NOBIDUpdated 2026-09-06
Risk
A schedule is a claim about what can go wrong and what it costs. Three things can go wrong here. Two of them are priced in basis points and the third is not priced at all, because it is not the kind of thing a wider quote protects you from.
Three ways this loses money
Not three ways it might underperform. Three mechanisms by which the desk ends a period with less capital than it started with.
| The risk | Priced by | Basis points | What would actually fix it |
|---|---|---|---|
| The overnight gap | Out-of-hours term | 202 | A hedge in the underlying, held through the close. |
| Inventory with no exit | Venue term, stranded tier | 420 | A venue. |
| The reference itself | Nothing on the schedule | — | A price feed the desk does not supply to itself. |
One — the overnight gap
A curve denominated in a tokenised equity trades all 168 hours of the week. The equity behind it trades 6.5 hours a day, 5 days a week — 32.5 hours in total. For the other 135.5 hours, 80.65% of the week, the desk is quoting off a reference that cannot be transacted at and cannot be hedged against.
That is what the out-of-hours term buys: 202 basis points, from z = 0.84 against an assumed single-name overnight σ of 2.4%. At the eightieth percentile the desk expects to be wrong about one night in five and paid for the other four. That is not the failure mode; it is the design, and a term that was never wrong would be a term nobody would trade against.
The failure mode is the night that is not in the distribution at all. σ describes an ordinary overnight move in a retail-heavy name; it does not describe a halt, an earnings release, a corporate action, or a Sunday announcement the tokenised wrapper has to catch up to on Monday. No single volatility assumption can, because a distribution fitted to ordinary nights is structurally silent about the nights that matter. The desk is short that tail and is not pretending 202 basis points cover it.
Two — inventory that never finds an exit
This is the one that could end the business, and it is uncomfortable in a specific way: it is the desk’s own thesis turned around and pointed at its balance sheet.
The argument for the desk is that 115,094 live curves — 47.89% of the 240,350 alive at block 55,223,440 — are denominated in assets with no pair against WETH or USDG on the canonical factory, so their holders cannot reach ether at any price. Every word of that is still true after the desk buys the position. The desk becomes the holder of exactly the thing it has just described as unexitable, on purpose.
The schedule prices the difference between the two kinds of warehousing rather than hiding it. A position the desk can offset against a listed instrument costs 145 basis points and is held until the next open, which is a date. A stranded position costs 420 — the extra 275 basis points is the entire compensation for holding something with no date at all. Whether that is enough is not a question this page can answer, because answering it means knowing how long the exit takes to appear, and nothing here measures that.
Underneath it sits a smaller version of the same problem: the desk does not yet know the shape of its own addressable book. The aggregate split is measured, but the per-curve census is not. The only quote asset confirmed by direct call is AMC, and the run that resolves the rest is scheduled for 2026-09-14. Until it lands the desk cannot say how concentrated the inventory would become in any one ticker, so it cannot size a concentration limit honestly either.
Limitation
There is no hedge for this one, today
Three — the reference itself
Every figure on the quoting page is a deduction off a reference notional, and the reference is an input: supplied by whoever calls the quote, with no price feed of the desk’s own behind it. The most load-bearing number in the waterfall is the one number the schedule does not produce, does not verify and does not price.
The chain already demonstrates what a wrong reference does at scale. The deepest curve on it prices in AMC Entertainment · Robinhood Token and reports a reserve of 1625. Read as though the quote asset were ether — which is what every screener, index and wallet here currently does — that is $3,996,184 at $2,459 per ether. It is 1625 AMC. The error is not subtle, not rare, and is being made continuously by software that is confident.
A desk making the same class of error makes it with its own balance sheet, one ticket at a time, and no basis-point term detects it: the waterfall applies identical deductions to a good reference and a bad one. This is the risk with the least written under it here, and that is not because it is the smallest.
What actually mitigates
Four things, structural rather than clever. The size bands cap what one ticket can do: the last schedulable band, Large block, charges 210 basis points and then stops, so a bad reference or a bad night is bounded per trade by construction rather than by somebody remembering to be careful. The schedule ceiling at $250,000 keeps the largest tickets off the automated path entirely — above it the endpoint returns a reason and a person decides. Settlement in ETH only leaves one inventory and one hedge to run instead of three, which is set out on the settlement page. And the 90-second firm window bounds the free option a counterparty holds over a stale quote.
A fifth is the publishing itself. A schedule that cannot be varied per counterparty is one the desk cannot talk itself into widening for a position it wants, and the discipline of defending 0.84 and 2.4% in public is worth more than either number.
What does not
None of the above is a hedge. A wider quote is compensation for a risk taken, not protection against it, and the difference matters most in the scenario where a desk would like to believe otherwise: many stranded positions, bought at a good schedule, in a week when the exit is further away than it was. The desk is not open and is not holding anything today, so nothing here has been tested against a real book. When it has been, the results belong on this page before anywhere else on the site.