FreshRetailNet-LT Leaderboard

• Recovery: the paper evaluates controlled censoring on non-stockout periods with observable sales, masking a suffix of within-day hours, primarily after 4 PM. Actual unmet demand during real stockouts is not directly observed.
• Forecasting: predict the next 7 days using raw sales or daily aggregates of recovered demand. The paper evaluates during operational periods without stockouts.
• WAPE: lower is better. WPE: zero is ideal; negative values indicate underestimation and positive values indicate overestimation. Ranking by WPE uses its absolute value, while the table retains the sign.
• MAE / MSE: lower is better; values are reported on the manuscript's evaluation scale. ρDS: a weighted correlation between stockout ratios and recovered demand; shown as a diagnostic, not direct ground-truth recovery accuracy.
• The future promotion, weather and calendar covariates are assumed available in the manuscript. Exact data splits, masks, preprocessing, aggregation and covariate-availability rules must be versioned with the evaluation code before external submissions are accepted.

Source manuscript: FreshRetailNet-LT: A Stockout-Annotated Censored Demand Dataset for Latent Demand Recovery and Forecasting in Fresh Retail, Tables II–III. No publication venue or acceptance status is asserted here.

Table II also reports the raw-sales diagnostic ρDS = −0.60, without WAPE or WPE; it is excluded from the ranked recovery entries.

Chronos2-Zeroshot is the forecasting component's label in the paper. It does not establish that an entire pipeline including a trained recovery model is zero-shot.

Community submissions: not yet open. A reproducible evaluation protocol and submission guide will be added after team review.