Experiment

We tested a prophecy

Not argued about it. Tested it — against volcanic, climate and sea-level records, under decision rules written down and cryptographically frozen before any data was downloaded. This is what that costs, what it produced, and the one result that settled it in a way no argument could have.

The claim

There is a widely circulated chronology holding that catastrophe arrives on a fixed schedule: a 138-year cycle running back through recorded history, with longer cycles of 792 and 394.5 years layered over it, anchored to a specific recent year. It comes with a ladder of dated nodes — 108 CE, then every 138 years, up to 1902 — and a mechanism: the sun darkens at each node.

It is a good claim to test, and that is rarer than it sounds. Most predictive systems are built so that nothing could contradict them. This one names years. Years can be checked against records that were compiled by people who had never heard of the theory.

The rule we set before looking

The entire value of what follows rests on the order in which it was done, so it is worth being precise.

Five protocols were written, each specifying the data source, the statistical test, the null model, and — critically — the number that would count as support. Each was hashed and the hash recorded before the corresponding dataset was downloaded. Those hashes still sit in the results directory alongside the outputs.

A prediction is only a prediction if it was written down before the answer was available. Everything else is a description.

This is the same discipline the app on this site imposes on a personal experiment, and the same one the angel numbers piece argues is missing when people count their own sightings. It is easy to state and genuinely uncomfortable to follow, because it removes every escape route in advance.

Test one — the anchor year

The chronology anchors on 2004. If that year is structurally special, it should show up as a breakpoint in long instrumental series.

Three independent series were tested, with the registered rule that the claim is confirmed only if at least two of the three select the anchor.

Result: none of the three did. Zero of three, where chance alone would be expected to deliver about half a win. Zero out of three is the single most likely outcome under the null. In the sea-level record — the stream the chronology leans on hardest — the anchor year performed worst of all six candidate anchors tested.

Tests two and three — the ladder against catastrophe

Next, the 138-year ladder itself, against catalogues of real disasters compiled by institutions with no stake in any of this.

Against the Smithsonian's global volcanism record, the ladder's hit rate was 0.333 against a null average of 0.449. Read that twice: the real ladder caught fewer catastrophes than ladders of the same shape shifted to random offsets. It was worse than arbitrary.

Against the twenty largest eruptions of the common era, one fell within three years of a node (p = 0.62). Against a fixed list of fourteen canonical mega-catastrophes, again one — the Black Death of 1347, near the 1350 node.

The misses are more informative than the hit, because they are exactly the events the theory most needs:

Test four — does the mechanism leave a fingerprint?

This is the test worth stealing, whatever you think of the subject.

The claim is not merely that bad things cluster on the ladder; it is that the sun darkens at each node. That is a physical assertion, and a darkened sun leaves a cooling signature in temperature reconstructions. So: build a detector, and before pointing it at the theory, prove the detector works.

The detector was run first on seven events where sun-blocking is independently established — 536, 1257, 1452, 1600, 1815, 1883, 1991.

The positive control is what makes this decisive. Without it, a null result is ambiguous — maybe the method is too blunt to see anything. With it, that defence is gone: the method sees real sun-darkening events clearly, and sees nothing at the predicted years.

The result that decided it

The chronology's own source list contains entries from two eras: a large set of dates before about 800 BCE, and a smaller set from the first three centuries CE.

They differ in one respect that matters enormously. The early dates are unverifiable — no independent year-precise record exists, so they can be assigned freely. The common-era dates fall in the Roman period, where external documentation constrains them.

Tested separately against the same lattice, with the same null model:

The pattern is present exactly where the dates cannot be checked, and inverts exactly where they can.

That is not a null result. It is a positive finding about how the chronology was built. No appeal to insufficient data survives it, because both halves come from the same document and were tested identically — the only variable is whether reality had a vote.

Two claims we checked by hand

Statistics can be argued with. Individual dated assertions can simply be looked up.

1626 — "North America resurfaced; the Great Lakes created." Champlain mapped Lake Huron in 1615–16, a decade earlier. The lakes are glacial and date back more than ten thousand years. The notable event of 1626 was the Wanggongchang explosion, a gunpowder magazine detonating in Beijing.

1764 — "Vulcan darkens the sun; half a million witnesses; royal records." The historical record contains individual observers reporting dots transiting the sun through telescopes. There is no darkening and there are no mass witnesses.

The verdict, and what it is not

Every registered endpoint returned null. After the first two phases the chronology's dated claims stood at 0-for-8 against reality under its own fixed rules, and no later phase changed the direction. The mechanism test came back with the wrong sign against a detector proven to work.

What this is not: proof that nothing patterned exists in history, or a claim that anyone involved was dishonest. Building a chronology like this takes real labour and real erudition. The failure here is not effort or intelligence. It is that the dates were fitted where fitting was possible, and the fit did not survive contact with records that were compiled independently.

It is also worth saying plainly: this result was allowed to come back negative. That is the only condition under which a positive result would have meant anything.

Why this is on a site about manifestation

Because it is the same structure, at a different scale.

A catastrophe chronology and a personal manifestation practice both rest on the same move — a pattern noticed after the fact, with the criterion for success adjustable in hindsight, and the misses never counted. Given a long enough list and a free hand with the dates, something always lines up. The 138-year ladder is that failure written large enough to be measured, and measuring it was possible only because the claim was specific enough to be wrong.

This is why the app here makes you write your criterion down and then locks it. Not to be strict, and not because your month is being graded — but because the version of you on day thirty is a worse witness than the version on day zero, and the only defence against that is a sentence that cannot be edited.

A prophecy that names years can be tested. A goal that names a checkable outcome can be tested. Everything else is a description of what already happened, wearing the clothes of a prediction.