Every change to
how costs are computed.

Orders larger than the deepest depth we sample.

• The order book is sampled at $1k, $5k, $25k and $100k. An order above $100k was previously priced at the $100k rung and still labelled Measured, which understated its cost. The liquidity line and the impact cost line now say that nothing is measured at that size, and why, instead of showing a figure about a smaller order. The same rule applies where a card or a basket is scaled to the amount entered: above $100,000 the impact line is withdrawn rather than scaled, and it is not counted in the total.

First engine.

• Fees from a version-controlled catalogue; Kraken spot and futures verified against the venue's own API, the rest labelled Published but unverified.
• Funding costed at each symbol's actual settlement interval rather than an assumed 8 hours, and reported as a distribution rather than a mean.
• Impact measured at the order's size from our own order-book samples, refused where no samples cover the period.
• Signal-to-fill drift measured on 1-minute data. An earlier 25%-of-bar-body placeholder overstated this cost by 6-8 bps per side and was withdrawn before launch.
• OHLC spread estimators restricted to wide-spread symbols, where calibration shows they work; they fail on majors and are not used there.