Blog
Core Macro begins, twelve days late
The Core Macro cohort is the part of this site that is meant to carry the evidence. Its questions are not chosen: every month the same five variables are forecast for the following month-end, whatever they look like. The first instance was due on 1 October at 12:00 UTC. It was published on 13 October. Its cutoff is the actual publication time, not the due date, and the coverage table on the track record shows it as a late instance rather than an on-time one.
The five forecasts (target: end of November 2026)
| Question | Median | Central 50% | Benchmark (persistence) |
|---|---|---|---|
| Brent spot, EIA | 109.4 $/bbl | 94.6 – 125.8 | 125.44 |
| US 10-year par yield | 4.97% | 4.70 – 5.27 | 5.24 |
| EUR/USD, ECB reference rate | 1.1147 | 1.0941 – 1.1399 | 1.1181 |
| Gold, LBMA PM | 4,083 $/oz | 3,830 – 4,354 | 4,190.6 |
| Fed decision, 28 October | hold 79.6% · hike 16.5% · cut 3.9% | — | — |
How the numbers were made
Each forecast starts from what markets already imply, then lets the engine update it. The entry prior is a market observable with its own uncertainty: futures that settle near the target date, implied volatilities, and fed funds futures for the meeting. It is recorded before the cutoff and then everything downstream is produced by the engine. If the engine merely copied the market, it would add nothing. If it moves away from the market, the scores will say whether that was skill or noise.
The benchmarks are deliberately simple: the last official value available at the cutoff. Beating persistence over seven weeks is not a high bar for a well-calibrated distribution, but it is the bar that the first resolutions on this site failed to clear three times out of four.
Three details worth knowing
Brent spot is not Brent futures. The question resolves on the EIA's Brent spot series, which since mid-September has traded 11 to 33 dollars above the futures. That gap is why the persistence benchmark (125.44) sits so far above the forecast's median. Whether the gap closes by the end of November is most of what this question is about.
The Fed question closes at the statement. The cohort's rule says the Fed question resolves on the first decision after publication, so it now closes at the 28 October statement (18:00 UTC), not at the end of November. The schedule had been ignoring that rule; it was fixed before this instance was published.
Gold resolves on a licensed benchmark. The LBMA's historical price tables are now behind a licence. The day's price is still shown on the LBMA website, so the November value will be recorded, with evidence, on the day it is published. If it cannot be documented within fifteen days, the question is void, as its rule already says. Its benchmark uses the free COMEX futures close instead of the LBMA price, for the same reason.