Trading terminal execution quality 2026: what a swap costs once the app's own fee is removed

Bench 268 reads real user swaps on Solana and five EVM chains for the trading terminals, wallet apps and Telegram bots in the cohort, and values what the user received against the pool's own state before the trade. Loss is the shortfall in basis points of the trade, and it decomposes per swap into the terminal's fee, the network cost (priority fee, inclusion tips, token-account rent), the pool's LP fee and price impact, and the protocol cuts (launchpad curves, routers). This bench publishes the same quantity with the terminal component removed, subtracted on each individual swap before the median is computed: median(loss) minus median(fee) is a different statistic and differs materially where an app's fee is not constant across its swaps. Network, pool and protocol costs stay inside, because routing through a launchpad curve rather than a migrated pool, or attaching a larger tip, is a choice the app makes on the user's behalf. Same rolling 24 hour window, same publication floors, same flow weighting across a product's chains, medians with a bootstrap interval and p90.

• ·The measurement is bench 268's, with one subtraction. Every detail of sampling, user attribution, quote handling, pool identity, reference price and loss bounds is that bench's; see its methodology. Nothing is re-read from chain and no additional RPC is spent.
• ·Execution loss = loss − terminal fee, computed on each swap, then the median over the window. The subtraction happens before the quantile, never after: median(loss) − median(terminal) is a different number, and the gap is largest on an app whose fee differs across its own swaps (FOMO charges 2 % on Solana and about 0.5 % through Relay). It moves with the cohort from one refresh to the next, so read it from the two columns on the board rather than from any figure quoted here.
• ·What is removed is the terminal's own cut: what reached its fee wallets on Solana, the router's residual after the pool and the gas on the EVM rows, the app fee on Relay legs. Referral and cashback legs count as terminal fee where the terminal's list holds them.
• ·What stays in: network (transaction fee the user paid, inclusion tips including the terminal's own relay, the deposit of token accounts the swap creates), pool (the final pool's LP fee and price impact) and protocol (launchpad and curve fees, router cuts, referral payouts). Each is a consequence of how the app routed the trade, not of what it charges for doing so.
• ·A swap whose fee exceeded its loss reads negative: a rebated fee, or a reference that moved in the user's favour between the previous trade and this one. Those swaps are kept: removing them would trim the left tail and push every median up.
• ·Publication and ranking follow bench 268 exactly: same floors, same health and ranked flags, same per-chain pooling weighted by flow. A product present on one board is present on the other, so the two can be read side by side without a row appearing in only one of them.
• ·The two rankings differ: a product whose fee sits far from the cohort's usual 1 % changes place once the fee comes out, in either direction. How far each row moved is read from the two live columns, not from this sentence. Read this board with the 95 % interval under each median. As on bench 268, most adjacent pairs are not separated by the measurement.