Snapshot Fees: Pricing Liquidity From Within a Batch Auction

This piece proposes Snapshot Fees, a fee mechanism that prices each trader’s standing liquidity by what it offers the market. The reading comes from the market itself: a batch auction seals the whole book at one instant, and every fee is computed from that sealed snapshot. Each fill is priced by whether its liquidity stood on one side of the book or both, and when both, how near the market’s center. The mechanism is built up in layers, each with an interactive lab for trying it directly.

Understanding Batch Auctions

Most electronic markets run on a continuous limit order book. Orders rest in a book sorted by price, and when a new order crosses the spread it trades immediately against the best resting quote. Ties at the same price go to whoever arrived first. In a continuous limit order book, every fill, quote, and cancel is processed one at a time, in arrival order.

Processing in arrival order rewards whoever reacts first when the market moves. When news changes the fair price, a market maker’s resting quotes go momentarily stale, and until the maker cancels them, anyone who reaches them first can trade against the old price at a profit. A race follows each such move, and it is a lopsided one: the maker’s cancel is a single message running against the orders of every firm trying to snipe the quote, and more often than not the quote is picked off before it can be pulled. These races are known as sniping, or latency arbitrage, and they are a form of adverse selection that concentrates on market makers. Makers respond by quoting smaller size and wider spreads, and that added cost is paid by everyone trading for ordinary reasons.

Aquilina, Budish, and O’Neill measured the race directly, using exchange message data that records the losers as well as the winners. In FTSE 100 stocks they found races running about once per minute per symbol, typically decided in five to ten microseconds, accounting for roughly 20% of trading volume, with the top six firms on one side or the other of more than 80% of them. Eliminating latency arbitrage, they estimate, would cut the market’s cost of liquidity by 17%, about $5 billion a year in global equities. This additional liquidity cost is structurally inherent to the design of continuous order books.

Frequent batch auctions are the design change Budish, Cramton, and Shim proposed: instead of processing orders continuously, collect them for a short window, seal it, and clear every crossing order at one uniform price. Inside a window there is no first. A momentarily stale quote is no longer a prize for the fastest, because while the auction window is open, everyone has the same chance to react. With that form of adverse selection neutralized, a maker can quote tighter and with larger size.

The design space has kept moving since. One example is Jump Crypto’s 2025 dual-flow batch auction design, which splits each window into two auctions, maker bids against taker sells and maker asks against taker buys, so taker flow always trades against standing quotes instead of netting against itself. While the Snapshot Fees mechanism proposed in this piece could work for any batch auction implementation, it is likely a more natural fit for a dual-flow batch auction, for reasons covered in The Dual-Flow Fit near the end of the piece.

Plain or dual-flow, batch auctions improve on continuous books the same way: latency arbitrage is neutralized, and with it the widened spreads it forces. Discrete windows also provide something continuous books cannot: a clean snapshot of the market’s state at each close, with every order in it committed and fillable at that instant. This piece asks whether that snapshot could be used to price orders directly, and whether pricing them well could improve the market further, with even tighter quotes and more size behind them.

Trading Fees Today

Exchanges everywhere charge for execution, and nearly everywhere the same two ideas shape the price: liquidity provision is treated better than liquidity demand, and activity is rewarded, with rates that fall as volume grows. In traditional markets these arrangements are more layered and bespoke, running through negotiated maker programs and rebate schedules that vary by venue and firm. This piece focuses on crypto markets, where fee structures are simple, public, and already quoted in bps. The typical schedule is compact: a maker/taker split, where takers pay a few bps of notional and makers pay less, sometimes nothing, occasionally a rebate. Tiers then scale those rates with each account’s trailing volume.

Volume-based pricing still works on a batch venue, but the sealed snapshot makes a different kind of pricing possible. Jump Crypto’s dual-flow paper alludes to it: maker fees could be priced by how far an order stands from the midpoint. This piece takes that idea and expands it into a full pricing mechanism.

Pricing Each Auction’s Liquidity

If the objective is to use fees to improve the market’s quotes, that objective can be priced directly:

Liquidity standing on one side of the book pays the cap, the highest rate any resting order is charged. Liquidity standing on both sides pays less. Liquidity standing on both sides, close to the mid, pays the least.

