Market structure

Why an order book, not a pool

Tokenised equities arrived on Solana at real scale and kept the market structure of a memecoin. This is the case for the other one — and why the argument stops being a preference and becomes a requirement once the asset has no published price at all.

1. The assets arrived. The market structure did not.

Tokenised stocks on Solana did roughly $5.77B in Q2 2026, about 96% of the category across all chains. Ondo Global Markets listed 200+ US stocks and ETFs; xStocks carries 130+ instruments. Whatever one thinks of the idea, the demand is no longer hypothetical.

Almost all of that flow sits in constant-product pools. Every equity market in the world runs on a central limit order book — that is not tradition, it is what the instrument requires — and on-chain the same asset trades on a curve designed for tokens with no external price. The asset class was ported. Its market structure was not.

2. What a pool is actually doing

A constant-product AMM does not form a price. It quotes a curve around its current inventory ratio and waits to be corrected. When the outside world reprices the asset, the pool is still quoting yesterday's number, and an arbitrageur takes the difference. That transfer has a name — loss-versus-rebalancing — and it is not a bug in a particular pool, it is the mechanism working as designed.

This is tolerable when the asset is volatile in both directions and the fee income covers the bleed. It is much less tolerable for an equity, which spends most of its time flat and then gaps on an earnings call, a guidance revision, a lawsuit. The pool cannot step back before the gap because it cannot cancel. A maker on a book can, and does, which is the entire reason market makers exist as a profession.

3. Pre-IPO breaks the assumption completely

Everything above assumes there is an outside price the pool is lagging toward. For a private company there is no exchange forming one. For Anduril, Neuralink, Kalshi and Polymarket nothing publishes a price at all.

For OpenAI and Anthropic the picture is more interesting, and worth stating precisely rather than rounding off. Pyth does publish a feed for both — Equity.Index.OPENAI/USD and Equity.Index.ANTHROPIC/USD, 24/7. But look at what it is: an index, not an exchange price, because there is no exchange to take a price from. It is derived from wherever these claims actually change hands — secondary venues, tokenised markets, broker indications. It reports price discovery. It does not perform it.

Honest note. This matters for the argument, so we would rather be exact than rhetorical: a price for OpenAI does exist, published by Pyth. What does not exist is a venue where that price is formed. An index is a mirror; somebody still has to be the market it reflects.

Either way the AMM's implicit contract fails at the first clause. Where nothing is published, an LP who seeds a pre-IPO pool is not providing liquidity around a known price; they are asserting one, with capital, against anyone who disagrees. Where an index exists, anchoring to it is circular: the index is derived from the thin venues the pool is supposed to be improving on, so the pool ends up quoting a reflection of itself. Neither case is price discovery.

An order book has no such dependency. It does not need to know what the asset is worth. It collects what people are willing to pay and accept, sorts them, and the best of each is the price — which is exactly how price discovery has always worked, and the only mechanism that works when there is nothing to copy from, or nothing worth copying.

4. Thin floats favour books, structurally

A pool must hold inventory across every price simultaneously. That is what the curve is: capital committed at prices nobody asked for. To show a usable spread it therefore needs depth, and pre-IPO floats are thin by construction — these are secondary claims on private shares, not a free-floating supply.

A book inverts that. A maker commits capital only at the price they chose, cancels when they change their mind, and pays nothing for the levels they are not quoting. That is why order books function in markets far too thin to support a pool at all — corporate bonds and small caps have traded this way for a century on a fraction of the capital a pool would need.

PreStocks state the consequence on their own site: their tokens carry “no guaranteed secondary-market liquidity.” That is a market-structure problem with a known solution, not a fact of nature.

5. The same order, both ways

The section below is computed live, not asserted: it walks the real resting book on devnet and prices the same quantity against a modelled constant-product pool.

Why a book, not a pool

A curve has to guess. A book already knows.

An automated market maker quotes off a formula anchored to a reference price. For a private company there is no reference price to anchor to, and the float is thin, so the curve is steep exactly where it matters. A book quotes where makers are actually willing to trade.

Buyofagainst a

reading the book …

Trade it on devnet

6. What a book buys you that a curve cannot

A real limit order.“Buy at 180 or better” is the most ordinary instruction in equities and a curve cannot express it. You get the curve's price or you do not trade.

