Before each CPI print we put a forecast on the record and grade it after, misses included. Our headline call is the Nowflation CPI Forecast, an error-weighted blend of our in-house model, the Cleveland Fed nowcast and the Street consensus, each weighted by its graded track record. The Street consensus and the Cleveland Fed have been graded on 45 prints since December 2022. Nowflation's own live record is 3 pre-registered calls graded since June 2026 at 0.212pp mean absolute error. Everything before that is a vintage-true backtest. Full scope: the in-house model's walk-forward covers 89 prints (January 2019 to August 2026) at 0.129pp mean absolute error vs 0.328pp for a naive baseline. League-table scope: 39 of those prints fall inside the benchmark window before the first live-graded print, at 0.094pp. Graded values are the scoreboard rows frozen at grading time and are never rewritten when a source revises after the release; Nowflation's live calls are graded on the value in the receipt locked before the release, with the frozen row shown beside it where the two differ.
A call counts as live only when its receipt was written before the print: one JSON file per print month at /forecasts/index.json, since August 2026 locked the evening before release, SHA-256 hashed and never rewritten, then graded on the receipt's value against the first release. The first receipt, June 2026, was locked a week before its release without a hash, and a note in it (not the forecast) was corrected an hour after it was written. The standing call for the October 2026 CPI is 3.9% (3.86% unrounded; in-house model 3.97%, Cleveland Fed 3.77%), provisional until the lock the evening before the release.
| Status | Written (ET) | Nowflation Forecast | Actual | Error | Live? | Receipt | |
|---|---|---|---|---|---|---|---|
| 2026-10 | provisional | October 18, 2026 at 8:05 AM ET | 3.86% | pending | — | pending | JSON |
| 2026-09 | locked before the print | October 13, 2026 at 6:30 PM ET | 3.56% | 3.7% | -0.140pp | live | JSON |
| 2026-08 | locked before the print | September 10, 2026 at 6:30 PM ET | 3.34% | 3.4% | -0.060pp | live | JSON |
| 2026-07 | frozen after the print | September 1, 2026 at 5:23 PM ET | 3.42% | 3.4% | +0.021pp | not live | JSON |
| 2026-06 | locked before the print | July 7, 2026 11:18 AM ET | 3.94% | 3.5% | +0.435pp | live | JSON |
July 2026 row, not live: Receipt written September 1, 2026 at 5:23 PM ET, after the July 2026 CPI was released (August 12, 2026). The values are the scoreboard rows frozen from the engine's daily record, but the call was not on the public record before the print, so this row is not counted as a live call.
August 2026 receipt, note: The Street consensus is dated September 11, 2026, the release day, after the lock (September 10, 2026 at 6:30 PM ET): when this receipt was written, the engine dated the economic-calendar consensus to the release it describes. The receipt holds the consensus value the engine had when it wrote the receipt.
The standing call is also sealed 30, 14, 7 and 1 days before each release, one file per horizon, never rewritten. Sealing began with the September 2026 print, 14 days before its release. Each file's SHA-256 goes to OpenTimestamps when it is written (the proof anchors in the Bitcoin blockchain; verify it at opentimestamps.org with the file and its .ots), and the Wayback Machine captures the file once it is live.
June 2026 receipt vs graded row: the receipt file records 3.935% and the scoreboard row frozen at grading time records 3.895%. The live record grades the receipt, the number published before the print. The frozen row holds the engine's day-before value, which no receipt published before the release; it stays on the scoreboard as graded and is not the live call.
| Forecaster | Live prints | Live MAE (pp) | Live bias (pp) | Closest-call % | Backtest prints | Backtest MAE (pp) | Combined † |
|---|---|---|---|---|---|---|---|
| Street Consensus | 45 | 0.087 | +0.020 | 60% | — | — | 0.087pp · 45 |
| Cleveland Fed | 45 | 0.104 | +0.034 | 38% | — | — | 0.104pp · 45 |
| Kalshi Market | 4 | 0.112 | +0.043 | 25% | — | — | 0.112pp · 4 |
| Nowflation Forecast | 3 | 0.212 | +0.078 | 0% | — | — | 0.212pp · 3 |
| Nowflation Model (in-house) | 3 | 0.312 | +0.038 | 0% | 39 | 0.094 | 0.109pp · 42 |
Live = calls on the public record before the print (for the benchmarks, every print they were scored on; for Nowflation, the receipts locked before their release). Backtest = the in-house model's vintage-true walk-forward inside the benchmark window (the league-table scope, 39 prints at 0.094pp; the full walk-forward is 89 prints at 0.129pp). A Nowflation call that was graded but whose receipt was written after the print (July 2026) is in neither and in no average. Closest-call % counts only the calls on the record before each print. † Combined mixes both bases and is never used for the rank.
headline YoY: in-house model 43%, Cleveland Fed 57% (as of 2026-10-18). Weights are inverse mean absolute error over the prints every rated member has graded; a member with too few graded prints carries zero weight. Where the in-house model has zero weight, the Nowflation Forecast is a blend of the benchmarks alone.
| Actual | Street Consensus | Cleveland Fed | Kalshi Market | Nowflation Forecast | Nowflation Model (in-house) | |
|---|---|---|---|---|---|---|
| 2026-09 | 3.7% | 3.6% | 3.6% | 3.649% | 3.56% | 3.46% |
| 2026-08 | 3.4% | 3.4% | 3.38% | 3.366% | 3.34% | 3.229% |
| 2026-07 | 3.4% | 3.4% | 3.42% | 3.347% | 3.421% (receipt written after the print; not a live call) | 3.46% (receipt written after the print; not a live call) |
| 2026-06 | 3.5% | 3.8% | 3.92% | 3.809% | 3.935% (receipt; the row frozen at grading reads 3.895%) | 4.025% |
| 2026-05 | 4.2% | 4.2% | 4.18% | — | — | 4.3% |
| 2026-04 | 3.8% | 3.7% | 3.565% | — | — | 3.633% |
| 2026-03 | 3.3% | 3.3% | 3.252% | — | — | 3.496% |
| 2026-02 | 2.4% | 2.4% | 2.406% | — | — | 2.472% |
Each print month links to its own print-day page with every forecaster's call, the miss sizes and the receipt. All 13 print-day pages are listed on the analysis page.