# How big is too big? Sizing a trade against the depth of a pool > Canonical: https://blazephoenix.xyz/learn/sizing-your-trade > License: CC BY 4.0 (attribution + link) · © 2026 BlazePhoenix > Updated: 2026-08-13 For a constant-product pool the fraction of value lost to impact is exactly d/(x+d) — your order against the pool's reserve of the token you pay with, and nothing else: about 1% of the reserve costs about 1%, ten percent costs about 9%, half the reserve costs a third of your money. Inverting it gives d = x·t/(1−t), the largest size that fits a target impact — and the reserve that matters is that one pool's, never the protocol's TVL. The previous article showed a single trade costing 9.59%, of which 9.09% was price impact — the cost of your own size. This article turns that into something you can decide with: given a pool, how large an order can you send before the cost stops being acceptable? The answer is one fraction, and you can compute it in your head. ## The one fraction Take a pool holding x of the token you are paying with, and y of the token you want. Ignore fees for a moment. If you pay in an amount d, the rule that the product must not shrink means the pool keeps x + d of your token and hands you back what the other side gives up. Divide what you receive by what you paid, compare it against the label price y / x, and almost everything cancels. What is left is: ``` fraction of value lost to impact = d / (x + d) ``` That is exact, not an approximation, for any pool that follows the constant-product rule. Check it against the number from the previous article: a pool with x = 1,000 and an order of d = 100 gives 100 / 1,100 = 9.0909%, which is precisely the 9.0909 tokens lost on a 100-token trade there. Read what the fraction says. Your loss depends on your order compared to the pool, and on nothing else. Not on the price. Not on the token. Not on the chain. A 100-token order into a 1,000-token reserve costs the same 9.09% whether the tokens are worth a cent or a thousand dollars. ## The table you can carry Same pool, x = 1,000, fee switched off so the effect stays clean: ``` order d/(x+d) you keep 1 0.0999% 99.90% 10 0.9901% 99.01% 25 2.4390% 97.56% 50 4.7619% 95.24% 100 9.0909% 90.91% 200 16.6667% 83.33% 500 33.3333% 66.67% ``` Two things are worth noticing before moving on. First, the cost is not proportional: going from 50 to 100 does not double the cost from 4.76% to 9.52%, it takes it to 9.09%, slightly less than double. Second, and more useful, the damage accelerates in a way that punishes exactly the trade a newcomer is most tempted by — the one that empties an account into a small pool. At 500 into a 1,000 reserve you are giving up a third of your money to arithmetic, before a single fee. ## Turning it around: the size that fits your tolerance Usually you do not want to know the cost of a size; you want the size for a cost. Solve the fraction for d and you get, for a target t: ``` d = x * t / (1 - t) t = 1.0% -> d = x / 99 (1.0101% of the reserve) t = 0.5% -> d = x / 199 (0.5025% of the reserve) t = 0.1% -> d = x / 999 (0.1001% of the reserve) ``` Check the first one so you are not taking it on faith. With x = 1,000, d = 1,000 / 99 = 10.101. Then d / (x + d) = 10.101 / 1,010.101 = 1.0000%. It closes. The rule of thumb that falls out is easy to remember and accurate enough to trade on: about 1% of the reserve costs about 1% in impact. Ten percent of the reserve costs about 9%. Half the reserve costs a third of your money. Note which reserve. It is the pool's balance of the token you are paying with, in that one pool. It is not the protocol's total value locked, which sums every pool a protocol operates and tells you nothing about the one you are trading through. A protocol with a billion locked can hold a pool with a thousand, and your order meets the thousand. ## What depth actually buys you Send the same 100-token order into pools of different sizes and the fraction does the arguing: ``` reserve x 100 / (x + 100) 1,000 9.0909% 2,000 4.7619% 5,000 1.9608% 10,000 0.9901% 100,000 0.0999% ``` Doubling the depth slightly better than halves the cost. Multiplying the depth by ten divides the cost by roughly ten. This is why the deepest pool is usually the right answer for a large order, and it is also the reason an aggregator that can reach more pools has something real to offer, rather than a slogan: it can find the reserve that makes your fraction small. It is also why the next article is about splitting. If one pool of 1,000 costs 9.09% for your 100, two pools of 1,000 might cost less than that — and the fraction above is what tells you so. ## Where this rule stops being true The fraction is exact for constant-product pools, and every pool is not one. Two families behave differently, and pretending otherwise would make this article a model rather than a measurement. Concentrated-liquidity pools let providers place their capital inside a price range instead of across all prices. Inside that range the pool behaves as though it were far deeper than its total balance suggests, so the fraction overstates your cost. But once your order pushes the price out of the range, the depth backing you can fall off sharply, and the fraction then understates your cost badly. Stable-curve pools do something similar for a different reason: near the point where the two assets are equal, they are much flatter than constant-product, and far from it they are not. Which means the rule of thumb is a way to think, not a way to price. The way to price is to measure. ## The two-quote test, which needs no formula at all Ask for a quote at your intended size. Then ask for a quote at a tenth of that size, and multiply the result by ten. Compare the two numbers. If the small quote scaled up returns 100.0 and your full-size quote returns 90.4, then your own size is costing you 9.6% and no theory was required to find it out. If they are within a fraction of a percent, the pool is deep relative to what you are doing and you can stop worrying about impact. This works on any curve, any venue and any chain, because it does not assume a formula — it asks the same contract the same question twice. That is the general discipline this site keeps returning to: where a measurement is available, prefer it to a model, including the model on this page. **Verify it yourself:** Request two quotes for the same pair, one at your full size and one at a tenth of it. Multiply the small result by ten and divide the full-size result by it. The shortfall is your price impact, measured rather than modelled. Related: https://blazephoenix.xyz/learn/three-prices · https://blazephoenix.xyz/learn/why-splitting-wins · https://blazephoenix.xyz/learn/price-impact-explained · https://blazephoenix.xyz/learn/what-is-tvl · https://blazephoenix.xyz/learn/amm-math-taxonomy