Two chains.
One price.
Held honestly.

01What Twinpad does

A Twinpad coin exists twice: as a pump.fun token on Solana and as a Pons token on Robinhood Chain. Each side has its own bonding curve, its own buyers and its own price. Left alone, the two prices drift apart within minutes. Twinpad's job is to stop that.

Every launch gets its own set of wallets. Twinpad's pool fronts each launch with working capital, the coin goes live on both chains, and a per-coin maker watches both prices every second. When the fully diluted value on one chain runs more than 5% above the other, the maker sells a small clip on the expensive side and buys a small clip on the cheap side. Inventory never crosses chains. Nothing is bridged.

The pool's capital is a loan, not a gift. Each side of a coin is launched with a fixed reserve, and that reserve is all the pool will ever put into that side. Creator fees from both venues are claimed once a minute while a coin is live: half is bought back and burned on the chain that paid it, the other half repays the pool. A coin that draws no outside holders is closed by policy and its capital returned.

02Where twin launches break

The first generation of twin launches on other platforms showed the same failures. We saw the early shape of each one in our own first coins and traced them in code before writing a line of the fix.

One-sided inventory

The Solana side opens with a large dev buy, so its wallet holds tokens and almost no cash. The Robinhood side opens with a little ETH and no tokens at all, because parity sizing lands the opening price below the Pons floor. From the first block the maker can only trade in one direction: sell Solana tokens, buy Robinhood tokens. When Robinhood runs ahead there is nothing to sell there, and the only lever left is buying the falling Solana side with fresh money.

Open-ended top-ups

When a maker runs out of cash on a side, the pool tops it up. If that top-up has no ceiling tied to what the coin has earned, a one-way sell-off on one chain turns into a pipe from the pool into a falling token. That is the mechanism behind the losses in the launches that went wrong: not a hack, not a bug, a rule that kept buying.

The operator is the biggest seller

To get its capital back, a maker sells its opening position into every rise. Done every tick, that makes the operator the largest seller of its own coin in the first hours, exactly when holders are deciding whether to stay. The pool gets its loan back faster and the coin pays for it.

A peg that cannot scale

On a bonding curve a few hundred dollars moves the price 10%. After graduation to a liquidity pool, the same 10% at a $1M market cap costs about $2,000 of buying, and the flow that opened the gap keeps coming. A maker sized for the curve stage becomes noise at scale, and a project that promises parity at any size is promising something it cannot pay for.

Nobody can see the reserves

Holders were asked to trust that the operator's wallets held what they claimed, on two chains, with fees flowing to addresses nobody could check. When the peg slipped there was no way to tell a temporary gap from an operator quietly exiting.

03The rule set

Seven rules. Each answers one failure above. The first four were chosen by running candidate rules through the simulator in Section 04 and keeping the one that held the band with the least pool money and the least operator selling.

3.1 Two-sided from the first block

Every launch opens a Robinhood position with its ETH front, even when parity math says zero. Both wallets hold tokens and cash before the first outside buyer arrives, so the maker can sell on whichever side runs ahead. Section 04 shows this single change is what moves the peg from "holds sometimes" to "holds almost always" during the curve stage.

3.2 One reserve per side, and the pool never adds to it

Each side of a coin is launched with its opening position plus a cash reserve of twelve clips. That reserve is the whole peg budget for that side. The pool never tops it up. Sells on a side refill its cash and fee income adds to it; nothing else does. The maker buys the cheap side only while the reserve has cash, so the reserve is the cap and there is no second rule to reason about.

The worst case the pool can lose on a coin is the amount it fronted, and that number is on the coin page before the first outside buyer.

3.3 The maker is not the top seller

Front repayment sells at most one clip a minute per side, and only once that side trades 25% above the maker's entry. Outside that, the maker sells only to close a gap, and every such sell is matched by a buy on the other chain. Repayment takes longer this way. Because the reserve is fixed, the pool's exposure does not grow while it waits.

