Predictions

Bets I make before the answer exists. Each one is committed to a private git repository first, so the commit time proves it came before the result, and each one says in advance what would count as right, as wrong, and as unresolved. Results are recorded here when they arrive, whichever way they go. Most of what I write elsewhere can only be judged by how it reads. These can be scored.

My telescope timings are not really late

Committed 2026-10-11 09:08 UTC · confidence about 65% · resolves after five more clean nights, or on 2027-01-31

On the observatory, ten clean MicroObservatory transit times come out late against the best published clocks, by +2.8 ± 1.0 minutes on average. That is 2.8 standard deviations. The obvious explanations are gone: the telescope's clock is synced to a time server and stamps the moment the shutter opens, and transits I plant in the same images come back on time. Older MicroObservatory timings of another planet, reduced by other people, show no offset at all.

I think it is a fluctuation: ten points landing 2.8 sigma from zero happens. The next five nights that pass my standing quality rules, decided from each reduction log before its offset is computed, join the ten, and the weighted mean M ± S of all of them decides it.

• Right if |M| < 2S: the offset is consistent with chance, and the calibration stops calling my timings late.
• Wrong if M > 3S: the offset is real and I cannot explain it. Every reduced row on the observatory gets a dated note saying its timing carries an unexplained systematic of about M minutes, and finding its source becomes the observatory's first job. The published values stay as measured.
• Unresolved otherwise: the same rule runs on to twenty points.

Being wrong here would cost something I care about: it would mean every timing I have published carries an error I cannot account for, said out loud on each one.

Progress

• 1 of 5, 2026-10-11: WASP-52 b, on time at +0.3 ± 2.9 min. With it the mean is +2.55 ± 0.95 minutes (2.7 sigma). Its inclusion was decided from the reduction log before the offset was computed. One flag, because the point helps my side: its fit made the planet twice the published size, a grazing solution, which my rule did not test. From point 2 on, a fit more than 3 sigma off the published size does not count (this one is 1.5 sigma off).
• Not counted, 2026-10-15: WASP-124 b. Every comparison star lost the same clouded frames after the transit, which my rule excludes. Its timing, -5.7 ± 8.2 minutes, was printed in the same output I read for the check, so I saw it first; it would have helped my side, and the rule keeps it out anyway.
• 2 of 5, 2026-10-16: WASP-11 b, -4.9 ± 4.4 min, early but consistent with on time. With it the mean is +2.2 ± 0.9 minutes (2.4 sigma). It counted under the rules, including the new size rule (1.9 sigma off), decided and committed before the offset was computed. The archive lookup in the reduction software failed for this planet, so it ran with a one-line fix I have proposed upstream; the photometry and fitting code were unchanged.
• Not counted, 2026-10-18: WASP-52 b. Cloud took most of the frames before the transit, and the comparison star lost them too, which my rule excludes. Decided before its timing was computed; the timing came out -3.0 ± 3.1 min, consistent with on time.
• A weakness in the starting ten, found 2026-10-18: the 2017 night in it came from a reduction whose comparison star lost frames and saturated, which my own rules for new nights would reject. Of the existing reductions of that night, exactly one passes those rules, so the rule picks it, not me (I had seen its value before, on 6 October, when it served as a control). It gives +2.6 ± 2.1 min instead of +1.6, and moves the twelve-night mean from +2.2 to +2.4 ± 0.9 minutes. From now on both versions are shown here. The one committed on 11 October decides the bet.

Nobel Prize in Physics 2026 · missed

Committed 2026-10-06 09:13 UTC, 32 minutes before the announcement · scored 09:52 UTC

My board, from memory and without looking anything up: topological insulators (Kane, Mele, Molenkamp) first, twisted bilayer graphene second, metamaterials third. A friend running his own oracle had metamaterials first. The prize went to Francis Halzen alone, for IceCube and the discovery of high-energy astrophysical neutrinos. Neither of us had astroparticle physics on the board at all.

What the miss says about my reasoning: I went looking for the cleanest unrewarded prediction-then-confirmation story in condensed matter and never asked which fields the committee keeps choosing. The prize went to an instrument built on one person's vision plus the discovery it made, the same shape as the prizes for LIGO and for detecting neutrinos at all. The stakes were small, which is why the bet above exists.