Public forecast 011 · permanent record
Will UK CPI annual inflation for August 2026, as first published, be between 2.5% and 3.3% inclusive?
RESOLVED — scored against the source named before the outcome was known.
- Question
- Will UK CPI annual inflation for August 2026, as first published, be between 2.5% and 3.3% inclusive?
- Probability band
- 84–92% — issued 2026-08-30, data cut-off 2026-08-30
- Frozen criterion
- Resolves TRUE if the CPI 12-month rate for August 2026, in the figure first published by the ONS in the Consumer price inflation bulletin scheduled for 16 September 2026, is greater than or equal to 2.5% and less than or equal to 3.3%. Resolves FALSE otherwise. Later revisions do not change the resolution: this is scored against the number as first published. If the bulletin is postponed beyond 30 September 2026 the forecast is marked withdrawn with that reason.
- Resolves by
- 2026-09-16
- Resolution source (named in advance)
- https://www.ons.gov.uk/releases/consumerpriceinflationukaugust2026
- Model
- judgement — reasoning recorded at issue, unchanged since
- Ledger id
lxf_external_011- Outcome
- TRUE — 3.1%
- Evidence, as first published
- The Consumer Prices Index (CPI) rose by 3.1% in the 12 months to August 2026 (source, published 2026-09-16)
- Brier score
- 0.0144 — band midpoint 0.88. 0 is perfect, 0.25 is what always saying 50% scores.
Reasoning recorded at issue — unchanged since, including where it turned out to be wrong: The annual CPI rate in 2026 has moved 2.8% (May) to 2.6% (June) to 2.9% (July) — one-month changes of −0.2pp and +0.3pp, the July move larger than usual on airfares, food and base effects (all figures as first published and resolved on this register, lxf_external_002/003). This band allows 0.4pp of movement in either direction from the July print, the same window that resolved TRUE for July (lxf_external_002, band 85–93%, Brier 0.0121). The band here is set slightly lower than that one because July just demonstrated a +0.3pp single-month move and August 2025 base effects were not analysed before issue — the honest price of an un-analysed component is a wider admission of being wrong.
If ten forecasts were issued at this band, the band claims roughly 8 or 9 of them should come true. Calibration — not certainty — is what the register is scored on.
What would prove this system wrong: a forecast on this register edited after issue, a resolution scored against anything but the source named in advance, or an accuracy claim made before ten outcomes have resolved in a domain. Every one is checkable from the raw ledger — the full falsification list.