The best quote says yes. Your size says no.
Every weekend-tradable rToken on Bitget, 90 of them, walked at real order sizes with Sounding’s own engine, at Bitget’s published 5 bps rToken taker fee. What a quote cannot show: what your size actually costs.
names where the best ask is inside a 50 bps ceiling but a 5,000 USDT buy is not, at a 5 bps fee.
the same at 25,000 USDT; 1 more book is too thin to fill it at all.
25,000 USDT buys within 50 bps at the published 5 bps but over at the 10 bps list rate: where knowing your own fee changes the answer.
One recorded snapshot, 2026-09-20 09:03 UTC, a Sunday weekend session; recomputed when this page was built. Median all-in for a 5,000 USDT buy at 5 bps: 23.61 bps. The 6 at 5,000 USDT: rSPMO, rBABA, rSQQQ, rCOHR, rEWY, rAEHR.
Every size, every ceiling.
| buy budget | ceiling | over at full size, 5 bps fee | best ask yes, full size no | within at 5 bps, over at the 10 bps list rate | too thin to fill |
|---|---|---|---|---|---|
| 1,000 USDT | 30 bps | 15 / 90 | 3 / 90 | 1 / 90 | 0 |
| 1,000 USDT | 50 bps | 5 / 90 | 2 / 90 | 1 / 90 | 0 |
| 5,000 USDT | 30 bps | 29 / 90 | 17 / 90 | 14 / 90 | 0 |
| 5,000 USDT | 50 bps | 9 / 90 | 6 / 90 | 2 / 90 | 0 |
| 25,000 USDT | 30 bps | 70 / 89 | 58 / 89 | 1 / 89 | 1 |
| 25,000 USDT | 50 bps | 31 / 89 | 28 / 89 | 17 / 89 | 1 |
Cost against each book’s displayed mid, with the taker fee charged on the traded amount. Books too thin to fill the order are counted in their own column and left out of the other denominators. An independent reviewer recomputed every cell at 5 bps with its own script, adding the fee on top, and found the same counts; charging the fee on the traded amount changes none of them on these books.
Is one snapshot typical? Measured every 30 minutes.
The schedule was written before the first capture: every eligible book, raw, every half hour until 2026-10-08, across weekend, overnight and US sessions, with a dozen names re-read in batches scheduled 5, 15 and 30 seconds after each sweep, to see how fast an answer goes stale. The gap is measured from the two exchange timestamps: the +30 s batch came 30.9 to 89.8 s after the first read, median 40.9 s. Failures stay in the record.
174 rounds analysed (plus 2 of 176 attempted where Bitget's metadata could not be read; kept in the record, left out of every rate), 2026-10-03 23:58 to 2026-10-08 11:46 UTC · 15660 of 15660 books read · sessions: weekend 24, pre-market 39, regular 39, after-hours 24, overnight 48
| session | best ask yes, full size no · 5,000 USDT buy, 50 bps | same · 25,000 USDT | answer changed at the +30 s re-read (median 40.9 s later) · 5,000 USDT, 5 bps fee, buy and sell side-pairs |
|---|---|---|---|
| weekend | 152 / 2160 (7%) | 585 / 2160 (27%) | 0 / 552 |
| pre-marketorders routed to the US market | 102 / 3510 (3%) | 1033 / 3510 (29%) | 19 / 936 |
| regularorders routed to the US market | 43 / 3510 (1%) | 689 / 3510 (20%) | 4 / 936 |
| after-hoursorders routed to the US market | 47 / 2160 (2%) | 509 / 2160 (24%) | 6 / 576 |
| overnightorders routed to the US market | 127 / 4320 (3%) | 1190 / 4320 (28%) | 29 / 1152 |
Only the weekend and holiday rows describe the book an order meets. In US sessions Bitget routes rToken orders to NASDAQ/NYSE, so those rows read a book the order does not fill against; they stay in the record because they were on the schedule, and Sounding refuses orders in those sessions.
What this cannot show: fills, hidden or off-book liquidity, hours that were not sampled, or that one week is typical. A missed round stays missing.
pnpm exec tsx scripts/census-inversions.ts evidence/census-20260920/eligible-census-RAW-20260920T0902Z.json 5000 5 50 pnpm exec tsx scripts/atlas-analyze.ts