Skip to main content

Economics and fairness

This game favours well-capitalised snipers, and a hider can shoot back

The price of a shot rises as squares run out, but the pool and the payout rise faster. A sniper who can afford to keep firing is paid more in expectation than it spends, and a sniper who can afford to fire every remaining square at once can make a profit that does not depend on luck (see the burst; at the 10% rake that margin is thin from a fresh round and grows as the board fills). A hider whose answerer is down for a whole answer window loses to a one-shot timeout bot. The hider's tools (a low base fee, an early extraction, a reliable automatic answerer) reduce the damage; none of them reverses the edge.

The chain cannot tell a hider's second address from a real sniper. A hider with a second address can exit without ever facing a hit and can collect real snipers' fees for about 10% of the collateral (see the second-wallet limitations). This is an accepted property of the design, not a bug waiting for a patch. Snipers in user-created (community) rounds are exposed to it. Rounds hosted by platform-approved addresses ("House rounds", see House rounds) remove the anonymous stranger, but the contract cannot prove that a House hider does not shoot its own round either. Nothing here is financial advice, the contract is not audited, and ZK-Sniper is not deployed.

All numbers on this page use the contract's own rules and the SDK's math (sniperPricingMatrix, sniperHitPayout, sniperEscapePayout), which an anvil test checks against ZkSniper.sol. They come from packages/sdk/scripts/sniper-ev.mjs (the same text is committed as packages/sdk/docs-sniper-ev.md) and are for a base fee b of 0.01 BNB. Every amount scales linearly with the base fee, up to rounding. Gas is not included. The hider's square is assumed to be uniformly random over the 64 squares; a hider who picks a predictable square does worse.

The rules the tables rely on​

  • The next shot costs b * 64 / remaining (rounded down), or half the pool once two or fewer squares remain. remaining counts every fired shot, whether it is pending or answered.
  • The hider's collateral 64 b seeds the pool. Every shot fee joins the pool first.
  • A hit, or a shot left unanswered past its deadline, pays that shot's sniper the whole pool minus a 10% rake. The sniper's own fee is part of what it wins.
  • Several shots can be pending at once, one per square. Only one can ever be the hit; the others are void and their fees stay in the pool.

1. A sniper who fires every square, one at a time​

Shot k is fired at square k after shots 0..k-1 all missed. "EV of shot" is the value of that one shot to a sniper who has reached it: the chance 1 / remaining times the payout, minus the price. "EV if stopping here" is the expected profit of the plan "fire shots 1..k and then stop", including the money already spent when none of the earlier shots hit. Negative is a loss. The model assumes the hider never extracts and no other sniper is playing.

The selected rows (the full 64-row table follows):

ShotRemainingPricePool beforePayout if hitP(hit)EV of shotSpent so farEV if stopping here
1640.01000.64000.58501.5%-0.00090.0100-0.0009
2630.01020.65000.59411.5%-0.00070.0202-0.0016
4610.01050.67050.61291.6%-0.00040.0410-0.0026
6590.01080.69160.63221.6%-0.00010.0625-0.0030
7580.01100.70250.64221.7%0.00000.0735-0.0029
8570.01120.71350.65231.7%0.00020.0847-0.0027
16490.01310.80940.74022.0%0.00200.18250.0048
24410.01560.92220.84402.4%0.00500.29780.0245
32330.01941.05930.97083.0%0.01000.43870.0586
40250.02561.23391.13354.0%0.01970.61950.1109
48170.03761.47481.36125.8%0.04240.87240.1877
5690.07111.86561.743011.1%0.12261.29670.3021
6050.12802.21482.108520.0%0.29371.70280.3842
6230.21332.50282.444533.3%0.60152.07610.4376
6321.3580 (duel)2.71613.666750.0%0.47533.43410.4525
6412.0371 (duel)4.07415.5001100.0%3.46305.47120.5066

All amounts are in BNB.

