Our forecast for the next CPI print, locked before the release with a dated, immutable receipt — value, method, and every competing forecaster (our in-house model, the Cleveland Fed, the Street consensus, Kalshi markets) on the same tape. A live countdown runs to the 8:30 AM ET release.
The forecast is an error-weighted ensemble: each member — the in-house bottom-up model, the Cleveland Fed nowcast and the Street consensus — is weighted by its own graded record over the same recent prints, validated in a walk-forward backtest built from the data as it was published at the time. After each print, everyone gets graded on the scoreboard — wins, misses, and mean absolute error, in public.
Today's call for the October 2026 CPI, released November 10, 2026 at 8:30 AM ET, is 3.9% (3.86% unrounded) year over year; the in-house model alone reads 3.97%. Until the evening before the release this is the current forecast: it moves as the inputs move, and each day's value is on the record. At 6:30 PM ET the evening before the print it is locked into a receipt — a JSON file with its SHA-256 hash, published before the number lands — and that locked value is the one graded. The standing call is also sealed 30, 14, 7 and 1 days ahead, so the record shows how it moved as the release approached. Sealing began with the September 2026 print, 14 days before its release.
How to read it: the forecast answers a different question from the Gauge. The Gauge measures prices today; the forecast is our best estimate of the number the BLS will print for a month that has already happened, which is why it can sit above or below today's reading. A forecast is only worth what its record says it is worth: so far 3 live calls have been graded, at a mean absolute error of 0.21pp — too few to rank anyone on, and the longer backtest sits beside it on the scoreboard, labeled as a backtest.