3.4 Half of every fee is bought back and burned

Both venues pay the coin's creator wallet on every trade: Pons a 2% creator tax during the curve stage, pump.fun 0.30% of each curve trade and a market-cap-tiered share after graduation. Twinpad claims both once a minute. Half of each claim buys the coin on the venue that paid it and burns the tokens, timed so the buy never lands on the side that is already expensive. The other half repays the front and then flows to the pool.

On Solana the burn is a token-program burn, which reduces supply on chain. On Robinhood Chain it is the token's own burn where the contract has one, otherwise a transfer to the dead address, which no key controls. Either way the transaction is linked from the coin page.

The burn is a commitment to holders, not a peg tool, and the sizing is stated plainly: per $100,000 of curve-stage volume it burns about $1,000 on Robinhood Chain and $150 on Solana. After graduation Pons pays no creator fee and pump.fun's share tiers down toward 0.05%, so the burn is largest exactly when the coin is young. Over the three simulated launch hours it moved the peg by nothing and cost the pool the burned amount.

3.5 The maker has modes

• Curve stage: parity keeper. Clips sized to curve depth, band 5%, both sides selling and buying inside the reserve.
• Float stage, above a market cap stated on the coin page: budgeted band-keeper. Clips scale with venue depth, the reserve stays fixed, and the maker never adds pool money. Gap-closing sells are capped per hour so that the maker cannot liquidate the pool's position into a falling market in the name of parity. Separately, the pool sells its inventory in tranches into strength on both chains, taking profit on the way up instead of defending a level on the way down.

3.6 Reserve rebalancing is manual and announced

Inventory never crosses chains automatically. When the pool's SOL and ETH drift out of proportion the system raises an alert and an operator swaps, and the swap is logged on the status page. No maker ever holds a bridge key.

3.7 Proof of reserves on every coin page

Each coin page shows the coin's wallets on both chains, live balances, what was fronted, what has been repaid, fees claimed, tokens burned per chain, and the maker's remaining reserve. Anyone can check any number against the explorers. Section 06 lists the fields.

What Twinpad does not do

• Raise the Pons creator tax to fund the peg. Holders pay 2% on that side and that is where it stays.
• Buy its own coin up to a target market cap. Market cap comes from outsiders or it does not come.
• Promise parity at any size. Above the stated market cap the two sides float, and the page says so.

04Simulation

Before committing to the rules in Section 03 we built a simulator against the real curve mathematics and the maker's real parameters, and ran candidate rules through the order flows that broke earlier launches. The code ships in the repository and every table here can be regenerated with one command (Appendix).

Method

• Venues. pump.fun bonding curve at its real virtual reserves (30 SOL against 1.073 billion tokens) with its 1.25% fee, of which 0.30% reaches the creator. Pons launch configuration 0 as read from chain: 1 billion supply, 1.68 ETH phantom quote, 1% fee, 2% creator tax, graduation at 4.2 ETH raised. After graduation both venues are modelled as constant-product pools seeded from the curve's final state; the pump.fun creator share is taken at its 0.05% floor tier there and Pons pays none.
• Rules compared. Live: the rule in production before this paper, which buys the cheap side whenever a gap is open and tops up from the pool inside a per-coin cap. Ceiling + seeded: the same with an opening Robinhood position and a cap on tokens the peg has bought per side. Steward: rules 3.1 to 3.3, a twelve-clip reserve per side, no top-ups, repayment sells one clip a minute at 25% over entry. Steward, scaled: the same with clips sized to 0.5% of venue depth. Steward + burn: steward plus rule 3.4.
• Flow. Seven order-flow regimes at two intensities: Solana holders dumping while Robinhood buys and its mirror, each side dumping alone, both dumping, both being bought, and two-way noise. A "30% dump" sells 30% of outsiders' holdings over three hours; a "30% buy" spends 30% of the venue's starting depth.
• Measure. Each run lasts three hours at a 3-second tick (production now runs the maker every second, which only tightens the results), repeated over 20 random seeds. We report the share of ticks inside the band, what the maker sold, what the pool topped up, and the pool's outcome against a baseline where the maker is switched off under the same flow.

