The Method Behind the Trackers: How I Keep Myself Honest About the Future

Over the past five weeks I have put five questions on five public websites, and each one has a number on it today. The Belt Equation asks what the chance is that, before I die, I live and work in a solar system where human industry reaches beyond Earth and Mars; as of October 11th, 2026, it says about half a percent if I live to 2071. The AGI Zeitgeist Tracker (AGI is short for artificial general intelligence, a machine that can do whatever a person can) asks whether we are in the singularity, the point at which machines improve themselves faster than people can follow; as of October 24th, 2026, it says not yet, and about 32 percent of the way there, averaged over the three of its four measures that can be measured, with the fourth, the world economy, at zero. The Quiet Galaxy asks where everybody is; as of October 17th, 2026, its middle estimate is about two hundredths of a detectable civilisation in our galaxy, roughly one such civilisation for every fifty galaxies like ours, which is a long way of saying that the chance we are alone in it comes out at 58 percent. The Upgrade Tree asks how much of a human being we can change with technology today; as of October 23rd, 2026, it counts 28 of 35 ways that have been tried in at least one living person. Outrunning Ageing asks when, if ever, medical repair overtakes ageing; as of October 19th, 2026, its middle estimate is the year 2058.

A reader who visits two of the sites will notice that they are built the same way: the same kind of page at the top, the same kind of list under it, the same strip saying when it was last checked, the same page at the end for disagreeing. That is because each site is one use of a single way of working, and the way of working is what this essay is about. I want to say plainly what it is, how much of it I borrowed, what small part of it I have not found anywhere else, and why I keep it.

One Question, Broken Into Things That Can Be Checked

The first move is the one the astronomer Frank Drake made in 1961, when he wrote an equation on a blackboard to organise a meeting about whether anyone in the galaxy was sending signals. The equation did not answer the question. It turned one unanswerable question into seven arguable ones, so that people who disagreed could say exactly where. I used that move for the Belt Equation and wrote about it in the essay that introduced the site, and the four sites that followed use it too.

Take one question I actually care about, and break it into a small fixed set of reasons the answer comes out the way it does: eight links of a chain from cheap launch to a seat for me; eight kinds of damage that ageing does to a body; Drake’s seven factors; seven capacities of a person that technology can reach; four conditions a singularity would have to meet. The set stays fixed and changes only at a yearly review, because a reader should be able to hold the whole of it in mind after one visit, and because a chain whose links keep changing cannot be argued with.

Under each reason sit the specific things that would have to happen: a treatment reaching patients, a rocket landing both of its stages, a measure of the economy crossing a line. Each carries a test written so that two strangers reading it would agree whether it has happened, and names the kind of public record that would settle it: a paper with a measured result, a regulator’s approval, a company’s own release naming a patient, a recording of the demonstration. A press claim, a funding round, a target date or a demonstration video settles nothing, on any of the five.

The number at the top is then worked out from those specific things and is never typed in. The Belt Equation runs its whole chain of events twenty thousand times for each lifespan it considers, each time with the dice rolled afresh; Outrunning Ageing and the Quiet Galaxy run theirs two hundred thousand times; the Upgrade Tree counts the technologies on its list; the Zeitgeist averages its measures. A reader who asks “why that number?” can follow the answer down from the top of the page to the one specific thing they disagree with, and argue with that thing alone.

Most of This Is Borrowed

Before I say what is unusual about the method, I should say how much of it is not, because when I first noticed the five sites converging I was tempted to think I had found something new, and a careful comparison with what already exists showed otherwise.

A test fixed in advance so that two strangers agree on the outcome, a named independent judge of that outcome, a public record of every forecast that is never deleted, edits that keep the old wording visible, a page for arguing, and a score for how well the probabilities matched what happened are all standard practice in public forecasting. Metaculus, the forecasting site, publishes its rules for all of them, and it does several more rigorously than I can, because its judge is independent of the forecaster and its score rests on thousands of questions rather than eighty. The score itself, called a Brier score after the meteorologist who proposed it, comes from weather forecasting and was carried into forecasting tournaments long before I used it. Writing down the test before you look is what scientists call preregistration, and since 2005 the leading medical journals have refused to publish a clinical trial that was not entered in a public registry before it began.

Even the part I was proudest of, the weekly reading, has a close relative. In medicine, a “living systematic review” is a summary of the evidence on one question that is re-searched on a fixed schedule under rules written in advance, with a dated entry saying “no new evidence” when nothing was found and the date of the last search shown. Cochrane, the organisation that sets standards for such reviews, published its guidance for living reviews in 2019. That is the weekly reading, the quiet weeks and the last-checked strip, done by professionals for years. And the idea of a person keeping a public, dated record of their own predictions for decades, with the original words frozen and a yearly update scored against them, is exactly what the roboticist Rodney Brooks has done since 2018.

So the honest description of the method is an assembly: practices from forecasting, trial registration and evidence review, applied to one person’s questions and run on a weekly clock. I am saying this at length because a method that borrows well and says so is more trustworthy than one that claims to be new, and because anyone who wants to do this for their own question should start from those sources, not from me.