Two properties follow directly from pricing this way:

• Day-one competition. A new maker can pay the best rates from their first window. Since the rate follows the quality of the quotes, a newcomer willing to quote tight and two-sided starts on equal fee terms with the largest incumbents.
• Per-market competition. Fees are calculated market by market, window by window. A maker’s standing in one market buys nothing in another, so every market, however small, is independently contestable: a specialist quoting one obscure pair well competes on equal terms with the largest firm in it.

Together, these two make liquidity supply responsive. When interest arrives in an asset, makers can arrive with it at full strength without the need to build up history first.

From here the piece builds the mechanism outward: the price of placement for a single order, then a full one-maker book, then the market-wide version, each with its own lab.

The Base Fee

For liquidity nearest the mid to pay the least, distance from the mid has to carry a price. The base fee is the first leg of that price: it sets the lowest rate a resting order can be charged given its distance from the market’s midpoint, the Measured Price, $M$, which is calculated from the sealed book. Whether an order actually gets that rate is the rest of the mechanism’s job.

The schedule around the Measured Price takes a simple shape. Nearest to $M$ sits a band with a fee level of its own, zero in everything this piece shows. Beyond the band’s edge the fee grows with distance, at rates the market configures; under the default calibration used here it grows gently through the zone where makers do much of their everyday quoting, then more steeply past it. The fee never exceeds the cap the rule referred to: $F$, the most a resting order can be charged. Market orders pay a rate of their own, never below $F$, so no resting order is ever charged more than a market order.

To calculate the base fee, seven values are necessary, and they fall into three groups.

The first value is dynamic and recomputed every window:

• $M$, the Measured Price: the market’s midpoint, calculated from the sealed book. Its formula is introduced in The Measured Price below, and the sections after it refine how the reading works.

The market’s configuration settings, fixed from window to window, though the regions they define all sit relative to $M$:

• $B$, the Inner Band: the region $M \pm B/2$, declared by the market. The base fee there is zero throughout this piece.
• $Z$, the Maker Zone: the working radius past the band edge.
• $k_1$, the Zone Slope: fee per bps of distance inside the zone.
• $k_2$, the Far Slope: takes over past the zone edge and runs the fee to the cap.
• $F$, the Fee Cap: the ceiling of the resting schedule. No resting order ever pays more than $F$, and no market order pays less.

And the one value belonging to the order being priced:

• $d$: an order’s distance, in bps, beyond the band edge. It starts counting at $B/2$ from $M$, and inside the band it is zero. The fee is a function of $d$ and nothing else.

Put together, the base fee for a single order is:

Read left to right: distance inside the zone is charged at $k_1$, distance beyond it at $k_2$, and the total is clipped at $F$. Inside the band, $d$ is zero, so the fee is zero.

Alongside the fee, the lab below plots a second curve: the net edge, an order’s distance from $M$ minus the fee charged at that distance. Distance is roughly the edge an order is asking for: if it fills, it earns that many bps of price improvement relative to $M$, and the net edge is what remains of that improvement after the fee. The curve ignores fill probability, adverse selection, and inventory, so it is not a profit model, but it does show what the schedule rewards. With $k_1$ below 1, a wider quote still nets more edge per fill, just less per bps. And once the fee reaches $F$, distance stops mattering: every placement past that point pays the cap.

The Order Fee

The base fee set the floor: the lowest rate an order can pay at its distance. What an order actually pays comes down to one rule:

Every resting dollar of notional is either paired, with a dollar on the other side of the book behind it, or unpaired. A paired dollar pays the worse of its two legs’ base fees. An unpaired dollar pays the cap.

A market functions when there is something to trade against at any moment, in either direction. Judged against that need, orders fall into three classes. Non-directional resting liquidity, quotes standing on both sides of the book at once, serves both directions at the same time; it is what lets a market function continuously. Directional resting liquidity, a limit order with nothing behind it on the other side, still provides something real: it stands committed and fillable, adding depth the market can trade against, but only on one side, and only because an opinion happened to rest there. A market order provides nothing to trade against; it is demand alone. The schedule prices the three in that order: non-directional liquidity earns the discounts, by placement; directional resting liquidity pays the cap; and market orders pay a rate of their own, never below it. How far the market-order rate should stand above the cap leads the Design Notes.