All 64 rows (base fee 0.01 BNB)
ShotRemainingPricePool beforePayout if hitP(hit)EV of shotSpent so farEV if stopping here
1640.01000.64000.58501.5%-0.00090.0100-0.0009
2630.01020.65000.59411.5%-0.00070.0202-0.0016
3620.01030.66020.60341.6%-0.00060.0305-0.0021
4610.01050.67050.61291.6%-0.00040.0410-0.0026
5600.01070.68100.62251.6%-0.00030.0516-0.0028
6590.01080.69160.63221.6%-0.00010.0625-0.0030
7580.01100.70250.64221.7%0.00000.0735-0.0029
8570.01120.71350.65231.7%0.00020.0847-0.0027
9560.01140.72470.66261.7%0.00040.0962-0.0024
10550.01160.73620.67301.8%0.00060.1078-0.0019
11540.01190.74780.68371.8%0.00080.1197-0.0012
12530.01210.75970.69461.8%0.00100.1317-0.0003
13520.01230.77170.70561.9%0.00130.14400.0007
14510.01250.78400.71691.9%0.00150.15660.0019
15500.01280.79660.72852.0%0.00180.16940.0033
16490.01310.80940.74022.0%0.00200.18250.0048
17480.01330.82250.75222.0%0.00230.19580.0066
18470.01360.83580.76452.1%0.00260.20940.0085
19460.01390.84940.77702.1%0.00300.22330.0107
20450.01420.86330.78982.2%0.00330.23750.0130
21440.01450.87750.80292.2%0.00370.25210.0156
22430.01490.89210.81632.3%0.00410.26700.0183
23420.01520.90700.83002.3%0.00450.28220.0213
24410.01560.92220.84402.4%0.00500.29780.0245
25400.01600.93780.85842.5%0.00550.31380.0279
26390.01640.95380.87322.5%0.00600.33020.0315
27380.01680.97020.88842.6%0.00650.34710.0354
28370.01730.98710.90392.7%0.00710.36440.0395
29360.01781.00440.91992.7%0.00780.38220.0439
30350.01831.02220.93642.8%0.00850.40040.0485
31340.01881.04040.95332.9%0.00920.41930.0534
32330.01941.05930.97083.0%0.01000.43870.0586
33320.02001.07870.98883.1%0.01090.45870.0641
34310.02061.09871.00743.2%0.01190.47930.0698
35300.02131.11931.02663.3%0.01290.50060.0758
36290.02211.14061.04643.4%0.01400.52270.0822
37280.02291.16271.06703.5%0.01530.54560.0889
38270.02371.18561.08833.7%0.01660.56930.0959
39260.02461.20931.11053.8%0.01810.59390.1032
40250.02561.23391.13354.0%0.01970.61950.1109
41240.02671.25951.15754.1%0.02160.64610.1190
42230.02781.28611.18264.3%0.02360.67400.1275
43220.02911.31401.20884.5%0.02590.70310.1364
44210.03051.34311.23624.7%0.02840.73350.1457
45200.03201.37351.26505.0%0.03120.76550.1555
46190.03371.40551.29535.2%0.03450.79920.1657
47180.03561.43921.32735.5%0.03820.83480.1764
48170.03761.47481.36125.8%0.04240.87240.1877
49160.04001.51241.39726.2%0.04730.91240.1995
50150.04271.55241.43566.6%0.05300.95510.2120
51140.04571.59511.47677.1%0.05981.00080.2250
52130.04921.64081.52107.6%0.06781.05000.2388
53120.05331.69001.56908.3%0.07741.10340.2533
54110.05821.74341.62149.0%0.08921.16160.2687
55100.06401.80161.679010.0%0.10391.22560.2849
5690.07111.86561.743011.1%0.12261.29670.3021
5780.08001.93671.815012.5%0.14691.37670.3205
5870.09142.01671.897314.2%0.17961.46810.3401
5960.10672.10811.993316.6%0.22551.57480.3613
6050.12802.21482.108520.0%0.29371.70280.3842
6140.16002.34282.252525.0%0.40311.86280.4094
6230.21332.50282.444533.3%0.60152.07610.4376
6321.3580 (duel)2.71613.666750.0%0.47533.43410.4525
6412.0371 (duel)4.07415.5001100.0%3.46305.47120.5066

What it says:

  • A single shot is worth taking in expectation from shot 7 on (price 0.0110 BNB, pool 0.7025 BNB; its EV is barely above zero and shows as 0.0000). The first six shots have a negative EV on their own.
  • Firing all 64 squares in turn costs 5.4712 BNB if the hit is the last square, and wins 5.5001 BNB. The plan's EV against a random hider is +0.5066 BNB, which is 9.25% of the worst-case spend.
  • The profit sits in the late shots, where the price is high but the chance of a hit is high too. Only a sniper who can afford the whole climb captures it, and the two duel shots (half the pool each) are the expensive part. That is the sense in which the curve favours deep pockets.

