A conversion price is not a fact you look up. It is the output of a machine that someone built. Two machines dominate, and they fail in opposite directions.
When you run a balance conversion — say you're holding USDT on Tron and you want to settle in USDC on Base — a number comes back. It looks like a quote. It isn't a quote in the way a fiat exchange rate is a quote. It's the answer a market structure gave to the question "what is the most someone will pay, or the least someone will take, right now, for this size." There are two structures that produce that answer, and they reason about it completely differently. This post is the conceptual version of that difference, written for the merchant deciding what to settle in — not the trader running the desk.
We're not quoting a single rate here. We're explaining how the rate gets made.
The orderbook: a list of intentions
An orderbook is exactly what it sounds like — a stack of resting orders. Some people have said "I will sell at this level or higher." Others have said "I will buy at this level or lower." The highest bid and the lowest ask sit facing each other, and the gap between them is the spread. When you convert, you cross that spread: you take the best price someone has already committed to, then the next best, then the next, walking up the book until your size is filled.
The mental model that matters: the orderbook prices your trade against stated intentions. Every level is a real order a real participant placed. If a lot of people want to sell at a tight cluster of levels, your conversion fills cheaply and predictably. If the book is thin above the touch — a few small orders, then a gap — your fill walks into worse and worse prices, and the price you end up with is an average across all the levels you ate.
This is why orderbook venues feel sharp for liquid pairs and brutal for illiquid ones. A deep book on a major stablecoin pair barely moves when you cross it. A shallow book on an off-the-run asset gives you a headline price at the touch and a real price several levels worse.
The AMM: a formula instead of a counterparty
An automated market maker has no resting orders and no counterparty waiting on the other side. It has a pool — a reserve of two assets — and a formula that decides the price from the ratio of those reserves. The canonical one is the constant-product curve: multiply the two reserve amounts together, and that product must stay constant through your trade. You put asset A in, the pool's A reserve grows, its B reserve must shrink to keep the product constant, and the amount of B you get out falls out of that math.
The mental model: the AMM prices your trade against a curve, not against people. Nobody had to want your trade. The pool will always fill you — there's no "the book ran out." But the more you trade relative to the pool's depth, the further you push along the curve, and the curve gets steeper as you go. That steepening is slippage, and on an AMM it's deterministic: same pool, same size, same answer, every time, with no human on the other side deciding whether to show up.
The orderbook can refuse you by being empty. The AMM can never refuse you — it can only make you pay more as you push the curve. Those are different failure modes, and they matter for different sizes.
Why this is the merchant's problem, not just the desk's
You might reasonably ask why a merchant settling invoices should care which machine is upstream. The answer is that the two structures behave differently exactly where your money lives: at small sizes on liquid pairs, and at the long tail.
For a routine conversion between two deep stablecoin pairs, both structures give you a number that's close to fair, and the difference is noise. Where they diverge is size relative to available depth and how off-the-run the pair is.
- On a thin orderbook, a large conversion eats through the visible levels and you find out the real price only after you've crossed them.
- On a shallow AMM pool, the same large conversion pushes far along the curve and the formula quotes you a price that's mechanically worse the bigger you go.
- On either, a tiny conversion against deep liquidity is approximately free of structure effects — the spread or the curve-step is too small to feel.
The practical reading: the headline price you see is most trustworthy when your size is small relative to the depth behind it, and least trustworthy when it isn't. Neither machine lies to you. They both tell you the truth about a market that may simply be too thin for the size you're trying to move. That's the same logic that governs which settlement currency is comfortable to hold — assets with deep two-sided liquidity convert predictably; assets without it convert at the mercy of whichever structure is upstream.
What "deep liquidity" actually buys you
People say "deep liquidity" like it's a vibe. It's not. It's a measurable property with a concrete payoff: the price you're quoted survives the size you trade. On a deep orderbook, the levels above the touch are fat, so crossing them barely moves your average. On a deep AMM pool, the curve near the current ratio is nearly flat, so pushing along it barely changes the rate. Depth is the thing that makes the headline number and the filled number agree.
Shallow liquidity is the opposite — and it's shallow for boring reasons, not sinister ones. Newer assets, smaller chains, and odd cross-asset pairs simply have fewer participants standing ready and smaller pools. The structure upstream didn't get worse; there's just less of it.
The reorg-aware, per-chain confirmation behavior we've written about elsewhere is about whether a payment is final. Liquidity is the adjacent question: once it's final and credited, what does it cost to reshape it into the asset you want to hold? Both questions decide whether settling in a given asset is comfortable or a recurring tax.
The operating rule
You don't get to choose the machine, and you mostly shouldn't try to. What you can do is read its output correctly:
- Treat the quote as size-dependent, not absolute. A good price on small size says nothing about a good price on ten times that size.
- Favor pairs with depth behind them. Two assets that lots of people trade against each other convert predictably regardless of which structure is upstream.
- Be suspicious of clean headline prices on off-the-run assets. That's exactly where a thin book or a shallow pool hides the real cost a level or two deep.
If you want the mechanics of what conversion is and where it sits in the platform — that it moves value between crypto assets only and is not a fiat off-ramp — start with what balance conversion is. This post is the layer underneath: why the number it hands you is the output of a structure, and how to read that structure before you trust the number.
The structure is doing its job. Your job is knowing which one is answering.
M. Voss, network research