01/the transaction

20 atomic swaps.
One transaction.

A single signed transaction on Solana mainnet executed twenty isoFi swaps back-to-back, all-or-nothing, on one isolated market. No Address Lookup Table. Three quarters of the wire budget. Sixteen percent of the compute ceiling. This is what the rest of the network cannot do.

atomic swaps

total accounts

lookup tables

920/1232

wire bytes

224,050

compute units

16%

of CU ceiling

02/the three walls

Every transaction hits three hard limits

Swap-stacking dies at whichever wall bites first. iso-xmm is the only profile on Solana that sits clear of all three at once.

WIRE / PACKET

1232 B

The entire signed transaction. Every distinct account costs 32 bytes here before a single instruction is added. iso-xmm reuses the same 8 accounts across all 20 swaps.

used 920 B · 75%

COMPUTE

1.4M CU

Max compute requested per transaction. At ~11.2k CU per iso swap and no tick-array or bin walk, twenty swaps spend only 224,050, with room for ~124.

used 16% · headroom ×6

ACCOUNT CAP

256

Distinct accounts a v0 message may reference. Because all 20 fills hit one market, iso-xmm's set never grows past 8, so the cap is irrelevant.

used 8 · 3%

03/the field

iso-xmm vs the Solana liquidity network

Measured live from mainnet: real per-swap account footprint, real compute, and the most swaps of its own kind each program has ever been seen bundling in a single transaction.

DEX / AMM	Accounts / swap	CU / swap	Ceiling in 1 tx	Most ever bundled (mainnet)	Binding wall
iso-xmmOURS	8	11,200	~31 · 124 cu	20	compute (huge headroom)
Phoenix (CLOB)	9	21,300	~15†	1	wire / account
Lifinity v2	12	19,200	~20†	1	account cap
Raydium CPMM	13	30,000	~23†	1	account cap
Orca Whirlpool	14	38,700	~16	2	compute + account
Raydium CLMM	16	78,600	~15	2	compute + account
Meteora DLMM	17	30,200	~15	2	account (bin fan-out)
Raydium AMM v4 (OpenBook)	19	15,500	~14†	2	wire / 256-cap
Meteora DAMM (dyn-vault)	16	118,000	~11	2	compute
pump.fun AMM	25-27	97,700	~10†	2	compute + account

Accounts/swap and CU/swap are measured live from real mainnet swaps. Ceiling in 1 tx = realistic same-shape swaps before the binding wall. † requires a pre-built Address Lookup Table to approach that number; iso-xmm needs none. The heavy-liquidity majority of the network (CLMM, DLMM, DAMM, pump, OpenBook) is hard-walled at 2-17 by compute or account fan-out. In live mainnet traffic no other program has ever bundled more than 2 of its own swaps in one transaction; iso-xmm did 20.

04/the verdict

Can the others do 20-21? No.

Heavy AMMs: hard floor Meteora DAMM (118k CU/swap), pump.fun (98k CU, 27 accounts) and the CLMM family (Whirlpool, Raydium CLMM, PancakeSwap v3, byreal) are capped at ~2-17 by compute and account fan-out. Each tick or bin walk inflates both at once. They physically cannot reach 20.

Big-footprint AMMs: byte wall Raydium AMM v4 (19 accounts) and pump.fun (27) spend their entire 1232-byte budget on account keys: 608-864 bytes before one instruction. They need a pre-built Address Lookup Table just to reach the low teens.

Nearest rivals: still short The CLOBs (Phoenix, Manifest, 9 accounts) and a few light constant-product AMMs have the compute headroom, but for real routes the per-instruction byte tax and the 256-account cap wall them in the mid-teens, and every one of them needs an ALT iso-xmm never builds.

iso-xmm: the only profile 8 accounts, 11,200 CU, a 1-byte selector, same-pool reuse, zero ALT. The single profile on Solana that clears the byte wall, the compute ceiling, and the account cap simultaneously. One isolated market = one tiny fixed account set, so swaps compound where every competitor's accounts multiply.

20 atomic swaps.One transaction.

Every transaction hits three hard limits

iso-xmm vs the Solana liquidity network

20 atomic swaps.
One transaction.