Accuracy

Inspect the evidence. Understand the limits.

Published forecasts and historical benchmarks answer different questions. Start with the record, then check the population, instrument and scoring rules behind each comparison.

The Lewsearch Report

Forecasts, grades, pending calls and corrections. The original grades remain visible.

50-state benchmark

A calibrated retrospective comparison with published poll toplines. Read the full result alongside the selected best 80%.

Methodology

How the panel works, what has been measured and which capabilities need separate human evidence.

Blind Challenge

Compare an agreed instrument with a completed human study whose results you withhold. Your first qualified U.S. consumer challenge is free.

The public record

  • 3 wins
  • 2 partials
  • 0 misses
  • 3 pending

Both partials were first graded as misses under their frozen rules: Issue 06 on September 23 and Issue 07 on October 17. We added the Partial category on October 18, 2026 and kept the original grades on each Issue. Category policy.

All 8 Issues and their grades
Issue 08
SCE October mean probability of higher U.S. unemployment in 12 months

Forecast 47.33%. Official Pending · November 9

Awaiting grade
Issue 07Partial
Issue 06Partial

Evaluate the work you need to do.

A historical topline match does not establish accuracy for a new subgroup, message ranking or simulated conversation. Agree on a relevant human comparison before relying on a new use.

Book a demo