The only one that shows you when it is wrong.

The track record is here, all of it

Every reading in the watching state opens a row that closes 48 hours later, with the result written exactly once. The database refuses any change afterwards.

The track record numbers are not reachable right now. We would rather show nothing than show an old value without saying so — see status.

One engine, one conclusion

There are no panels contradicting each other. Each asset has exactly one official reading, and it always shows the evidence it rests on.

1. 1 · Validated data

   Price by weighted median of three independent exchanges. No price is published with fewer than two agreeing. Derivatives, macro and sentiment come in through the same path, all validated before they touch the database.

2. 2 · Six conditions

   Always the same six, computed by fixed and public rules. A condition that cannot be evaluated shows up as unmet, with the reason written out — never as neutral.

   • 4-hour trend
   • Momentum outside the neutral band
   • Funding at a 30-day extreme
   • Open interest diverging from price
   • Volatility above the regime
   • Volume above usual

3. 3 · The reading, in words

   A language model reads the evidence and writes the official reading. It never sees prices, and every number it writes has to exist in the evidence it received.

4. 4 · Safety rules

   Before any text is published it is verified: no price targets, no predictions, no confidence percentages, and no claiming more than the data supports. What fails is refused whole.

5. 5 · Auditable record

   The reading enters the public record and is confronted with what happened 48 hours later. Written once, never edited.

Pricing

The method, the track record and the system status are public and free forever. What you pay for is full coverage and real time.

Hard questions

Does this tell me what to buy?

No, and it never will. It tells you which conditions are aligned, which are not, and how old the data underneath is. The decision is yours.

I've been burned by signal services before. Why is this different?

Because the track record is public and cannot be edited — the database refuses changes to results already written. It includes the 2,295 observations from the engine that preceded this one, which performed negatively out of sample. We deleted nothing.

Does the AI make things up?

It is structurally prevented from doing so. The model never sees prices, only the already-computed conditions, and every number it writes has to exist in that evidence. Text that fails verification is thrown away and replaced by rule-generated text — and that substitution is counted in public.

What if the data is down?

You see it. Every value shows its age, every source shows its last success, and a panel with stale data dims with the reason written out. Nothing pretends to be fresh.

See for yourself

Nothing is locked behind a sign-up. Open the market, read one piece of evidence, then check the track record to see whether the earlier readings were worth anything.