The burst: a sniper who fires every remaining square​

Because shots are concurrent, a sniper does not have to wait for answers. It can fire every unshot square back to back. The hider's square is among them, so the sniper always holds the winning shot: either the hider proves the hit, or the shot times out in the sniper's favour.

The total of all 64 prices does not depend on who fires them or in what order, because the price depends only on how many shots were fired before. So, without sponsorships, the final pool is fixed. For b = 0.01 BNB it is 6.1112 BNB, and the winner is paid 90% of it, 5.5001 BNB. A sniper who fires everything still unshot after n shots has spent final pool - pool(n), so

burst profit = 0.9 x final pool - (final pool - pool(n)) = pool(n) - 0.1 x final pool

and pool(n) is at least the collateral, which is only just more than 10% of the final pool (0.6400 BNB against 0.6111 BNB). The profit is positive at every stage, but from a fresh round it is thin:

Shots already fired and missedPool before the burstBurst costWinner is paidBurst profitReturn on the burst
00.64005.47125.50010.02890.5%
80.72475.38655.50010.11362.1%
160.82255.28875.50010.21134.0%
321.07875.03255.50010.46759.3%
481.51244.59885.50010.901319.6%
561.93674.17455.50011.325531.8%
602.34283.76845.50011.731646.0%

All amounts in BNB, base fee 0.01 BNB, computed with sniperPricingMatrix and sniperHitPayout. The hider loses the collateral of 0.6400 BNB in every row.

At the 10% rake the sweep from a fresh round clears only 0.0289 BNB (0.5%) on 5.4712 BNB laid out, so the gas of 64 transactions, which this page does not model, can turn that row negative. The margin widens quickly once part of the board has been fired: 9.3% after 32 shots and 46.0% after 60.

This is not a risk-free money machine, and the honest limits are:

  • Capital. The burst from a fresh round needs about 5.5 BNB for a round with 0.64 BNB of collateral: roughly 8.5 times the hider's stake, locked for up to an answer window.
  • Gas. 64 transactions, plus gas that this page does not model.
  • Competition. The pool is fixed, so two bots bursting the same round split the squares and only the owner of the hit square is paid; the other loses what it spent. The edge is shared in expectation, not doubled.
  • Older pending shots. A timeout on any square ends the round and voids the others, and an older shot expires first. A sniper that bursts into a round that already has an older pending shot from an address it does not trust can lose everything it fired. See what a hider can do with a second wallet.
  • House rounds are not exempt. The sweep needs no knowledge of the square, so it works the same on a round hosted by an approved address. On a fresh round it is worth about 0.045 x C (0.0289 BNB at C = 0.64) to the bot, before gas. See House rounds.
  • A reaction time. After the hider calls initiateExtraction, shots are accepted for 100 blocks, about 45 seconds. A bot that watches for ExtractionInitiated can run the burst inside that window, so an extraction does not stop a bot; it only stops people who are too slow.

2. The first-shot-timeout bot​

A bot fires one shot at every new round (price 0.0100 BNB, square of its choice) and never fires again. If the shot wins (a hit, or nothing answers it within the answer window) it collects 0.5850 BNB, a net 0.5750 BNB. If the hider's answerer is working it answers: a miss (63 times in 64) costs the 0.0100 BNB fee, a hit (1 in 64) pays the same pool. With q the fraction of rounds whose answerer is down for the whole answer window (default 267 blocks, about 2 minutes on BSC at about 0.45 s per block):

EV(q) = q * W + (1 - q) * M, with W = net when the shot wins = 0.5750 BNB and M = EV against a working answerer = -0.00086 BNB.

Answerer down for the whole window, qEV per round (BNB)EV per fee paid
0.00%-0.00086-8.59%
0.10%-0.00028-2.83%
0.50%0.0020220.19%
1.00%0.0049048.99%
2.00%0.01066106.57%
5.00%0.02793279.33%
10.00%0.05673567.26%
25.00%0.143111431.05%
50.00%0.287072870.70%
100.00%0.575005750.00%

The break-even is q* = 0.149%. Above that the bot is profitable before gas. In terms of the shot: a probe costs one base fee and a win pays 0.9 x 65 = 58.5 base fees, so the break-even chance of winning is 1 / 58.5, about 1.7%, of which 1.56% is the blind chance that the probe lands on the hider's square.

