method on any prior study, to prove the technology works on your data.
Twin‑2K‑500 — survey synthesis benchmark
Apr 2026, Public benchmark, Toubia et al.
0.005 categorical MAE across 694 variables: 64% below the sampling-noise baseline. 100% novel rows. External discriminator only 4.9 pp above real-vs-real.
Pricing & Promo — year-ahead pricing validation
May 2026, Customer blind holdout, Fortune-class FMCG
172-unit median miss across 195 held-out product-week cells. 83% inside product-level weekly volatility. 4 of 4 products with the correct price-response sign.
Data reduction — sensory product testing
Sept 2024, Customer validation, N. American beverage co.
80% reduction in data requirements — reconstructed a 2,782-person study from 500 obs. Bayesian changepoint corroborated.
Bring a study you already fielded. We'll predict the holdout.
Validate on your data: we hold out a portion of your data, train Simulacra on the remainder, generate predictions over the holdout, and send you the scorecard. Standard NDA, no contract required.
What we test: