Clicking “Buy YES” on an AMM-powered prediction market triggers a pipeline: pool state updates, price moves, shares minted or transferred, fees collected. This chapter walks that pipeline—what happens in the contract, what you see in the UI, and how that differs from a limit-order fill on a regulated book.
Step-by-step trade flow
First the system reads pool state: reserves, outcome shares outstanding (q), parameter b, fee tier. The frontend quotes by simulating cost for desired shares or stablecoin spend. You execute—wallet signature or server confirmation. The curve updates; new q shifts implied probabilities for all outcomes. Settlement later pays $1 on the winning outcome via oracle rules separate from intrade math.
Pricing and resolution are decoupled. Oracle mistakes do not change how the curve repriced before truth arrived—but vague resolution still makes the whole market untrustworthy, as in the opening module.
Where liquidity comes from
Initial subsidy seeds the pool with cash and b. LP deposits add stablecoin; the pool honors buys and sells. Fee reinvestment may leave a portion of trading fees in the pool. Arbitrage supplies economic liquidity when bots trade dislocations versus books or models.
Without subsidy, LPs, and arb, curves are shallow and percentages lie.
Operators sometimes seed multiple pools per event (YES/NO tokens vs collateral pool). The user-facing pipeline is still read state → quote → sign → update curve; only the accounting layer changes. Traders should follow their venue’s settlement path, not a generic DeFi tutorial.
Oracle and timing risk
Between trade and resolution, oracle committees, challenge windows, or exchange rulebooks can delay payout. Intrade pricing ignores that timeline, but your capital is locked through it. Wide oracle risk often shows up as wider effective spreads as LPs demand compensation—even when the formula mid looks calm. Headline odds during a quiet week may look credible because arb links them to regulated books; an isolated niche pool with no twin can show the same UI pattern with no national forecast behind it.
Price vs probability display
The UI shows 63% YES. Underneath, LMSR uses softmax over q with b; constant-product designs use token balance ratios. Both map to implied probability for binary events when outcomes partition 100%.
Display lag matters on-chain: prices can change between quote and fill during mempool delay. Regulated apps often lock a server quote until confirm—different UX, same lesson: simulate immediately before signing.
Worked example: quote vs fill on a thin pool
Assume binary market, b = 50, pool TVL roughly $18k, display 52% YES. At T0 the simulator says $500 buys average 54.1%—save that output. At T1 you approve while pending; mid drifts to 53%. At T2 fill confirms at 54.4% effective. At T3 another trader buys $300 YES and your mark moves to 55% without you trading.
Your edge is not “I bought below 54%” unless you locked it. Gaps between quote and fill are slippage tax for takers and profit opportunity for bots—not rounding error.
Minting and burning shares
Many designs mint outcome tokens when you buy from the pool and burn when you sell back. Net supply of YES/NO reflects positioning. At resolution, winners redeem $1 and losers go to zero. LPs holding inventory bear outcome risk if they do not hedge—unlike many CLOB traders who flatten into the book, LPs often stay in the pool until explicit withdrawal.
Fees
Typical stack: protocol fee (percent of notional to treasury or LPs), implicit slippage (curve movement), and gas on-chain. Compare all-in cost to crossing a two-cent spread on a book before declaring the AMM cheaper. A book might show YES 62¢ ask with $8k at top; your $2k lift pays 62–63¢ all-in. A pool might show 61% mid but 63.5% effective on $2k after curve, fee, and gas. The cheaper headline is not always the cheaper trade.
Liquidity depth without an order book
Depth means how much notional moves price one percent. If $100 moves 50% → 50.5% but $10,000 moves 50% → 58%, the pool is shallow for your size—you are the market. Simulate 1% and 5% impact mentally before betting edge on the mid.
Informed flow vs noise
When news hits, informed traders buy one side and the curve ratchets—feature, not bug. LPs lose if they are on the wrong side; they earn fees if flow is balanced noise. Adverse selection appears as inventory bleed rather than picked-off limits. The pool learns from your order; informed flow is tuition.
Stale quotes and manipulation
Thin pools allow pumps with small capital and wash volume on weak surveillance. Mitigations include minimum pool size to list, oracle delays, volume-weighted displays, and hybrid CLOB overlays. On a thin AMM, moving the headline to 90% may cost hundreds; on a deep book plus overlay, touch may resist while the pool still moves on huge size.
Regulated book vs pool-first pipe
On a CLOB-first venue, matched limits at the touch move price; depth is size at best bid and ask. On pool-first or hybrid venues, curve plus arb plus optional book sets the print; depth is TVL, b, and simulator output. Election night on a mature regulated book might show one- to two-cent spreads with large size at top; a new geopolitical prop with $12k TVL is still pure curve risk—do not import marquee depth assumptions into every screen.
Splitting large orders
Because the curve is convex in trade size, splitting a large buy into smaller clips can cost more total slippage than one block—path dependence again. Conversely, splitting can reduce temporary headline impact if you care about signaling. For pure P&L, trust the simulator for the full size you intend to hold, not for a probe trade at one-tenth size.
Tie-in to informative prices
Module one framed markets as priced beliefs under clear resolution. AMMs change how fast beliefs update and who supplies the quote. 70% with $15k TVL is not the same information product as 70% with $2m book depth, even if resolution text matches. Cross-venue chapters treat who trades and what manipulation costs; every pool trade feeds the curve bots read.
Wallet and approval friction
On-chain pools add approval transactions, RPC failures, and failed simulations. A quote that looked good in the UI can revert on-chain when pool state moved in the same block. That is not LMSR-specific—it is reason to treat confirmed transaction logs as the source of truth for entry price, not the preview modal.
Withdrawals and LP exits
LPs who withdraw shrink TVL and steepen the curve for everyone still trading. Watch for governance votes or reward program endings that coincide with TVL cliffs—your slippage can jump without any news on the event itself.
Backend versus on-chain settlement
Some products show AMM-style percents while settling on a centralized ledger; others settle entirely on-chain. The trade pipeline in this chapter still applies—read state, quote, execute, update curve—even when the wallet step is abstracted away. Know where your shares live because that determines withdrawal and dispute paths.
Healthy signals include TVL large versus your trade, active 24h volume, small price impact on a $1k probe, and post-news gaps versus books closing within minutes. Red flags: TVL below your order, one-sided wash volume, >2% move on $1k, persistent five-cent gaps, or effective price more than three points from mid on your size.
After your trade confirms, the curve state is new—any screenshot of the pre-trade percent is already history. Refresh the simulator before adding size.
Summary sentence
Every click is a state transition on the curve—trade with the post-click world in mind, not the banner you saw three seconds ago.
Key ideas
AMM liquidity is pool capital plus formula, not posted bids. Every trade moves the curve—simulate size first. Fees are explicit percent plus slippage plus gas; compare all-in to books. LPs and operators subsidize depth; arbitrage aligns external truth when it exists.
Next: the b parameter—liquidity, sensitivity, and slippage dial.