Price-time priority. The maker who quotes first and tightest is filled first. That rule is what makes anyone quote tight; without it there is no reward for competing on price, so nobody does.

The right to change your mind. A maker cancels. A pool cannot, and that single asymmetry is most of what LVR measures.

A depth curve that means something. A book tells you plainly that it cannot fill your size. A pool always fills you — at a price that silently gets worse the larger you go. For a venue that wants to serve borrowing and liquidation later, a real depth curve is the thing you size against.

7. When a pool is the right answer

The honest version of this argument has to include where it loses. A constant-product pool is better when there are no makers at all: it is always quoting, with nobody at the keyboard. For a long tail of assets that nobody wants to make a market in, a mediocre always-on quote genuinely beats an empty book.

Pools are also simpler to integrate, compose trivially with routers, and for very small orders on a deep pool the slippage difference is noise. And a book with no makers is worse than useless — it looks broken. An order book is not free; it is a venue that has to be populated, which is why maker incentives are on our roadmap rather than an afterthought.

Honest note. Our claim is narrow and we would rather state it narrowly: for assets that have a genuine two-sided interest and no reference price, a book is the correct instrument and a curve is the wrong one. We are not claiming AMMs are obsolete.

8. Why this was hard on Solana until now

If books are so clearly right for this asset, the obvious question is why Solana has so few. The answer is structural, not ideological. The classic on-chain book keeps the whole thing in one account per side, so every maker write takes the same write lock and Sealevel serialises them: one writer per slot, however many quotes arrive. A venue whose makers re-quote constantly is exactly the workload that design punishes hardest.

TornaCurb runs on Torna, our own open-source index, where every B+ tree node is its own account. Quotes at different price levels touch different leaves, carry disjoint write sets, and commit in the same slot. The engineering case — prior designs, the alternatives we rejected, the measurements — is written up separately.

Read the engineering research →Or just trade it on devnet

9. What makes this real

Everything above runs on devnet against mock share tokens, and the honest question is what stands between that and a venue people actually trade on. Two things, in order, and the second is harder than the first.

The audit is the gate. The engine and the order book have been through in-house adversarial review, and a September 2026 pass still found and fixed critical issues, which is the argument for an external audit; it is pending. We are not putting an unaudited matching engine in front of real collateral, and no amount of hackathon pressure changes that. Nothing about the venue changes once it clears — listing a real token is a config row, not a protocol change, because the venue reads mints from a table and has no opinion about which ones.

We are not issuing anything. That distinction is worth stating plainly, because it is what makes the path walkable. The shares are issued, custodied and legally wrapped by their issuers — every pre-IPO name here is a PreStocks token. Those tokens already exist and already trade. TornaCurb does not create exposure to a private company — it gives exposure that is already trading somewhere worse a better place to trade. A matching venue for existing tokens is a different undertaking, legally and operationally, from an issuer.

Liquidity is the real gate. An order book with no makers is worse than a pool: a pool at least always quotes, while an empty book just looks broken. So the venue does not open with ten listings on mainnet. It opens with one, with a maker who has committed to quote it, and widens only as fast as someone is willing to stand behind each new name. Rebates on resting volume are the standard way to pay for that, and they cost nothing until somebody actually quotes.

Which suggests where a curve does belong.A newly tokenised name has no makers yet, and we said earlier that an always-on quote beats an empty book in exactly that case. So the honest sequence is not book-versus-curve but curve-then-book: launch a thin listing on a bonding curve, where something has to quote before anyone will, and graduate it into the order book once there is two-sided interest to carry it. Meteora's DBC graduates into an AMM pool today; graduating into a book instead is the same mechanism pointed at the venue the asset ends up needing. We have not built that — it is the shape of the liquidity answer, not a claim about what exists.

Honest note. We would rather show you a working venue and name the two gates than claim a mainnet we have not earned. Both are gates we can walk through; neither is one we can skip.

Market-size figures are third-party reporting on Q2 2026 tokenised-equity volume and listing counts, not our own measurements. The pool comparison above is a model, clearly labelled as one: no constant-product pool exists for a pre-IPO name, which is rather the point.