Method

This page is the arithmetic. For the rest of the surface — every section of every page, what it shows and why — take the tour.

How to read a belief bar

A Question shows one bar per candidate answer, plus one more: unknown. The bars are shares — they add up to everything that could be the case, including "none of the above". The unknown bar is always rendered, always last, and drawn differently from the others because it is a different kind of thing: it is the room left for an answer nobody has proposed yet.

Leading is not proven. The top bar on a Question is the branch the current evidence favours, not a finding. A 34% branch is a 34% branch; the other 66% is not silence.

Belief is derived, never stored

No one types a score into this system. Every number is recomputed from the graph of support and challenge that surrounds a claim. Adding evidence adds an edge; it never edits a score. That means every figure can be taken apart — on a claim page you can see each contribution that produced it, term by term.

Repetition is not evidence

When forty outlets run the same wire story, the model records forty restatements of one assertion — and gives them the weight of one. Independent legs (a separate source with a separate basis) move belief; a chorus does not. The same rule governs what gets ranked as "hot": attention here follows evidence and inference, not volume, so a syndication storm cannot push a story up the page.

The floor: ignorance is not 50/50

A hypothesis nobody has supported does not start at "maybe". It starts at the floor — near zero — and has to earn its way up. This is what makes it safe for the system to generate hypotheses openly: an unsupported theory has no power, so proposing one costs nothing and proves nothing.

The residual

Every exclusive Question carries an explicit "none of these" share. It rises when assertions arrive without discriminating evidence — which is exactly the situation where a normal news product would show growing certainty. A Question with a large and growing unknown share is a Question where the world is shouting and the evidence is not moving.

Inferences and their critical questions

Some claims here were not asserted by anyone; the model derived them by applying a named argument scheme (for example, Cui Bono — who benefited). Derived claims always show the scheme that produced them and the premises they rest on, together with that scheme's critical questions: the standard challenges that, if answered unfavourably, withdraw the inference. An inference whose critical questions are open is shown as exactly that — provisional.

What makes something "hot"

The front page has to rank things, and ranking is a judgement — so it is explained here like every other number on the site. A Question's heat is:

heat = activity × (1 + contention) + 10 × share-movement + 3 if the lead changed

Activity is a weighted, time-decayed count of what changed on that Question in the last week (half-life 36 hours). The weights are the honesty rule written as numbers: a new contradiction counts 5, a ruled-out combination 4, an inference 3, a new answer 2, a corroboration 1 — and a restatement counts zero, however many of them arrive. Corroborations from one source are capped at 3 a day, so a burst cannot substitute for evidence.

Contention is how close the race is, counting "unknown" as a contender — a Question the fog is winning is a live Question. Share movement is how far belief actually shifted. Together they mean the top of the page is where the argument is genuinely moving, not where the volume is loudest. That is also why the badges say "2 contradictions · lead changed" rather than a score: the reasons are content, the number is plumbing.

Lenses

Judgment is a view, not a fact. The same structure can be read under more than one model, and every number on the site names the lens it came from. Switching lenses never changes which branches exist — belief may rank branches; it may never hide one.

Evidence-weighted log-odds@1

Belief computed from the support/challenge graph in log-odds: independent legs add, challenges subtract, restatements add nothing.

Structural structural@1

No probabilities at all — only how well tested each branch is: evidentiary legs, independent sources, challenges. Untested and contested are different shapes, not different numbers.

Add ?lens=structural@1 to any page's URL to read it under the structural lens.

Two clocks

Everything the model records carries two times: when it happened (event time) and when we learned it (knowledge time). The gap between them is the fog of war. Adding ?asof=YYYY-MM-DD to a page renders it as the model stood on that date — which is how you can watch a story's confidence form, or fail to.

Nothing is deleted

A branch that loses is greyed, not removed. A contradiction that rules out a combination is recorded as such and shown when you try to walk that world. The record of what was believed, and when, is part of what this site is for.