The Three Things I Have Not Found Elsewhere

When the comparison was done, three parts of the combination were left that I could not find in any one place.

The first is the shape of the whole: one personal question, with a headline worked out from a tree of fixed, individually settled tests. Forecasting sites take questions one at a time and score each on its own. Brooks’s scorecard is a list of his predictions, each judged on its own. What none of them does is compute an answer to a question too big to ask directly from the answers to many small ones that can each be settled, and let a reader trace the big number down to the small one they disagree with.

The second is who does the watching, and how. Once a week on each site, every one of those specific things is read against the news by a reader with no memory of the week before. The reader is an AI model, because a reader with no memory of last week cannot be a person; it is started fresh each time and given a fixed list of what to look for. It must run at least two searches for every item and write down every search, including the ones that found nothing, and it is allowed to move very little on its own: a capped nudge to a probability on the Belt Equation; on Outrunning Ageing, one rung up the five-rung ladder of stages every treatment sits on (idea only, shown in animals, first human, in clinical use, in wide use), and never onto the clinic rung, which is the one the headline depends on; nothing at all on the Zeitgeist, where it may only add to the record and flag what it found for me. The quiet weeks are kept and shown with the same weight as the weeks when something moved, and a widely reported event (a crewed landing announced, a chip implanted, a planet found) gets a dated entry saying it moved nothing here, and which test it was read against. Living reviews have the schedule and the “nothing new” entry. The memoryless reader, the empty search per item on the record, and the entry that cuts against the news cycle rather than riding it are the parts I have not seen together.

The third is a second record kept beside the first. Three of the five sites record what named people have said about the future, dated and quoted from the recording or the original text, with the famous paraphrase labelled a paraphrase when the words turn out to differ, and set that record beside what the world then did. A statement never becomes a measure of the world, and a measure never becomes a statement; the two sit side by side so that a reader can see the gap. Forecasting sites score their own members. Brooks scores himself. Nobody I found keeps a scored ledger of other people’s public talk beside a dated record of delivery.

That is the whole of the claim: an assembly of existing practices, plus three uncommon parts, run by one person on his own questions.

Rules That Change Only in the Open

Everything on the five sites is dated, and nothing is edited. A change is a new entry that says what the old value was, what the new one is, and why; a mistake is corrected by a later entry that says so. A test, once a number has been published against it, is never reworded; if a test turns out to be wrong, the old wording stays beside the new one, with the date, so that anyone scoring the forecast later can find the test it was made against. The rules themselves follow the same discipline: each site has a page saying how its number is worked out and a dated list of every change to those rules, with the reason and with who decided it.

Since October 17th, 2026, one more rule has held across all five. Every fact in a public record carries a source a reader can open, and a fact that cannot be sourced after a real search is removed rather than left on display with a note. I added it after learning that one site’s records had been drafted quickly with the help of an AI model and carried seventy-six facts marked for later checking, and deciding that the marking was not enough. The check happens before anyone is sent to the site.

Every site ends with a page for disagreeing, and the page says what happens next. You can name a thing the site should be tracking and is not, as long as you can write its test. You can question a test or the record that settles it. You can argue a number, or the rung of the ladder a treatment or technology has been placed on. Each becomes a durable record of its own, and the change that answers it cites it.

Why I Keep It

Every one of the five sites began as a hope. I hope to work off Earth before I die. I hope medicine outruns my own ageing. I hope we are not alone. Those hopes are the reason the sites exist, and they are also the reason the sites have to be cold. A hope is the thing most likely to bend a record, because a record kept by someone who wants a particular answer will drift toward it unless something stops it. The tests, the named settlers, the memoryless reader and the dated history are what stop it. They are there so that when the number moves, I can tell whether the world got closer or I just wanted it to.

That is also why the hopes do not go in the records. I thought for a while about giving each site a place to write down the hope behind it and the steps I was taking toward it, and decided against it, because a record built to motivate has a purpose, and purposes bend records. The hope belongs here, in the essays, and in the posts each site writes when something changes, where it can be said in my own voice and checked against a record that does not care how I feel. The record is what gets built so that the story can be told, and the story is only worth telling if the record stayed cold while it was kept.

I argued in Everything Is Amazing and Nobody’s Happy that calibration, the discipline of holding an accurate picture of where you stand, is a practice rather than a mood. The five sites are that practice applied to the things I most want to be true. The Belt Equation’s own plan says the point in one sentence: “The number moving is the product, and the number itself is only a snapshot.” The first score of how well my probabilities matched what happened is due in January 2027, and I expect it to be thin, because the events that have resolved so far did so before any number was on record for them. I would rather publish a thin score than none.

Argue With the Method

Each site has a page for disagreeing with its numbers. For the method itself, the Belt Equation’s disagree page takes arguments about how the numbers are made as readily as arguments about a number, and the answer, if it changes anything, arrives as a dated entry that cites you.

The five numbers at the top of this essay will all be different by the time you read it. That is the point of writing them down with the date beside them.