How to read model performance vs your own results

What the model figure is

It answers one narrow question: given the leaders currently in the basket and their score-derived weights, what would that allocation have produced over the last 7 and 30 days, using each leader's publicly visible trading history?

That is a useful question because it tests the method rather than a marketing claim, and because every input is publicly verifiable on-chain. It is not a forecast, and it is not what your account did.

What it deliberately leaves out

The model has no view of your account, so everything specific to you is absent. Most of the gap between the model and a real follower's result comes from these omissions rather than from anything going wrong.

• Your start date — the model window is not your holding period.
• Your fills — latency, book depth, partial fills and rounding all move your average price.
• Your size — sleeves too small to express positions cleanly track worse.
• Costs — 0.1% of mirrored notional volume on mirrored volume, plus exchange fees and funding.
• Composition changes — the model shows the basket as it stands now.

How to compare the two fairly

Use the same window, and use one that is long enough to contain more than a handful of trades. Compare against the basket composition that was actually mirrored during that window, not today's. And compare cost-inclusive results to cost-inclusive results — a gross model figure against a net account figure is a guaranteed mismatch.

The most common self-deception is asymmetric attribution: crediting the method when your account beats the model, and blaming execution when it lags. Both directions are the same phenomenon.

Why per-leader attribution is the better lens

Because each leader runs in one Hyperliquid sub-account per mirrored leader, you can compare sleeve by sleeve instead of arguing about one aggregate number. That reveals the actual cause: a specific high-turnover leader diverging, a sleeve too small to size cleanly, or funding on a long-held position.

An aggregate gap tells you that something differed. Per-sleeve attribution tells you what.

What no performance number can tell you

Neither the model nor your own trailing return says anything reliable about the next period. Scores are backward-looking, baskets change, and regimes end. The transparency claim here is about verifiable inputs and stated method, not about predictive power.

Past performance is not indicative of future results. Perpetual futures are leveraged instruments and carry a substantial risk of loss, including the loss of your entire position.