Result 1: which rule, same capital

A fresh launch with today's capital, about $940 fronted per coin, plus the steward reserve where the rule calls for it. Each cell is the share of time inside the 5% band and the dollars the maker sold over three hours.

Pool top-ups per coin: live up to $507 and about $300 in every two-way flow, ceiling + seeded $103–305, steward $0. The $150 loss guard halted the live rule in three of the flows (red) and steward in none. The live rule's five-figure selling in two-way markets is churn: with nothing to sell on the cheap chain it trades every tick without closing the gap, paying fees each time, until the guard stops it.

Result 2: inventory on both chains

Today's launch shape under the live rule against a launch that opens tokens and cash on both chains with a disclosed locked allocation, run under the steward rule. The two-sided shape fronts $3,385 of opening positions plus $600 of reserves per coin, excluding the locked allocation. This is the launch shape Twinpad adopts.

With the two-sided shape every rule holds the band about equally well; the difference is what it costs. In two-way noise at 30% the live rule sold $1,049 of the coin over three hours and steward sold $211, with the band held 100% by both. The loss guard never fired under the two-sided shape.

Result 3: what market cap does to the peg

The same maker placed at six market caps. Depth is the quote sitting in the venue. The cost column is what it takes to close a 10% gap on the shallower side. "One-way" is a 30% Solana dump with Robinhood quiet; "two-way" is 30% noise. Live uses $25 clips and pool top-ups; steward scaled uses clips of 0.5% of depth and a reserve of twelve such clips per side, never topped up.

Two things follow. In two-way markets a peg is affordable at any size if clips grow with depth, and a fixed reserve does it without the pool sending more money later, where the live rule needed thousands in top-ups. In a one-way run above roughly $25k no rule we tested holds the band, and the money spent trying is money lost. That is the boundary between the curve-stage parity keeper and the float-stage band-keeper in Section 3.5.

Depth-scaled clips trade more. As simulated, the scaled steward closes gaps in falling markets too: at $1M in a two-way sell-off it sold $114k of the pool's position where the live rule sold $12k. In a two-way rally it sold $6k, mostly repaying its reserve, and those sells left the end price on both venues about 9% below the hold baseline, which marked the pool's whole position down by $35k. The float stage in Section 3.5 caps gap-closing sells per hour so that parity work never becomes liquidation. That cap is not in these runs.

Result 4: the bag, not the maker, is the exposure at scale

At scale the pool's risk is the size of the position it still holds. Keeping parity in a one-way dump adds 11% to 25% to that loss, for every rule: matching the falling chain means selling the quiet chain down to meet it, and the pool's position there is marked down with it. The same arithmetic runs in rallies. While the pool holds about a fifth of the supply, each dollar the maker sells at a $1M market cap lowers the mark on the whole position by several dollars, so the cash is banked and the price holders see is lower. Both facts point the same way. The float stage sells in tranches into strength rather than defending a level, it does not chase a one-way run on the other chain, and the locked allocation stays locked.

Result 5: the burn is real money, not a peg

Adding rule 3.4 to steward burned $27 to $110 per coin over the three launch hours in the two-sided shape and moved the share of time in band by at most one point in every flow. Above graduation, at pump.fun's floor tier and with no Pons fee, it burned $1 to $37 per three hours with today's small opening position. The burn's size follows volume, not the peg, which is why Section 3.4 states it per $100,000 traded rather than as a promise about price.

How to read the numbers

• Order flow is synthetic. It was shaped to reproduce the failure modes, not fitted to any one token's trades. Rankings between rules are the finding; dollar amounts are illustrative.
• Runs are marked to market at the end of a three-hour window. In a rising market a maker that sold into strength shows less than one that held, even though it banked real cash. That is an opportunity cost, not a loss from the pool.
• Gas, slippage rejections and RPC failures are excluded. They affect every rule equally.

