Verify an AI signal record
Four steps to confirm a single past algorithmic call yourself, no spreadsheet required.
There is no need to comb through an entire history to gauge an “AI” service's honesty. Run a single past call from end to end and you have learned the thing that counts: whether the record can be checked at all. The four steps that follow climb from the quickest, cheapest look to the one that settles it.
Verify in four steps
1. Begin with the sample size
Locate the total count of calls and make sure the failures are counted in. An accuracy figure with no number of calls behind it — or one quietly stripped of its losses — is out before you go any further. The desk pick puts its fast model at 67.5% over 308 day-trade calls; the 308 is the figure you are hunting for, and the percentage says nothing without it. The full version of this test lives on a re-runnable record.
2. Make sure it is live, not simulated
Hunt for a continuous, forward run of real-time calls instead of a curve smoothed after the fact. An “AI” model parading nothing but a glossy equity line is showing you a backtest until it proves otherwise. A real record states its window — here, 2026 to date — and keeps its ugly stretches in view.
3. Locate the outside reviewer
Confirm that a named third party has examined the underlying statements. A platform ranking is not an audit; a testimonial is not a review. The third-party competition results are at World Cup Championships.
4. Match one call against the ledger
Here is the step that decides it, and the one most services cannot withstand. Take one past call and line its published fields up against its Bitcoin stamp. Because that stamp was laid down before the trade resolved, a match certifies the fields were set ahead of time. One confirmed call is worth more than a hundred screenshots. This is what it looks like in practice:
The call shown here is invented for the walkthrough, not a real trade. The steps, though, are precisely what you would carry out on an actual published call.
- Grab the published call and its five fields. Suppose it reads: long the index ETF,
entry 412.80,target 414.20,stop 412.10,grade B,signal time 14:32:05 UTC. - Rebuild the digest. Those five fields are joined in a fixed order and fed through SHA-256, a one-way function that compresses any input down to a single fixed-length string. Identical inputs always yield the identical digest; flip one character and the output changes entirely.
- Pull up the ledger stamp. The OpenTimestamps receipt that shipped with the call names the Bitcoin block its digest was anchored in; its open verifier, or any block explorer such as mempool.space, lets you read that block for yourself. Check that the digest you rebuilt is the one sitting in that receipt.
- Read the timestamp. Find when that block was mined. If it was mined ahead of the trade resolving, then the call — entry, target, stop and grade as a set — was demonstrably locked beforehand. That is the entire proof.
Now try to cheat it: pretend the stop was slid from 412.10 to 412.40 once the market turned. Step 2 would now spit out a digest that no longer agrees with the receipt in step 3, and the edit is caught. That is why a verified stamp beats any screenshot — it breaks noisily the moment a field is altered.
In short: the first three steps take a few minutes and weed out most of the market; the fourth is the one no service can fake. Clear step four and a service has handed you a record open to inspection rather than mere applause. The machinery behind it is laid out on locked before the outcome.