Non-directional liquidity is determined from the sealed order book at auction close: a trader’s notional dollars on the bid side pair against their dollars on the ask, best-priced first, until the smaller side runs out. A pair forms only when the bid stands below the ask: a trader’s bid at or above their own ask would trade with itself, and both legs stay unpaired. Each paired dollar pays the worse of its two legs so that a tight bid cannot borrow a better rate than the far ask it is paired with.

The calculated fee rate is charged only when the underlying liquidity actually trades. The formula for the order fee is:

where:

• $q_i$, a slice: the portion of the order paired against a single opposite-side level.
• $b$: the slice’s own base fee.
• $b_i$: the base fee of the level the slice paired against.
• $u$: the notional left unpaired.
• $q$: the order’s total size, $q = \sum_i q_i + u$.

The formula is the combination of the base fee and the pairing rule. For a maker to lower its fee rate, it should place orders closer to $M$, where the base fee $b$ is smaller, and size them to what the other side of the book can pair, so nothing lands in $u$ paying the cap.

The Measured Price

Every fee in the schedule is priced as a distance from a single reference point: the Measured Price, $M$, recomputed each window from the sealed book.

Whatever form the calculation takes, the reference has two criteria to meet:

1. The Measured Price should reflect the real cost of trading.
2. The Measured Price should be influenced only by those who directly bear its cost.

Two familiar references both fail:

• The midpoint of the best quotes. One dollar of bids at the touch sets the same midpoint as a million, though what trading against them costs is entirely different. It fails the second criterion as well: a few dollars parked ahead of the real depth set a new best price and carry the midpoint with them, bearing almost nothing for where it lands.
• A weight over every resting order. This improves on the midpoint by the first criterion: depth now enters the number, and it moves closer to what trading at size would cost. It fails the second, and not in a way any range restriction could repair. A directional order’s rate is the cap wherever $M$ sits, so it bears none of that cost and has no incentive to place it well; size piled on one side just below the touch would drag the average toward itself, and being wrong would cost it nothing.

The Measured Price keeps the depth-weighting but restricts it to the liquidity for which being wrong is expensive: a trader quoting both sides is stating where they believe the price stands, their fees hang on $M$ through both legs, and a mispriced leg stands to be traded against.

Within that liquidity, weight comes from size and placement: more notional is a larger claim about where the asset trades, and notional standing closer to the mid is a more confident one. The whole calculation can be thought of as an election held each auction: standing paired, fillable size is how a trader votes on $M$, and size and proximity set the weight of each vote.

The calculation has boundaries. Quotes count only within $B/2 + Z$ of each side’s best quote, half the Inner Band plus the Maker Zone: the same working width the fee schedule discounts is the width $M$ reads, which is where the zone gets its name. The pairing must also happen inside that area, on both sides. A tight bid whose only ask stands beyond the zone gets no voice, even though the two still pair for fees; the fee’s pairing accepts any distance and prices it, while the reading requires both legs to stand within the area it covers.

The same width bounds the market itself. The best bid and best ask being read must stand within $B + 2Z$ of each other, the full working span. Two quotes farther apart than that are not a market with a wide spread; they are two claims about the price with no working region between them, and $M$ does not average them. A book that wide produces no reading at all.

The Measured Price, then, is not calculated from the raw book but from a filtered one. Per trader, keep only the two-sided size that pairs within the area, and pool what survives by price level. The ladder that remains is the eligible book, the only book $M$ ever reads. Comparing prices across it requires a common size to evaluate them at. $D$, the market’s typical demand, is that basis (how $D$ is derived from the market’s own activity, rather than set by hand, comes in Tracking Typical Demand). The walk takes $D$ dollars of notional into the eligible bids, best prices first, and averages everything it consumed by volume. That is the bid-side impact price, the price a seller of $D$ dollars would actually receive against the eligible liquidity. The ask side gives its mirror, and the Measured Price is the midpoint of the two:

Both impact prices are read from the eligible book alone; the rest of the book never enters the formula. Measuring at size is what makes $M$ expensive to move. A tiny order absorbs almost none of a $D$-dollar walk, so it barely moves the average; shifting $M$ requires real size near the mid. If paired depth within the area runs out before $D$ dollars, the missing dollars are priced at the area’s boundary; the walk never reaches for size beyond it.

The lab below runs everything so far on a single maker’s book: the walk, the Measured Price, and a fee receipt for every level.

Many Makers, One Price

Everything so far ran on a single maker’s book. In a real market the eligible book pools every trader’s paired, in-area size at once: the same filter and the same walk, producing one $M$ for the whole market.

