We built an alpha engine. It lost money.

Scienter began as a copy-trading intelligence pipeline: scan every Hyperliquid trader and vault, gate out the ones whose records fall apart under inspection, and score the survivors with the Deflated Sharpe ratio — the correction that asks how many candidates you sorted before you picked the winner.

The first full run answered clearly. Of more than 23,000 traders scanned and 150 judged in depth, none demonstrated skill beyond what the luckiest of a population that size would show anyway. Not a single one cleared even the loosest tier.

So we pointed the same instrument at our own signal engine over 90 days of live calls. It returned −0.14. The detector we built to find false positives in other people's track records found one in ours.

The tempting reading is that the verification machinery is worthless because it verified something bad. That is backwards. The Deflated Sharpe is a luck-detector, and it worked — on its own author, against its own interest, with nobody forcing it to.

The honest conclusion is narrower and more useful: producing durable alpha is a job that well-capitalised teams do in groups, and one person did not manage it. But the measurement apparatus — the population-scale deflation, the sample-size floor, the refusal to score a three-trade record — is genuinely rare, and almost nobody in this market applies it. That is the thing worth building on.

Every leaderboard you have seen ranks by profit. Ranking 44,000 traders by profit and publishing the top hundred is running 44,000 trials and reporting the luckiest, and the industry does it because the resulting table is more fun to look at than the true one. Ours is the true one, and it is mostly empty at the top.

Every directional call the engine made is hashed and committed before the market can move, then settled against the mid at horizon — so a losing call cannot be quietly dropped from the record after the fact. The losses are in there. That is the point of it.

Read the Alpha Ledger — every call, its commitment hash, and how it settled.

A leaderboard is only worth what its author's honesty is worth. There is no way to prove that from the outside — every site claims rigour, and the ones selling you a signal claim it loudest.

So here is the only proof we can actually offer: the harshest number on this entire site is about us. We ran our own product through our own judgment layer, it returned a negative result, and we put that on the front page rather than in a changelog. If we were willing to shade a number, this is the one we would have shaded.

We publish our own failure so you know we tell the truth about others.

Scienter does not sell signals and does not claim to know what a market will do. It answers one question about strangers with public track records: is this skill, or is this the luckiest of forty thousand coin flips? Reading the answer is free and always will be. We are paid a fee on trades you choose to route through our execution, and on nothing else.

Every scorecard is reproducible from public Hyperliquid data. Every number traces to a published equation. Anything that cannot be computed shows an em-dash instead of a guess.

The full specification — every input, every formula, every threshold, and a list of what it deliberately does not claim — is in the scorecard methodology.