Prediction methodology

How the release-timing model works

This page explains the forecasting logic behind the May 2026 release estimate and the official source data used to build it.

BackForecast page

Current model

Backtested business-day forecast

  • Each forecast uses the latest 8 official releases from the same release month only.
  • Candidate dates are business days only, so Saturdays and Sundays cannot rank.
  • The main signal is a business-day-index kernel with sigma 1.75.
  • Blend weights: 70% business-day index, 20% ordinal weekday pattern, 10% recency.
  • Conditioning removes already-passed business days in the active release month and renormalizes the remaining probability mass.

Backtest

Rolling holdout results

  • 24 eligible same-month holdout targets were used in the rolling evaluation.
  • Top-3 hit rate: 42%.
  • Top-1 hit rate: 13%.
  • Mean absolute business-day error: 2.8.
  • Mean log loss: 2.61.

Official sources

Government data only

Release timestamps for May 2026 come from the official Department of State Visa Bulletin archive and each issue's official PDF metadata. Filing-chart companion links come from USCIS.