Within the pooled walk, a trader’s voice is their share of the dollars it consumed. Better-priced quotes are consumed first, and quotes at the same price share pro-rata. A small maker’s size counts exactly in proportion, with no seniority and no minimum.

This produces a property worth stating plainly, because no existing fee schedule has it: a maker does not set their fee. They set their book. The formula is identical for everyone, but the $M$ inside it is calculated from every maker’s quotes at once, so the rate a given book pays depends on where the rest of the market stands. The dependence shrinks with weight, and weight comes from the same actions the schedule already rewards: a maker quoting larger and tighter pulls $M$ toward the center of their own quotes, which is where their rate is most predictable.

Pooling has one more consequence. Makers usually cluster, everyone’s quotes within reach of everyone else’s, and the filter produces a single eligible book. But books can also stand so far apart that neither lies within the other’s reading reach. Each region then forms its own candidate eligible book, and $M$ reads the largest one. The smaller candidate gets no voice: its quotes are priced against the larger book’s $M$. Moving $M$ away from the market therefore has a single price: standing more paired, fillable size than everyone else combined, within one working span.

Two situations produce no new reading. If no valid candidate stands anywhere, or two candidates that share nothing stand at exactly equal size, the market holds the last $M$, flagged as held, until a dominant book returns. And a market that has never had one holds nothing: from launch until the first candidate forms there is no $M$, and every fill pays the cap. The first account to stand a real two-sided book within the span sets the first Measured Price, and with it the first access to a discount.

The lab below is that market: your book on top, the aggregate of every other maker mirrored beneath it, both feeding one walk.

Tracking Typical Demand

To calculate $M$, the eligible book must be walked on both sides up to size $D$, the typical demand for that market. What remains is how that typical demand is determined.

Within any market, demand continuously changes. Therefore, $D$ should not be based upon some generic constant. Instead, it needs to reflect the market’s conditions as they actually are.

Doing that well means satisfying two criteria:

1. $D$ should follow the market’s demand as it changes, quickly enough to track a real shift within days and steadily enough that the measurement adds no noise of its own.
2. $D$ should be trustless, computed from published data by a fixed rule, predictable and verifiable like every other number in the design.

What follows is one mechanism that satisfies both. Unlike the fee rule and the Measured Price, it is not settled design: it is a demonstration of one possible solution. If preferred, other mechanisms could serve in its place.

A Proposed Mechanism for Tracking Demand

The mechanism runs on two clocks:

1. Every window: the market’s fills add to two running sums.
2. Every day: the sums are used to compute the typical demand for that day and $D$ then takes one bounded step toward it.

When the day is done, the sums reset to zero and start accumulating for the new day.

To run the mechanism in full, eight values are necessary, and they fall into three groups.

The value measured every window:

• $v$: a window’s volume, corresponding to the actual notional dollar volume that executed against the standing book in that window, both directions summed. Because $v$ is read from fills, it adapts to the venue’s design: in a dual-flow auction every taker dollar trades against the book, while in a single-cross auction buys and sells net against each other first and only the imbalance reaches the book, which is then genuinely what the book absorbed. $D$ is the typical window’s demand, not the typical order’s size.

The two sums carried through the day:

• $S_1$: the sum of every window’s $v$, the day’s total flow.
• $S_2$: the sum of every window’s $v^2$. Together the two sums give the day’s reading of typical demand, $S_2/S_1$.

And the market’s configuration, fixed from day to day:

• $\alpha$: the chase speed, how much of the gap between $D$ and the day’s reading a single day may close.
• $m$: the minimum sample. A day carrying less than $m \cdot D$ of flow does not update.
• $L$: the daily limit. One day moves $D$ by at most a factor of $L$, in either direction.
• $D_{\min}$: the floor, the smallest value $D$ may take.
• $D_0$: the seed, $D$’s value at launch. It initializes and constrains nothing afterward.

Put together, the daily update is:

The step line reads from the inside out: $S_2/S_1$ is where the day says demand stands, its ratio against $D$ is the gap, $\alpha$ sets how much of that gap one day may close, and the $\min$ and $\max$ hold the result to $[1/L,\ L]$. The second line applies it: a day that carried the minimum sample moves $D$ by the step, never below the floor, and a day that did not leaves $D$ unchanged.