05Safety rails

The values the rule set specifies. Where a rail is not yet live the roadmap in Section 07 says so, and the coin page shows the values in force for each coin.

06Transparency

Every coin page carries a reserves panel. Every field is an on-chain number or a signed record that resolves to one.

• Creator, maker and launcher wallet addresses on both chains, linked to Solscan and to the Robinhood Chain Blockscout explorer.
• Live token and quote balances per wallet, and the locked allocation's address and size on each chain.
• Fronted, repaid and retired amounts per chain, with the transactions behind each step.
• Creator fees claimed per venue, cumulative and last claim, and tokens burned per chain with the burn transactions.
• The maker's reserve on each side and how much of it remains.
• Maker state: parity keeper or band-keeper, the market cap where the mode changes, and any halt with its reason.
• Pool status: SOL and ETH held, free above the floors, and the log of operator rebalancing swaps.

Closes are logged step by step: sell, fee collection, sweep, refund. A closed coin's record stays public.

07Roadmap

08Limits and risks

• Parity is a service, not a law. Above the stated market cap, or during a one-way run, the two prices can and will separate. The page says so before anyone buys.
• Custody. Twinpad holds the per-coin wallets and the pool. Holders are trusting an operator, and the reserves panel exists so that trust can be checked rather than assumed.
• Burns are per chain and irreversible. Fees differ between venues, so the amounts burned on each chain differ too. Circulating supply diverges slightly between chains over time; parity is measured on nominal supply and the panel shows both burn totals.
• Venue risk. pump.fun and Pons are third-party programs. A change in their fees, graduation rules or creator share changes the economics here, including the burn.
• Infrastructure. Robinhood Chain RPC availability has already required failover once. A maker that cannot read a price does nothing, which is the safe failure but still a gap.
• Model risk. The simulator's order flow is synthetic and post-graduation pools are approximated. It ranks rules well; it does not forecast dollars.

09Appendix

Simulation assumptions

• Prices used throughout: SOL $102.75, ETH $2,532.60.
• pump.fun graduation at 85 SOL raised; pool seeded with 79 SOL and 206.9M tokens at 1.25% fee. Pons pool seeded with the 4.2 ETH swept at the curve's final price, pool fee 0 per launch configuration 0, no creator tax after graduation. Both treated as constant product, scaled to a target market cap with quote depth proportional to the square root of price.
• Creator income: Pons 2% tax on curve trades; pump.fun 0.30% of curve trades and 0.05% of pool trades, the floor of its market-cap tiers, so the post-graduation burn is a lower bound.
• Above launch every rule starts from the same opening positions, so tables compare behaviour rather than bag size.

Reproduce

cd server
SOLD=1 PUMP_CREATOR_BPS=30 npx tsx scripts/sim-peg.ts 0        # Result 1, 5
SOLD=1 PUMP_CREATOR_BPS=30 LOCK_PCT=0.15 FRONT_SOL=19.63 DEV_SOL=19.46 \
  FRONT_ETH=0.54 SEED_ETH=0.5287 npx tsx scripts/sim-peg.ts 0   # Result 2
SOLD=1 PUMP_CREATOR_BPS=5 npx tsx scripts/sim-peg.ts 1000000   # Result 3, 4: one cap;
                                # repeat for 25000 100000 500000 5000000

Glossary

• FDV — fully diluted value: price × 1 billion supply, in USD. The band is measured on FDV so both chains compare on the same footing.
• Front — the pool's loan to a coin's wallets at launch: the opening positions plus each side's reserve. Repaid from repayment sells and the pool's half of fees.
• Clip — one maker trade.
• Reserve — the cash a side holds for buying the cheap chain. Fixed at launch, refilled only by that side's own sells and fees.