Gold (spot): what the analysts said on 29 July 2026

Three independent readers look at the same day through deliberately different windows: one reads the debate, one ignores the debate and reads only the numbers, one reads only the news. If they all saw the same evidence they would simply agree with each other, which would tell us nothing. Today they genuinely disagreed.

The final step is arithmetic, not a fourth AI opinion. Version judge-2.0.0: each reachable reader's odds are averaged in log-odds space (so a confident answer is weighted properly rather than being flattened by a plain average), equally weighted because none had yet earned enough independent graded calls to count for more. The result is then pulled 20% back toward what this market normally does, keeping weak or incomplete evidence humble. Confidence calibration stayed off because the prior independent record had not yet earned it. Measured agreement between readers: 0.833 (1.0 would be identical); 3 of 3 answered.

Why the headline percentage isn't just the “lower” bar above. The bar shows all three outcomes, and one of them — barely moved — isn't a direction at all, it's a question about size. Reading “lower 42%” off that bar would suggest we mostly expect the other thing to happen, which isn't what the numbers say. So the call is stated the way the question is actually asked: when this market moves, which way does it go? Of the 68% of days it does move, we put 62% on lower. The chance it barely moves is reported on its own, and grading uses all three numbers untouched.

How strong is this call? Every debate commits a direction — we never sit one out. What varies is how much the call is worth, and we measure that against the market's own habit on the same scale: when this market moves, it goes lower 46% of the time historically, and we are saying 62% — a gap of 15.8 points. Calls at least 10 points clear of history count as strong, and the scoreboard reports those separately — so a record built on near-coin-flips can never read as skill.

Starting point before anyone argued: over 319 past stretches like this one, this market rose 38% of the time, fell 32%, and barely moved 30%.

Three independent readers look at the same day through deliberately different windows: one reads the debate, one ignores the debate and reads only the numbers, one reads only the news. If they all saw the same evidence they would simply agree with each other, which would tell us nothing. Today they genuinely disagreed.

The final step is arithmetic, not a fourth AI opinion. Version judge-2.0.0: each reachable reader's odds are averaged in log-odds space (so a confident answer is weighted properly rather than being flattened by a plain average), equally weighted because none had yet earned enough independent graded calls to count for more. The result is then pulled 20% back toward what this market normally does, keeping weak or incomplete evidence humble. Confidence calibration stayed off because the prior independent record had not yet earned it. Measured agreement between readers: 0.82 (1.0 would be identical); 3 of 3 answered.

Why the headline percentage isn't just the “lower” bar above. The bar shows all three outcomes, and one of them — barely moved — isn't a direction at all, it's a question about size. Reading “lower 44%” off that bar would suggest we mostly expect the other thing to happen, which isn't what the numbers say. So the call is stated the way the question is actually asked: when this market moves, which way does it go? Of the 71% of days it does move, we put 62% on lower. The chance it barely moves is reported on its own, and grading uses all three numbers untouched.

How strong is this call? Every debate commits a direction — we never sit one out. What varies is how much the call is worth, and we measure that against the market's own habit on the same scale: when this market moves, it goes lower 43% of the time historically, and we are saying 62% — a gap of 18.9 points. Calls at least 10 points clear of history count as strong, and the scoreboard reports those separately — so a record built on near-coin-flips can never read as skill.

Starting point before anyone argued: over 315 past stretches like this one, this market rose 40% of the time, fell 31%, and barely moved 29%.