The Reasoning

The dollar-weighted average. Consider a fictitious day with two windows of trading, one clearing $2,000 and one clearing $8,000. There are two ways to average them. Averaged over windows, the typical window carried $5,000. Averaged over dollars, the answer is different: of the $10,000 that traded, four of every five dollars traded in the larger window, so the typical traded dollar sat in a window of $0.2 \times 2000 + 0.8 \times 8000 = 6800$ dollars. $S_2/S_1$ computes the second average ($S_1 = 2000 + 8000 = 10000$ and $S_2 = 2000^2 + 8000^2 = 68 \times 10^6$), and it is the one the walk needs, because the walk’s question is what it costs to trade where the trading actually is. It is also inexpensive to maintain: computing it requires no count of windows, an empty window adds nothing to either sum, and a window that traded costs only two additions to record.

The fractional chase. The exponent controls how quickly $D$ chases the day’s reading. At $\alpha = 1$ the update would jump the whole way in a single day, and $D$ would inherit each day’s noise along with its signal. Below one, the update closes only part of the gap and leaves the rest for the days that follow. At $\alpha = 1/2$, a reading at four times $D$ steps it up by $\sqrt{4} = 2$, and the gap that remains falls to its square root each day after; at $\alpha = 3/4$ the same reading steps it by $4^{3/4} \approx 2.8$ and converges in fewer days. Where in that range a market should sit is a calibration choice the design leaves open.

The minimum sample and the daily limit. These two conditions protect the tracking from days that would mislead it. The minimum sample keeps a dead day from re-measuring anything: a day with only a handful of fills says very little about typical demand, so $D$ holds unchanged rather than update from it, and nothing decays while it waits. The daily limit caps how far any single day can move $D$, however strange that day’s trading was, a single enormous window included.

The floor. The floor exists for the reading’s sake rather than for the tracking. $D$ sets the depth the walk measures at, and a walk that has shrunk too far reads only the quotes at the touch, which is the size-blind midpoint $M$ was designed to avoid in the first place.

The lab below runs three copies of the machine over the same ten days of demand, one line per chase speed. The limit, the seed, and the floor are dials, and none of their values is settled design: they are calibration choices a venue would fit to its own market.

One Auction, Start to Finish

Every part of the mechanism has now been introduced in its own section. This section runs a single auction from start to finish. Once the window seals, measuring the book and matching the orders are independent of each other, both reading only the sealed snapshot, so a venue can run the two in parallel or simply perform everything top to bottom. Pricing the fills needs them both, and finalizing the window books the results. The chart shows what depends on what; the scheduling is the venue’s choice.

Each step below carries its computational cost. Three counts cover them: $n$ resting orders in the sealed book, $a$ accounts standing both a bid and an ask, and $\ell$ eligible levels in the winning candidate. $a$ is at most the number of accounts and $\ell$ is at most $n$, so every bound is against the book itself; most steps touch only a subset of it, and the bounds given are worst cases. The gray nodes are work a batch-auction venue performs whether or not this fee mechanism exists; only the rest is the mechanism’s marginal cost.

Seal the Auction

The auction closes with every resting order committed and fillable. Everything after is computed from this snapshot and published history; nothing else enters. Cost: $O(1)$, the snapshot is the book the venue already holds, kept price-sorted as the orders arrived.

Measure the Book

This branch turns the sealed book into the window’s reference price. It is also the branch a venue can omit: a window with no fills and no consumer asking for $M$ has nothing to price, and the reading keeps no state that needs maintaining, since it is recomputed from each window’s own book. A venue with downstream consumers of $M$, health series or reference prices, will compute it every window regardless.

1. Find the candidate eligible books. Per account, keep only the paired size standing within reach, and pool what survives by price level. Books near one another share a single candidate eligible book, books too far apart form separate ones, and a candidate whose best quotes stand more than $B + 2Z$ apart is dropped as incoherent. Cost: each candidate is found in a bounded number of passes over the quotes within its reach; $O(a \cdot n)$ in the worst case, $O(n)$ in the common one, where every account’s view converges to the same candidate.
2. Select the market. The largest candidate by eligible size wins, and $M$ reads only it. If no valid candidate exists, or two candidates that share nothing tie exactly, no fresh $M$ is possible this window. Cost: comparing at most $a$ candidate sizes, plus one $O(n)$ disjointness check in the exact-tie case.
3. Walk both sides. Consume $D$ dollars of the winning book on each side, best prices first, pro-rata at equal prices; depth the book does not hold is priced at the edge of the measured area. The two volume-weighted prices are the impact prices. Cost: $O(\ell)$, only the winning book’s levels.
4. Set $M$ and the band. $M$ is the midpoint of the two impact prices, and the band is $M \pm B/2$. The venue also records the impact spread, the gap between the two impact prices: tight when the walk was fed by real depth, wide when the book ran thin and the boundary fill had to price the difference. Downstream consumers gate on it as the measurement’s health reading. Cost: $O(1)$.

