Method

A traditional East Asian calendar turns a birth date into a cycle of years. Rules of our own read each year of that cycle and issue one reading a year for every leader. The rules stay in-house; what is published is the reading and the record behind it.

The calendar also wants the hour of birth, which the public record almost never gives. So every year is read for each possible birth hour, thirteen in all, and a reading is issued only where at least ten of them agree. Where they do not, there is no reading.

A year reads Strong, Firm, Even, Soft or Weak. Signals lift a year and drags lower it; the level is their balance. Each reading shows how many signals and drags stacked, never which.

A season's readings are published on 1 December of the year before and sealed before the year begins. Nothing is changed afterwards. If the rules change, the next season carries a new version and the seasons already sealed stand as published.

The year is the calendar year, January to December, and the stock is the company the leader runs. A year counts when the leader held the role for at least six months of it, once per leader. Two questions are asked of every leader-year: did the stock rise, meaning its return for the year was above zero; and did it beat the field, meaning its return was above the median return of the leaders in role that same year.

Beating the field compares leaders inside the same year, so a year in which everything rose separates no one. Where the S&P 500 appears it is the market year for context, not the benchmark.

260 leaders and 3,482 leader-years, every year in the role from 1995 to 2025, read with the rules now fixed:

These are development results: the rules were chosen on these years. How that was kept honest is the next section. The forward record starts with the 2027 season. The 2026 readings were computed in October 2026, so 2026 is a trial run.

Rules chosen on a history can fit its noise. Three things were done about that before any rule was kept, and one thing cannot be done until the seasons run.

The rules came first. Most of the counts are rules the tradition already states, read as written; the rest measure a year against the leader's own history. Each was tested as stated. Nothing in the reading was found by searching the data for what worked.

The other-leader check. The same years were read against another leader's chart, 200 times over. For each count, the share of leaders who rise more often in their flagged years has to sit above the band those re-pairings produce, or below it for a drag.

The year-shift check. The flags were moved one, two and three years either way. All six shifted versions have to read worse than the true one.

Half-samples. The leaders were split in two at random, 40 times, and each half read on its own: 80 looks at the same question. A finding that holds in nearly all of them is not carried by a few leaders.

One count each. Weights were tried and lost out of sample, as the table below records. So every count is one vote, and the level is the balance.

What was searched, and thrown away. The searches that were run on the same history, and what came of them. None of it is in the reading.

The reading itself, the levels the site publishes, through those checks:

And the nine counts behind it, labelled S1 to S6 for the signals and D1 to D3 for the drags, in an order that says nothing about which rule is which. 3 of the nine do not clear every check alone; the notes say why each is kept.

These checks rule out particular alternative explanations. They do not turn a history into a forward test. That begins with the 2027 season.

Three further rules are watched and not counted. One passes both checks but may be measuring something simpler than the calendar, so what it shows is not clear. Two fire too rarely to judge, and on the beat-the-field measure they add nothing. They will be sealed with the 2027 readings, unseen, so that season can judge them. Until then they appear nowhere on a leader's page.

Birth dates. Of the 325 leader-company records on the site, 205 rest on a filing that states an age, which fixes the year and not the day; 118 have no source recorded yet; 2 have a published source for the day itself; 0 are withheld because a source rules the held date out. Every date and what it rests on is in the ledger.

Birth hours. Almost none are public, which is why a reading needs ten of thirteen hours to agree, and why some leader-years have no reading at all.

The roster. Founders and chief executives of listed companies, mostly large and mostly American, with at least a full year in the role. Companies that did not survive to be listed are not in it, and the leaders with the longest records are the ones whose companies lasted.

Nothing here is investment advice. A reading describes a leader's year as read from a birth date; it is not a forecast of a price.