The numbers in this table did not change when the answer window moved from 3,600 to 267 blocks: they are scale-free in the base fee and do not depend on the window. What changed is what q means. Before, it was the share of hiders who happened to be away for 27 minutes. Now answers are automatic, so a hider being away does not matter if the zk-pad responder (or the House) is answering for them; q is the share of rounds in which the answering system itself is down for the 2-minute window.

Outages: what they cost and who can see them​

A responder outage of d minutes exposes a round for max(0, d - 2) minutes at the default window, because the first two minutes of an outage are covered by the answer window. The old 27-minute window hid short outages entirely; the 2-minute window does not. As an example, one 10-minute outage a month exposes 8 minutes, which is 0.018% of a 30-day month, below the 0.149% above. That comparison understates the risk: a bot does not fire at random. The responder's GET /v1/sniper/health endpoint is public, and so is the chain. A bot that sees ok: false, or a shot that has gone unanswered for 10 seconds, can fire one probe on every live round. If the outage lasts the whole window it wins every one of those pools. For any realistic failure (a crash loop, a provider down, a wallet out of gas) the chance that an outage that has already lasted 10 seconds lasts another 2 minutes is far above 1.7%.

The party who pays is the hider: the House in a House round, the hider in a community round. Snipers never lose money to a responder outage. For scale, the House's default of 3 live rounds at a base fee of 0.001 BNB puts about 3 x 0.065 = 0.2 BNB of collateral and fees at risk in a full outage.

The mitigations are operational, not contractual, and are listed under the responder: two redundant instances on different infrastructure with different keys and RPC providers, a backup that takes over in about 2 to 4 seconds, alerts that fire within 30 seconds, and senders funded for at least 200 answers. The contract has no circuit breaker: an owner pause() stops new shots, but it is entry-only, manual and the owner is a single key today, so automating it is a security trade-off and is not built. The web app disables "Shoot" on a round whose responder reports unhealthy; that protects honest users only, since a bot calls the contract directly.

This is why the table above still matters even though hiders no longer answer by hand: a bot can detect the one thing that makes it profitable.

3. The hider's side​

A hider who answers every shot and escapes after n fired shots (all misses) receives the pool minus a 10% rake on pool - collateral only, never on their own collateral. If a shot hits or times out first, the hider loses the whole collateral. An escape also needs the extraction countdown (default 100 blocks, about 45 seconds) to end with no shot pending.

Shots firedPoolEscape payoutHider netNet as % of collateral
00.64000.64000.00000.00%
80.72470.71630.076311.91%
160.82250.80420.164225.65%
321.07871.03480.394861.68%
481.51241.42520.7852122.68%
602.34282.17251.5325239.45%

All amounts in BNB. The probability that a uniformly placed square survives n misses is (64 - n) / 64; the table is the payoff if it does. A hider who is hit loses 0.6400 BNB (100% of the collateral).

Two costs the table leaves out. Every answer is a transaction of about 2.37 million gas, one for every miss. With the zk-pad responder, zk-pad pays that gas and meters it against the round's fees (see the hider guide); with a browser session key you pay it, about 0.0077 BNB for all 64 answers at 0.05 gwei (the app asks you to fund 1.5 times that), so at a small base fee the gas can be larger than what the hider wins. And a hider who is unlucky, whose answerer is down, or who is up against a bot loses the whole collateral, which is a far bigger number than the best case above.

What a hider can do​

ToolWhat it doesWhat it does not do
A low base feeCaps the absolute loss at 64 b and makes a burst cost less capitalDoes not change the sniper's percentage edge, which is the same at every base fee. Gas can swamp a tiny pool, on both sides
An early extractionThe pool is settled for the hider after about 45 seconds (plus any shot still pending) if nobody hits first. Against casual snipers this takes the profit in the table aboveDoes not stop a bot that watches for ExtractionInitiated and bursts. Shots fired in the last block of the countdown also hold the escape for up to an answer window
A working automatic answererRemoves the timeout loss and the one-shot timeout bot's edgeCosts gas for every answer, and does nothing about a hit. Two choices: the zk-pad responder, or a funded session key in an open browser tab
The zk-pad responderAnswers in seconds while the hider is away, within a cost policy checked when the round is admitted (at most 63 miss answers per round, and the base fee must cover the gas at a worst-case price when the round is admitted and created)Reveals the square to its operator, who can shoot it, and its outages are exploitable after about 2 minutes. See the hider guide
Picking an unpredictable squareAvoids squares that guessers try first (a corner, say)The tables above already assume a uniform square, so this only keeps you from doing worse than they show

