impact PnL

zero spread at 1000× only works if tiny moves do not pay in full. so a win is trimmed. small moves are trimmed hard and big moves barely. a loss is not trimmed. you lose what the price took and nothing more.

all of it is integer math in the program. nothing below is an estimate.

the win

a winning close runs five steps:

size  N   = margin × leverage
move      = favourable price change / entry price
m         = move − deadband                 if m ≤ 0, the win pays 0
raw       = N × m
G         = raw × (1 − b) × m / (m + c)     the trimmed gain
fee       = ceil(2% × G)
W         = G − fee                         what you keep on top of your margin

your margin always comes back first, in the same transaction, and is never trimmed. only the profit W can wait in the line.

the scale applied to the raw gain is (1 − b) × m / (m + c). it is the documented form (1 − b) / (1 + 1/(m·r) + N_ref/(10^6·m·p)) multiplied through by m.

the parameters

symbol	meaning	BTC	ETH	SOL	unit and bounds
deadband	the first part of a win that pays nothing	0.2 bp	0.2 bp	0.2 bp	stored as db_fp 20,000,000 of FP. bound 0 to 2 bp
b	the ceiling of the trim	1000 bps	1000 bps	1000 bps	10%. at most 2000 bps
r	curve rate	15000	15000	15000	1000 to 1,000,000
N_ref	reference size	$100,000	$100,000	$100,000	$1,000 to $10,000,000
p	per asset position multiplier	814.598	483.979	set at launch	stored as p_milli. 10,000 to 10,000,000
c	derived constant	189,426,606	273,287,203	set at launch	in units of 1e-12. must stay between 10^7 and 10^10
bust buffer	distance before margin hits zero	5 bp	5 bp	5 bp	1 to 9 bp

the devnet SOL listing uses ETH's p as a placeholder. no SOL calibration exists yet. mainnet values for all three: set at launch.

c is computed as ceil(1e12 / r) + ceil(N_ref × 1000 / p_milli), rounded up so the scale is never larger than the exact one. for BTC that is 66,666,667 + 122,759,939 = 189,426,606.

what the curve does

• for a big move, m / (m + c) goes to 1 and the scale goes to 1 − b, which is 90%. a huge win keeps at most 90% before the fee.
• when m equals c, the scale is 50% of 1 − b. for BTC c is about 1.9 bp, so a win of about 2 bp past the deadband keeps about 45%.
• for a tiny move, the scale is about m / c. it falls to zero as the move does.

the examples

BTC long, entry $83,000, $100 margin at 1000×. every number is from the exact integer function.

move	exit	raw gain	after the 0.2 bp zone	trimmed gain	share of raw	fee 2%	you keep
+0.0015%	$83,001.245	$1.50	$0.00	$0.00	0%	$0.00	$0.00
+0.005%	$83,004.15	$5.00	$3.00	$0.369143	7.38%	$0.007383	$0.361760
+0.1%	$83,083	$100.00	$98.00	$73.913146	73.91%	$1.478263	$72.434883
+0.5%	$83,415	$500.00	$498.00	$431.776320	86.36%	$8.635527	$423.140793
+1%	$83,830	$1,000.00	$998.00	$881.469167	88.15%	$17.629384	$863.839783
+5%	$87,150	$5,000.00	$4,998.00	$4,481.215975	89.62%	$89.624320	$4,391.591655

the first row is inside the deadband, so it pays nothing. the +0.5% row is also the worked example 15.1(a) of the spec.

the same numbers for a short work the same way when the price falls.

the loss

loss = min(margin, ceil(N × adverse move / entry))     rounded up
you get back  margin − loss, at once

the loss is rounded up, so you pay the rounding, not the house. it can never exceed the margin.

example: BTC −0.03%, exit $82,975.10. you lose $30.000000 and $70 comes back.

the bust

each bet has a bust price, set one bust buffer (5 bp) before the margin hits zero:

distance = 1 / leverage − 5 bp
bust     = entry × (1 − distance)        long
bust     = entry × (1 + distance)        short

leverage	distance to bust
1000×	0.05%
100×	0.95%
10×	9.95%
1×	99.95%

the distance is rounded toward the entry, so the trigger is never farther than the exact figure. a tick at or past the bust dusts the bet and the whole margin is forfeited.

example: 1000× long at $83,000 busts at $82,958.50. a tick at $82,950 dusts it. the bounty is $0.10 and the house side fee is $2.00 if the line is empty.

fees

fee	rate	taken from	when
open	none		never. no entry fee, no funding, no spread
win	2% of the trimmed gain, rounded up	your gain	only if the line is empty and the win is paid in full and the house holds the cash for the fee
loss	2% of the loss, rounded down	the house side, not on top of your loss	only while nobody waits in the line
bust	2% of the margin, rounded down	the house side	only while nobody waits in the line
activation	$1	your first deposit	once

the win and loss fee are owner settable, never above 500 bps (5%) each.

a win that has to wait in the line pays no fee. take profit and stop loss add no fee.

where a fee goes

a fee is split once, when it is paid:

stakers  = floor(fee × STAKER_SHARE_BPS / 10000)       STAKER_SHARE_BPS = 5000
dev cut  = fee − stakers
referrer = floor(dev cut × REF_BPS[tier] / 10000)      20% at tier 0, 40% at tier 1
dev kept = dev cut − referrer

the referrer comes out of the dev cut only. stakers and the house see no change.

worked split of the +0.5% win fee, 8,635,527 micro USDC:

to	no referrer	tier 0 referrer	tier 1 referrer
stakers	4,317,763	4,317,763	4,317,763
referrer	0	863,552	1,727,105
dev	4,317,764	3,454,212	2,590,659

orders and the price they use

• take profit. closes at exactly the trigger price, on the win path. the overshoot of the crossing tick stays with the house.
• stop loss. closes at the worse of the trigger and the tick that crossed it, on the loss path. a gap through a stop is paid by the trader, not by the house.
• limit order. fills at exactly the limit price. the jump past it stays with the house.
• a tick that is at or past the bust belongs to liquidate, never to a stop or a take profit.

DUST MODE

a DUST MODE round is binary. the payout is between −1× and +1× the margin.

g0 = min(L × move, 1)                  the move scaled by leverage, capped at 100% of margin
g1 = max(g0 − L × deadband, 0)         the cell's deadband in margin units
mh = g1 / L                            the price move that g1 equals
G  = M × g1 × (1 − b_dm) × mh / (mh + c_dm)
fee = ceil(2% × G)

adverse move:  loss = min(M, ceil(M × L × move))    there is no deadband on a loss
g1 = 0:        push, margin back, no fee, no $DUST

b_dm, c_dm and the deadband are set per cell: one for each leverage (5000×, 10000×) and horizon (5 s, 15 s, 60 s) pair. with c_dm at 0 the haircut is flat: G = M × g1 × (1 − b_dm).

examples with the pre calibration cell, deadband 0.1 bp and no haircut:

round	move	gain before fee	fee	you keep
10000×, $100	+1.3 bp	$90.00	$1.80	$88.20
5000×, $100	+1.3 bp	$60.00	$1.20	$58.80
10000×, $100	−1.3 bp	lose $100		$0
10000×, $100	+0.05 bp	push		$100 back

the devnet cell values are in parameters. a round can lose at most its margin and win at most its margin, so the house has no gap risk on it.

rounding

the rule is one line: whatever you receive rounds down, whatever you pay rounds up, a bust trigger rounds toward the entry, and emissions round down. the test suite checks it against exact fractions.