Match

The venue’s own cross computes the fills at the uniform price. Fees never enter price priority, so nothing the mechanism computes affects the matching. Cost: with the book already price-sorted, the cross walks only the orders that trade, and it is the venue’s existing work either way, not the mechanism’s.

Price the Fills

Fees are charged only on fills, and a filled order’s rate depends on its account’s whole sealed book (its pairing can spill across levels), so base fees and pairing are computed per account with fills, never for the market at large. Each level’s base fee comes from its distance beyond the band edge; each account’s bid dollars pair against its ask dollars, best-priced first; every paired dollar then pays the worse of its two legs’ base fees, every unpaired dollar pays $F$, and the fee applies to the notional that traded. The fee never changes how much traded: the matched quantity is fixed by the cross, and the fee is deducted from what the fill delivers at settlement, so no separate balance or reservation is involved. Cost: linear in the books of the accounts that filled, bounded by $O(n)$, and untouched by the books of everyone who did not trade.

Finalize the Window

The window’s executed volume adds to the demand sums $S_1$ and $S_2$, at $O(1)$. This is later used at the daily close when $D$ takes its one bounded step toward what they read. The state of $M$ carries across windows: fresh when a candidate won, held at its last value when none did, and absent entirely before a market’s first candidate forms, where every fill pays the cap. The venue publishes the sealed book, the fills, $M$ with its state, and the impact spread, and the window settles as one unit: the fills, their fees, and the day’s sums land together. From what is published, any participant can reproduce every base fee, every pairing, and every final rate; no step depends on operator judgment or on data the venue holds privately.

Summed, the mechanism’s marginal cost is the measuring branch and the pricing of fills: $O(a \cdot n)$ in the worst case, $O(n)$ in the common one, and nothing at all in a window where nothing traded and nothing asked. There are no convergence loops anywhere; every step touches each order a bounded number of times, which is what lets the schedule price at auction cadence and lets any participant re-run a window’s arithmetic as fast as the venue does. One implementation note: the demand sums accumulate squared volumes, so fixed-point integers must be sized for $S_2$.

Design Notes

These notes cover where the market-order rate stands and what its distance from the cap can fund, how Snapshot Fees composes with existing practice, what it leaves flexible, what it costs, what $M$ could serve beyond fees, and which auction design fits it best.

The Market-Order Rate

The schedule above prices resting liquidity and never fixes what a market order pays. The only constraint is that the market-order rate stand at or above the cap $F$. Setting the two equal is the simplest choice. Setting the market-order rate above the cap opens room the next notes use: the gap can fund negative maker fees and gives volume tiers room to discount.

Negative Maker Fees

Nothing requires the resting schedule to stop at zero. Shifted downward, the best-placed, fully paired liquidity is paid for its fills rather than charged. For this to work, what the filled market order pays must exceed the rebate paid out, and the excess is best kept a strict premium rather than a full transfer, so that a self-cross always burns money and wash trading is never free.

A natural construction is a flat market-order rate against the scaled resting schedule, with the venue collecting the difference, so its take moves inversely with the quality of the book it fills. With a 5bps market-order rate against a resting schedule running from 3bps of rebate at the center to 1bps at the cap, the venue nets 2bps to 4bps per trade. If the schedule instead ran down to a 5bps charge at the cap, the take would run 2bps to 10bps. In whichever fee schedule used, any rebates settle in-window from the same window’s market-order fees, and none are paid before a market’s first $M$ forms; until placement can be measured, everything rests at the cap.

Rebates also require knowing which side of a fill was resting. A batch has no aggressor, so the venue must classify, by declaring order classes at entry or by seniority, an order counting as resting once it has stood through a seal without filling. Any scheme must satisfy one requirement: no fill collects a rebate on both sides. The reading carries no seniority in any of this; a two-sided book standing at the seal votes on that window’s that window’s $M$ in full.

