How Faz works

Faz is our own intelligence, and it is a referee, not a forecaster. People take positions on each question, when will it happen, and defend them with evidence. Faz reads everything on the record, from official targets, research, data and polls to the arguments of the people on this site and what it finds on the wider web, and rules: a date with a range, or no date yet. Like a referee, it gives its reasons: the evidence the ruling rests on and what would change it. It keeps a public record of how close its rulings came to what really happened. For now its rulings are made by the Fazoh team with AI assistance, checked against the rules. Ruling well on the future is hard. Faz is built to get better at it.

1. It reads the debate

Everything people post under a claim, from answers and rebuttals to quick takes, goes into Faz's ruling. It keeps its notes and reads only what is new since last time, so a fresh argument can move the ruling. Arguments with sources count for more than opinions, and a rebuttal that lands counts for more than a pile of agreement.

2. It reads the evidence: FazBot

Every source cited under a claim goes into Faz's evidence file. Faz sends FazBot, its reader, to open those pages and check what they actually say, so a ruling rests on the documents themselves and not on summaries of them. Faz cites only pages it has read. It does not search the open web at this stage; when the evidence on a claim is thin, it says so and keeps its range wide. That is one more reason a well-sourced answer from a member can move the ruling.

3. How a ruling is made

No judgment is taken at face value. Faz works through the same steps on every question:

• The reference class. How long have comparable things taken, and how often did announced dates hold? Crewed programmes, new launch vehicles, drug approvals and treaties each have a history, and announced dates tend to slip.
• The critical path. The event cannot come before its slowest unmet step. Faz checks the questions on the road to it, and a date earlier than the chain allows is wrong.
• Both sides. It weighs the debate, it will happen and it will not, by the quality of its sources and reasoning, never by how many people agree. Where the evidence disagrees, it says so rather than splitting the difference quietly.
• The wider web. What has changed in the world since the last ruling: a test that flew, a programme cancelled, a law passed, a new data release.

How it weighs each piece is its own: the public sees the ruling and its reasons, never the numbers behind them. While the evidence is too weak for a date, the question is watching, with no date and the debate still open. Once it is strong enough, Faz rules the most likely year with a range. A dated question goes back to watching only when the evidence clearly weakens, so the date does not flicker on and off. Each ruling shows the evidence that carries it, with links, and what would change it. Each question has one page and one date; a question on the road to it is a question of its own, with its own date. The few older dated yes-or-no claims that have no question yet get a ruling in words: likely by then, an open question, or unlikely by then.

Monthly rulings

Rulings are revised at most once a month. A major event, such as a test that fails or a programme cancelled, allows an early revision, and the reason is written down with it. Every revision says from what to what and why, nothing is backdated, and the page shows whether the date has been drifting earlier or later. A question coming closer is news, and so is one slipping away.

4. It learns from being wrong

When a question's event happens, it moves to the archive, where a report card sets Faz's first and last rulings against the real date: how many years each was off, and how early each was made, next to the positions members defended. When time decides a milestone, Faz's last ruling on it is checked against what happened. Those results feed back into the engine. Once 20 claims are sealed, every new ruling begins with Faz's own report card: where it has leaned too far one way, or not far enough. Every ruling also records a lesson, such as how far announced dates in its field usually slip, and the lessons feed every later ruling. The whole record, misses included, is public on Faz's record.

How Faz gets better

The report card is the automatic part. The rest is people. Members rebut arguments that do not hold, reviewers flag sources that do not say what an answer claims, and editors seal every verdict. Our engineering team reads the archive and the sealed claims, the misses first, and asks what Faz would have needed to see sooner: a source it did not open, a base rate it ignored, a range that was too narrow. The answers become changes to how Faz reads, weighs and reports evidence, proposed with evidence and approved by an editor, and every change is dated in the log below. The engine will never be finished, and we do not intend it to be.

Engine log

Changes to how Faz works, newest first.

• 7 October 2026 One question, one page, one date. The dated yes-or-no milestones under a question ("before 2035?", "before 2040?") are no longer pages of their own: they asked the same question again and split its debate. Their addresses lead to the question. A question can now have questions on the road, each its own "When" with its own date.
• 6 October 2026 Faz rules dates; no public percentages. Faz is the referee, not a forecaster. Pages show its ruling in words, a date with a range or no date yet, and on milestones likely, open or unlikely by then, with its reasons. The numbers behind a ruling, source grades included, stay inside, and members' positions are shown as counts of who defends what.
• 6 October 2026 Topics read as a whole. Watching while the evidence is too weak, a date once it is strong enough, revised monthly.
• 5 October 2026 A reader of judgments. Every ruling now starts from the judgments already on record, weighs each by its evidence and by how such judgments have fared before, and names the ones it rests on.
• 5 October 2026 Evidence mode. Faz reads the sources cited under a claim and builds its ruling from them, citing only pages it has opened. Without web search, it keeps its range wide and says when the evidence is thin.
• 3 October 2026 Threshold claims. Long shots with no natural date stay open until they happen. Faz keeps watching for the day they do, and they are kept out of its record so they cannot flatter it.
• 3 October 2026 Incremental reading. Faz keeps notes on each debate and reads only what is new since its last ruling, so a fresh argument can move the ruling between scheduled updates.
• 3 October 2026 Report card. Sealed claims check Faz's final rulings against what happened, and the summary goes into every new ruling.
• 3 October 2026 First version. Rulings with evidence on both sides, source checks on members' answers, and proposed verdicts that an editor confirms.

What Faz is not

• Not a forecaster. Faz does not predict and has no view of its own to hand down. It rules on what is on the record, and its ruling is only as good as the judgments and evidence there, which is why it shows its reasons.
• Not the last word on a verdict. Faz proposes a verdict with sources when a milestone's date arrives; an editor seals it.
• Not infallible. For now its rulings are made by the Fazoh team with AI assistance, checked against the rules. Read the evidence and decide for yourself.
• Not for sale. Nobody can pay to move a ruling, and donors get no say over Faz.

Claims with no deadline

Some ideas are long shots with no natural date: will we ever confirm alien intelligence, will anyone live to 150? These are threshold claims. They stay open for as long as it takes, and they never close as "didn't happen". Faz keeps ruling on them and watching for the crossing. When one crosses, an editor seals it with the date it happened, and everyone who defended yes before that day is on the record, the earliest first. Because "no" can never be proven, threshold claims are not scored and stay out of Faz's record.

How people are scored

Members are scored on two things. Debate points are for what their arguments bring: sources that hold, rebuttals that land, answers that change minds. Reality points come once the answer is known: how close their position was, and how early they defended it. On a milestone, your last position taken at least 48 hours before the date is the one that counts. The team's positions never count. Faz's rulings are checked against reality in the open too, in the archive and on its record. More on the How it works page.