"What does this strategy return" is unanswerable without saying how much money it holds. The four paths' ceilings differ by three orders of magnitude, and the most attractive one is often the smallest. Some of the numbers below are small enough to end a discussion.
| Path | Bound by | What happens when you scale |
|---|---|---|
| Directional | book depth | your orders move the price and eat their own edge |
| Market making | rebate pool × your fill share | more competitors, smaller share |
| Arbitrage | remaining inventory | the count is fixed; extra capital has nowhere to go |
| Subsidy | no cap, but diminishing returns | chasing a tier requires size, and slippage eats the tier |
Full-level replay across more than 1,600 short-tenor windows:
| Percentile | Top-of-book size |
|---|---|
| median | about $94 |
| 25th | $37 |
| 10th | $13 |
From which the order size that extinguishes the edge follows:
| Basis | Net return reaches zero at |
|---|---|
| Counting book-walking only | about $200 |
| Plus price drift during execution | about $90 |
An edge has a capacity, and on short-tenor boards it is often only a few hundred dollars. Scale ten times and the edge may already be zero or negative — the strategy did not stop working, you overran it.
The right question is not "what does this return" but "at what size does it still return". Different questions — and a backtest answers only the first by default.
Market-making income has a hard bound: total taker fees × the rebate share, divided among participants by fill share. It is a closed pool — however hard you work, you can only take part of it.
| Category | Best wallet achieves | Population median |
|---|---|---|
| Crypto up/down market making | about $95/day | −0.74¢/share (47 wallets) |
| Sports market making | about $545–3,739/day | — |
On crypto up/down market making, being the single best participant in the ecosystem earns $95 a day. And the population median is negative — most people doing this are losing money.
More telling still: the wallets at the top of the maker-rebate table are almost entirely in sports, with not one of their recent thousand-odd fills in crypto up/down. The people best at this business have voted with their feet against the category.
Sports is an order of magnitude higher — but that is a different business with entirely different capability requirements.
This path's capacity is not set by your capital but by how many opportunities remain. I censused the lane: live executable inventory was zero.
So its ceiling is $0 — not "low returns" but nothing to do.
Which generalises: an opportunity requiring no judgement and paying a certain return will have its capacity driven to zero, because it is equally obvious to everyone and the fastest participant clears it. If you can still see it, that is usually because it cannot be executed, not because nobody noticed.
| From | To | Gain | Cost |
|---|---|---|---|
| Gold (18%) | Platinum (32%) | +0.24pp | roughly 5× the stake, paying 0.45–1.5pp of slippage |
The cost is two to six times the gain. So this path's real ceiling is not in the tier table — it is in how large an order your strategy can absorb, which returns you to §2.
| Do | Do not |
|---|---|
| Ask "how much can this path hold per day" first | ask "what is the annual return" first |
| Read the population median | read the best wallet's result |
| Compute returns at the size you can absorb | assume unlimited liquidity |
| Establish what the best participant earns | assume you will beat the best participant |
Row one has a very practical use: if a path's ecosystem ceiling is $95 a day, then however good your method, it cannot support a full-time occupation. That judgement can be reached before you start, and it costs two lookups.
A return figure without a capacity figure means nothing. A path whose ecosystem ceiling is $95 a day with a negative population median should not be entered however good the backtest looks — and that judgement can be completed before you write a line of code.