What a hider can do with a second wallet​

The hider's own address cannot shoot its round, but a second wallet can, and the chain cannot tell whose shot is whose. This is a known limitation that the contract does not mitigate. It is an accepted property of a design without identity, not a bug waiting for a patch. Two variants matter. Both cost the hider the same constant: 0.1 x (C + F), where C is the collateral and F the fee of the second wallet's shot (plus gas). At C = 0.64 BNB and F = 0.01 BNB (the first shot) that is 0.065 BNB.

Variant 1: the self-hit exit. At any moment while the hider's own square h has not been shot, the second wallet fires at h. The hider answers it as a hit with a genuine proof. The second wallet is paid the pool less 10%, the round ends at once, and the hider's side has taken the pool without an extraction countdown and without any chance of a real sniper hitting h. The contract test test_knownLimit_hiderSelfHitExit runs it with a real proof: the second wallet receives pool - 10% and the treasury receives the rake.

Why the cost is 0.1 x (C + F): an honest escape pays the 10% rake on the fees only; the self-hit exit pays it on the whole pool, collateral included. The shot fee F is not lost, because it joins the pool and comes back at 90%. (The design review that found this first, at the earlier 5% rake, quoted 0.05 x C plus a whole fee, which counts the fee as spent. The same mistake at 10% would give 0.074 BNB. The test shows 0.065 BNB. Use 0.065.) The cost barely moves with how far the round has gone: a later shot pays a bigger F, but only 10% of the difference (0.068 BNB when the fee is 0.04 BNB).

It is worth it whenever the hider's chance of being hit, times the collateral it would lose, is more than the 0.065 BNB it costs: at C = 0.64 BNB, a hit probability above about 10%. This also means an extraction countdown is not a real risk to a hider who is willing to cheat: the self-hit exit skips the 100 blocks in which snipers could still fire.

Variant 2: the timeout void. The second wallet fires at a square the hider knows is empty, as early as it likes, and never answers. Every shot has the same answer window, so a shot fired earlier always expires earlier. A real sniper then fires the hider's actual square and the hider does not answer that either. When the second wallet's deadline passes, its owner calls claimTimeout: the round ends in the second wallet's favour, every other pending shot is void, including the real sniper's shot while it is still inside its own window, and the second wallet is paid 90% of the whole pool. The contract test test_knownLimit_hiderParkedShotTimesOutBeforeTheRealSnipersShot shows the earlier deadline being claimable first and the real sniper receiving nothing.

Both variants collect real snipers' fees. The pool contains every fee real snipers paid. If R is the total real snipers paid into the round after the second wallet's shot, the hider's side ends up 0.9 x R - 0.1 x (C + F) ahead of where it started, at C = 0.64 and F = 0.01 BNB:

Real snipers have paid in (R)Hider's side gainsReal snipers get
0.0310 BNB (after 4 shots in all)-0.0371 BNBnothing
0.0747 BNB (8 shots)+0.0023 BNBnothing
0.1725 BNB (16 shots)+0.0902 BNBnothing
0.4287 BNB (32 shots)+0.3208 BNBnothing
0.8624 BNB (48 shots)+0.7112 BNBnothing

The break-even is R of about 0.072 BNB, reached with the eighth shot in all (the parked shot and seven real ones at a 0.01 BNB base fee). That is the same 90% of the fees an honest escape would keep, for an extra 0.065 BNB, with no risk of a hit. Gas is not included. In variant 2 the hider also avoids losing the collateral if a real sniper fires h; in variant 1 the real snipers never get a chance to.

Other things a hider's second wallet can do:

  • It no longer blocks real snipers from firing. Shots are per square, so a parked shot holds only its own square, and the hider cannot escape while any shot is pending.
  • It can wash the price curve. Firing at known-empty squares and answering them costs the hider 10% of the faked fees plus gas and raises the visible pool. Do not read pool size or shot count as a sign of how many people play.

Why the contract cannot fix this. Any rule that tells the hider's shot from a real one needs to know who the hider is, and addresses are free. Paying the earliest expired shot, the latest one, only after every shot has expired, or splitting the pool among them can each be beaten by a hider who places shots before or after the real one (a split by flooding squares). A rule that charges the hider more (a bond) can make variant 2 expensive, but not variant 1, because a bond is refunded when the hider is hit. See House rounds for the deferred burned-bond design and what it would and would not buy.

