We market made on BULK's mainnet on day one. Here is every number.

The specifications matter more than usual here, because they decide whether a small operator can take part at all. BTC-USD is the only market with a $1 minimum notional. The other nineteen sit between $25 and $50. Tick size on BTC-USD is 0.001 and maximum leverage is 40×.

That single fact shapes everything below. The one market cheap enough to test on is not the one worth earning on, and the markets worth earning on demand a position size that makes every experiment cost real money.

• Off-chain matching, Solana settlement, USDC collateral only.
• 20 perpetual markets, all trading, post-only available everywhere.
• BTC-USD: $1 minimum notional. Every other market: $25–50.

Two full bot sessions followed. The first quoted BTC with a $13 clip at 2× leverage and ran just under seven minutes: five fills, $77.71 of turnover, 100% maker, zero fees, −$0.0009 realised, ending flat. The second used a $31 clip at 4× for five minutes: two fills, $61.88 turnover, 100% maker, zero fees, −$0.0005 realised, also flat.

Seven fills. $139.59 of turnover. Fourteen hundredths of a cent of loss. We are publishing that number rather than the round trip that made money, because two sessions of a live bot are the honest sample and a single hand-run cycle is not.

• 7 fills, $139.59 turnover, 100% maker, $0.0000 in fees.
• Net result: −$0.0014. Both sessions ended flat with no open orders.
• Every position was closed and verified before the session was recorded as stopped.

Here is the arithmetic that makes this venue interesting at all. Our round trip captured 4.35 basis points of spread on the leg, and both legs filled at exactly the limit price with no slippage. Had both legs crossed as taker at 3.5 bp, the cost would have been $0.000725, more than the trip earned. The zero maker fee is not a nice-to-have here. It is the only reason the economics close.

There is also a market-share rebate that scales with your share of each market, and it had accrued nothing for us: makerSharePpm 0, makerShareRebateBps 0.0. Rebates need volume before they mean anything, and we had none.

• Maker 0.0 bp at tier 0 with a zero threshold: the base rate.
• Taker 3.5 bp. Both legs as taker would have sunk the round trip.
• Market-share rebate exists but had accrued nothing at our volume.

The check is not degenerate. Something moved, so this is not zero equalling zero, a mistake we had made on their testnet earlier.

One honest limit on that claim. The exact figure by price was 0.000450619 and the venue recorded 0.00045, so it rounds its accounting at roughly 1e-5 USD and carries that rounding into collateral consistently. The identity is proven to 1e-5 USD, not to eight decimal places. The extra digits come from the output format, not from precision.

• Collateral delta, realised PnL, fees and funding reconcile with zero residual.
• Proven to 1e-5 USD, the venue's own accounting resolution.
• On a $1 clip that quantisation is worth up to about 5 ppm; on $350, 0.014 ppm.

BTC-USD, the market we traded, carries a one-tick spread 98.7% of the time. There is nowhere to stand inside it. ETH-USD was worse in a different way: its book did not move for over ten minutes, confirmed across two separate observation windows. SOL-USD sat at 1.46 bp, also collapsed.

The wide spreads are elsewhere. MEGA-USD showed a 23.5 bp median and stayed wider than 5 bp 100% of the time; JTO-USD 29.5 bp on the same terms; MON, JUP and ZEC between 21 and 31 bp. Those are markets with room to quote inside, and a $25–50 minimum notional.

So BTC is valuable for exactly one thing: it is the only book cheap enough to verify that your machinery works. It is not where the money is.

• BTC-USD: one-tick spread 98.7% of the time. Nowhere to stand.
• ETH-USD: book unchanged for 10+ minutes across two windows.
• MEGA, JTO, MON, JUP, ZEC: 21–31 bp median, but a $25–50 minimum notional.

That is one participant's grid, not a crowd. Quoting inside a spread like that is not competing with a market. It is playing against a single opponent who can see your order the moment it lands.

And here is what we did not measure, stated plainly: turnover and fill frequency. A wide spread with one market maker and a $9.5k book is explained equally well by "a market with no competition" and by "a market with no flow." Until that is measured, "there is room to quote on MEGA" is a hypothesis about profitability, not a finding.

• 17 of 20 books: about $9.4–9.7k a side, one order at the best level.
• Real depth only on ETH, SOL and partly BTC.
• Turnover was not measured. A wide spread does not prove opportunity.

Our measured median reaction from touch to re-quote is 5,324 milliseconds. That is three orders of magnitude slower than the window. The head start is real and we cannot take it with our current cycle. A snapshot TTL of three seconds and a loop that thinks in seconds cannot outrun a 25 ms door.

We publish this because it is the kind of thing a venue comparison normally skips. A protective mechanism you are structurally too slow to use is worth nothing to you, whatever the documentation promises.

• Taker orders delayed 25 ms; post-only and cancels pass immediately.
• Our touch-chase median: 5,324 ms, three orders of magnitude slower.
• The mechanism is described as temporary, to be replaced by a protective order type.

The delegated agent key cannot move funds out. Every attempt to withdraw externally, to authorise a second agent, or to divert proceeds returned unauthorized signer: a terminal rejection, not a rate limit.

But it can create a subaccount and transfer between the owner's own accounts. Both actions succeeded. Money cannot leave the owner; it can move around inside. That is a meaningfully different permission boundary from "the key can only trade," and on the testnet it is invisible.

The general lesson: prove a delegated key's permission boundary in the same network you will trade in. A testnet result is not evidence about mainnet, and we have the counter-example to show for it.

• Withdrawals, second-agent authorisation and proceeds diversion: all rejected.
• Subaccount creation and internal transfers: both permitted.
• Testnet permissions did not predict mainnet permissions.

A post-only order is rejected precisely when the price has moved away from it, which is exactly the moment an emergency exit exists for. The result was 143 seconds of an unhedged position on a live mainnet account while the exit kept bouncing.

The fix is not clever: an emergency exit crosses the spread. It is now an immediate-or-cancel order, and the response to a protective send is checked rather than assumed. We mention it because anyone building the same machinery will reach for post-only for the same reason we did.

• Post-only exits are rejected exactly when you need them most.
• Cost: 143 seconds of unhedged exposure on a live account.
• Emergency exits must cross, and the venue's response must be checked.

Zero maker at the base tier is rare, and it is what makes small passive quoting arithmetically possible at all. Against that: the liquid markets have no spread to capture, the wide markets carry one participant and unmeasured flow, and the minimum notional on everything except BTC makes each experiment cost real money.

What we would want before committing size, and have not got: measured turnover and fill frequency on the wide markets, and an adverse-selection number from the engine itself rather than from a script. Without a markout, any cost-per-million figure for this venue is an estimate wearing a decimal point.

• In favour: zero base maker fee, clean settlement, books that reconcile.
• Against: no spread where there is liquidity, no measured flow where there is spread.
• Missing: turnover per market and an adverse-selection measurement.