Volume-Based Discounts Compose

Nothing in this design requires abandoning volume tiers, and they can attach to any rate: the market-order rate, the cap, the resting schedule. The composition works under one rule:

Volume lowers rates toward floors; it never reorders them.

Concretely, two invariants survive every tier. At equal volume, resting liquidity stays the better deal; no discount makes a market order cheaper than standing in the book. And where fees run negative, no combination of an account’s own orders nets a profit; what an account pays on one side always exceeds what it can collect on the other. A market might run its resting schedule from a 5bps cap down to a 2bps rebate at the center, with volume lowering only the ceiling, from 5bps toward 2bps; its market-order rate might run from 10bps down to 4bps on volume alone. The ordering holds at every tier, and no pairing of the rates nets a gain. One recalibration comes with the venue rather than the mechanism: a batch auction prints less volume for the same natural flow, since the latency-race volume is gone, so volume thresholds would be set against a smaller total.

Market Fee Profiles

Each market needs a calibration, but bespoke settings per market are unlikely to be the end state. More likely is a small family of vetted fee profiles, perhaps two to five, correlating with volume and volatility: deep markets running tight bands and strict precision pricing, volatile ones running wider bands and gentler slopes.

Extending this further, a market could shift its profile as conditions change, by governance or by an automated rule: $M$ already recomputes per window and $D$ per day, so the band, the zone, and the rest could follow slower metrics of the market’s own volume and volatility. All of it stays market state.

Maker Fee Uncertainty

On every venue running today, a maker knows their rate before they quote. Here the rate is known only up to the Measured Price: the schedule is fixed, but where a quote falls on it depends on where $M$ lands, so the fee is an interval rather than a number until the window seals. That is a real cost, one more variable to model alongside fill probability and adverse selection. Its bounds are also real: the schedule and formula are fully public, the worst case is the cap $F$, and the interval narrows with the same size and tightness the schedule already rewards.

Consuming the Measured Price

The mechanism’s byproduct is a reference price. Each window exports $M$ with its state and the impact spread, all recomputable from published data, and a price that is measured at size and expensive to move is a candidate input for services beyond fees: settlement prints, indexes, and eventually higher-stakes consumers.

At the highest-stakes end of that spectrum, a liquidation engine should never consume a single window’s $M$ directly; it would want machinery built around $M$, smoothing across windows, staleness bounds, health gating, so that moving the reference would mean sustaining a distortion across the whole lookback rather than winning one window. All of these are extensions around $M$, not changes to it; the per-window measurement stays exactly as described.

The Dual-Flow Fit

The mechanism runs on any batch auction, but a dual-flow venue fits it naturally. Its flows are declared rather than inferred, so the standing book is an explicit, labeled object and $M$ reads only it. Its fills split pro-rata with no seniority carried across windows, which is already how the walk shares and how the fees judge. And its takers always trade against the standing book rather than netting against each other, so gross flow walks the maker book, the flow the fee slopes are calibrated for. The two also push in the same direction, since the venue’s structure favors tight quotes at size before any fee enters and the schedule pays for the same behavior.

Conclusion

Snapshot Fees prices each auction’s liquidity from the auction itself: one snapshot, one schedule, and a rate set by what the filled liquidity offered the market. Whether anyone would run it cannot be settled on paper.

Adoption is likeliest as part of a new venue rather than a retrofit, and whether the machinery, light as it is, earns its place there is the venue’s judgment. For regular traders the mechanism is mostly invisible: they pay the market-order rate, volume discounts intact, though whether it feels that simple to them is another matter. The real question belongs to makers, and it is a genuine balance. A batch venue removes latency-driven adverse selection, and this schedule removes the volume gate at the bottom of the fee ladder, so a new maker can compete on quoting alone from the first window. Against that, the fee machinery is novel, and modeling a rate that is an interval rather than a number is a real switching cost. Where that balance lands is for makers to decide.

The piece opened with two questions: whether the snapshot could price orders directly, and whether doing so would improve the market. The first is answered by the mechanism’s construction. The second can only be answered empirically, by a market where the schedule runs and the comparison that follows. If the mechanism works as argued, the evidence would show in the book: tighter spreads, greater depth at the touch, and new makers appearing in markets that previously had none.