Who is exposed. Snipers in community rounds, where the hider is whoever created the round. At launch the intended model is that only platform-approved addresses host rounds (House rounds). That changes who the hider is. It is not a proof: the contract cannot check that a House hider owns no shooter address. What a sniper can do in any round: treat a pending shot that is older than yours and not yours as a risk to your own shot, because its timeout comes first (getRound shows pendingBitmap, getShot shows each shot's sniper and deadline), and do not assume an old unanswered shot means an inactive hider.

The security page lists it among the known limitations, and the developer threat model (docs/THREAT_MODEL.md, section 3.6) records it as an accepted, unmitigated risk.

Accepted limitations at a glance​

These are accepted properties of the current design. Each is visible to you before you pay: the web app's "read before you pay" block says them, and so does this page.

LimitationWhat it costs whomCost to the cheating side
Sybil timeout order. The hider's earlier parked shot expires first and voids a real sniper's shotThe real sniper loses its fee and the hit0.1 x (C + F) plus gas: 0.065 BNB at C = 0.64 BNB, F = 0.01 BNB
Self-hit exit. The hider's second wallet shoots the hider's own square and the hider answers a hitReal snipers lose the chance to hit during the countdown, and their fees go to the hider's sideThe same 0.1 x (C + F) plus gas
Burst sweep. A bot fires every unshot square and wins by hit or timeoutThe hider loses the collateral; other snipers' squares are void or splitNone to the bot beyond capital: profit is pool(n) - 10% of the final pool, 0.5% to 46% of the capital laid out (thin from a fresh round, before gas)
First claimer picks the winner. When several shots have expired, the first claimTimeout caller chooses who is paidThe snipers who were not claimedNone. Claim in the first block you can
Answerer outage. If the hider's answerer is down for the whole answer window, a sniper can claim the pool by timeout, and a bot can watch for outagesThe hider (the House in House rounds) loses the collateral; snipers do not loseOne base fee per probe. Break-even at a 1.7% chance of winning, which an observed outage can exceed. See section 2
Pause during an extraction. An owner pause right after initiateExtraction rejects fireShot while the countdown keeps runningSnipers lose their last-chance shotsNone. This is an admin-trust assumption, see security

What this means for you​

  • As a sniper, the late shots and the burst carry the profit, a one-shot timeout bot beats a round whose answerer is down for a whole window, and your risks are capital, competition, gas, and older pending shots in the round you are firing into. In a community round the hider may be playing against you with a second wallet and can exit for about 10% of the collateral; in a House round you are relying on the platform's policy, which the chain cannot check.
  • As a hider, you are the side that has to keep an answerer running, pay for ZK answers (or accept that the responder knows your square), and accept that a patient or well-funded counterparty has the better expected return. Choose the base fee as a stake you are willing to lose.
  • As an observer, do not read the pool size or the number of shots as a sign of how many people are playing; a hider can inflate both.

Not fixed, by design​

  • Hider self-shooting through a second address is not preventable on chain. Both variants let a hider avoid losing to a hit and keep the real snipers' fees, at 0.1 x (C + F) plus gas. See above.
  • A sybil that parks a shot first wins the timeout order. The contract does not rank expired shots.
  • A burst sweep is profitable on any round it is run on (unless an older pending shot of an address the bot does not control exists), whoever hosts it, House rounds included.
  • The position commitment has no hider binding: a front-runner who sees a positionHash in the mempool can squat it, and the honest hider re-salts.
  • A round whose answerer is down for a whole window loses by timeout, and the shorter window (about 2 minutes) means an outage does not have to last long to be exploitable.
  • The price curve favours well-capitalised snipers.

Reproduce the numbers​

pnpm -C packages/sdk build
node packages/sdk/scripts/sniper-ev.mjs --stdout # tables 1 to 3 for a base fee of 0.01 BNB
node packages/sdk/scripts/sniper-ev.mjs --base-fee 0.02 --stdout # another base fee

The burst table uses sniperPricingMatrix(b): the final pool is rows[63].pool + rows[63].cost, the winner is paid sniperHitPayout(rows[63].pool, rows[63].cost), and the burst cost after n shots is the final pool minus rows[n].pool.