Method

A leveraged ETF is built to deliver a fixed multiple of its index's return. SOXL targets three times SOXX; SOXS targets minus three times the same thing. That multiple is set in the fund's prospectus, not estimated by us. Measured from a common anchor over a single session, it reduces to one line of arithmetic:

Everything left over is the residual, quoted in basis points. It is not a forecast and it carries no view about where anything is going. It says only that two prices the venue is publishing at the same moment cannot both be right.

rTokens route to NYSE and Nasdaq while those venues are open, which on Bitget is 04:00 to 20:00 ET. Outside that there is no market behind the price. Each night is anchored on the close of the 19:00 ET bar — the 20:00 session close — and every hourly bar from 20:00 through 04:00 is checked against it.

Anchoring both legs at the same instant is what makes the daily reset of a leveraged fund irrelevant here. The reset matters for holding across days; within one night it does not.

The same arithmetic run between 10:00 and 16:00 ET is the control: arbitrage is live then, so the residual should be small. It mostly is. But hourly closes on two different instruments in a fast market are not simultaneous, and that timing mismatch manufactures residual of its own. For a few pairs the session reads wider than the night for exactly that reason.

So the headline here is the absolute breach rate, not the ratio between session and night. The absolute figure does not depend on the control being clean.

Holding one dollar of the leg against |β| dollars of the base means 1 + |β| dollars working for every dollar of signal, and you pay on the way in and the way out. At 10 bps a side that is 80 bps round trip on a 3× pair and 40 on a mirror pair. Nothing in the record clears it.

The research desk runs on a model with three tools wired to Bitget's public market API: a spot price, recent candles, and the day's movers. It is instructed never to answer a price question from memory — every figure it quotes came back from a call it made while you watched, and the chips above each answer name the tool it reached for. The board's own state is handed to it as context, so "what is breaking tonight" is answered from the same reading you can see.

• —The arithmetic proves one of the two quotes is wrong. It never says which one.
• —Hourly bars, not tick data. A breach that opens and closes inside an hour is invisible here.
• —A bar with no trade is absent rather than carried forward, so a genuinely frozen quote is dropped rather than scored.
• —Fund expenses, borrow and the daily reset all move the identity by a few basis points a day. That is noise against a 20 bps tolerance, not against a 2 bps one.
• —Coverage of the dark window is only meaningful from 2026. Everything earlier is too sparse to grade.
• —This measures quote consistency. It is not advice, and it is not a claim that either price is the right one.

Everything comes from Bitget's public spot endpoints — no key, no account, no signing. scripts/fetch.py pulls hourly candles for the fourteen tickers the identities touch; scripts/audit.py grades them and writes audit.json, which is the only source for every number on this site. The browser re-implements the same residual, so the live board needs no server of ours at all.