Home › Track record
A forecast that cannot be marked wrong is not a forecast. These are the rules we published first and have not changed since, plus the arithmetic behind every number on this site.
That third rule is deliberately unkind to us. A three-scenario forecast that covers most of the plausible range is easy to get “right”, which is exactly why we also publish the probability weight on each scenario and a written falsification condition — see below.
Since the first report on 30 August 2026:
Three of the fifty-nine are not in the denominator: today’s open call, the weekly call currently open (graded on Friday 6 November), and the 20 September analytical special, which carried no price scenario of its own. Every other page made a call and is counted, weekly calls included.
The accuracy figure is therefore correct calls divided by graded calls, and the newest open forecast is always excluded until it closes.
10 September 2026. We put the likely range too high and the close finished outside all three published scenarios.
Week of 21–25 September. Against a 16-of-16 bullish Wall Street survey we called consolidation at $4,300–$4,420. The direction was right (gold fell $93) but the close, $4,285, was $15 below our range, and the criterion we had pre-registered said that counts against us.
Week of 28 September–2 October. We called $4,150–$4,420 across three scenarios and tied a break to US data. The break came on Monday through oil and Hormuz, and the week closed at $4,140, ten dollars below our lowest number. A fresh reading published on Tuesday beside the original (not in place of it) landed correctly; the original did not, and the original is what counts.
Every one of these pages is still on the site, unedited, with the verdict published the following morning above the next forecast.
On 26 September we wrote that the $15 bucket miss of the 25 September weekly call “counts against us”. It did not enter the tally that day; the running figure kept showing one miss and 95–96%. On 3 October, while grading the second weekly miss, we found the gap and corrected it in public: both weekly calls now count as wrong, and the accuracy figure fell from 96% to 90% by our own hand.
An earlier correction ran the other way: for several weeks our social posts summarised the record as “two wrong calls, 10 and 11 September”, when the 11 September page only carried the red mark because grading publishes a day late. The number has to be right in both directions.
On 18 September Kitco’s Wall Street survey came back 16 bullish out of 16. We published the opposite view the same weekend: no rally, consolidation at $4,300–$4,420, with a 20% pullback scenario. Gold fell $93 that week. On direction, the consensus was wrong and we were right; on the bucket, we missed by $15 and counted it as a miss. A week later the same survey was split three ways, and two weeks later it was 6 bearish, 3 bullish, 4 neutral.
The rules we changed after two weekly misses in a row are on the 3 October page: a weekly range of at least $400 while daily moves exceed $50, oil and Hormuz as a mandatory channel, trend continuation ranked first when the market is trending on one open headline, a confirming number beside every scenario, and every published call entered in the tally the day it is graded.
Every dated report since 30 August 2026 is public. The reports are currently written in full in Arabic; English editions are being published from here forward, and the archive is being translated. Every figure quoted on this page appears on the Arabic page carrying the date it was published on.
Documented disagreements English forecast hub Full Arabic archive