Run a prediction market that actually makes money.
把预测市场做成真正赚钱的生意。
A practical operator's field guide to AMM prediction markets: what to list, how to seed, when to pull liquidity, and the one place an AMM is a genuine profit center. Read the playbook — then model your own market.
One rule decides everything: the fee earns from retail round-trips and nothing else. So you run the book on the heads, the AMM as a cold-start engine for the tail, and a real profit center in exactly one niche. Catalog, capital, risk and the business — four chapters.
Model operator P&L for any market: seed, window, terminal price, turnover and fee. See fees, convergence loss, net result and breakeven turnover update live.
Count is sports, money is politics. A barbell catalog of ~120 markets, demand mechanics that manufacture the round trip, and hard kill criteria at days 30 / 60 / 90.
Routing by contestedness, seeding to expected volume, withdrawal that stops future loss, and the single most-resourced function: opening price.
按胶着程度路由、按预期量铺仓、撤池只为阻止未来亏损,以及投入最重的职能:开盘定价。
Chapter 25 min
03
RiskAdvanced
Risk Rules
风控铁律
Six rules ranked by what actually kills operators — adverse selection, correlation, resolution, bankroll, ops and abuse. Diversification counts clusters, not markets.
Revenue lines ranked, the KPIs that gate every listing, and the three year-one calls: book-first, fee structure, and which categories to actually back.
收入线排序、决定每次上架的 KPI,以及第一年的三个决策:订单簿优先、费率结构、到底押哪些类别。
Chapter 44 min
★
StrategyStart here
The Operator's Doctrine
运营总纲
The one-paragraph thesis the whole playbook hangs on: who you actually earn from, who you must refuse, and where the only real AMM profit center lives.
整本手册的一段式核心:你到底从谁身上赚钱、必须拒绝谁,以及 AMM 唯一真正的利润中心在哪。
Overview2 min
StrategyCore · 18 min read
The Operator's Doctrine
运营总纲
Where an AMM makes money, where it loses it, and the operating rules that keep you on the right side.
AMM 在哪里赚钱、在哪里亏钱,以及让你站对一边的运营规则。
PPrometheX Research · Updated Jun 2026
The fee monetizes retail round-trip flow and nothing else — informed flow is fee-immune by construction and causes all the inventory loss. So: order book first for the head markets (flow capture, zero inventory risk), the AMM as the cold-start machine for the long tail (seed = paid marketing with a measurable cost), and a true AMM profit center in exactly one niche: high-churn contested sports and news.
"List what noise traders love, refuse what insiders love, cap every correlated cluster, keep half the bankroll out of the pools."
01Catalog & Launch
Count is sports, money is politics, P&L is decided by ~10 head events. The catalog is a barbell of roughly 120 markets — a thin head that carries the P&L, and a long tail that exists to fill the board and cold-start liquidity.
Demand mechanics — manufacture the round trip. Live-sports cadence, a prominent cash-out button (turns one-way bets into fee-bearing round trips), weekly P&L-percentage leaderboards, sign-up credit withdrawable only after 5× wagering, and community market requests gated on pre-committed demand.
Kill criteria
Day 30 — Tier A sports ≥ 8× turnover per weekly cycle; tail markets < 1× over two cycles get delisted. Day 60 — delist any category whose flow is > 70% one-directional. Day 90 — sports must be profitable in both outcomes at 150–200 bps, otherwise the audience isn't there and more capital only scales the loss.
02Capital & Market-Making
"Opening price is the most-resourced function. A 0.30 open resolving toward 0.85 burns 53.5% of seed."
Routing. AMM only for contested retail markets (p̄ ≤ 0.8) and the long tail; book-only for anything decisive (p̄ ≥ 0.9) or insider-prone; both on heads — a thin AMM for always-on UX with a quoting bot inside its spread.
Seeding. Seed ≈ 0.5–1× expected daily volume, floor 10–20× typical trade size. Add only after ≥ 1×/day turnover for 3 days, never past 1× daily volume.
Withdrawal pays only to stop future convergence (the accrued loss is sunk). Automate a pull trigger at forecast-vs-price gap > 15 points: on a $10k pool at 0.70 with news pointing to 0.95, pulling saves ~$4,800 of future loss against ~$45 of forgone fees.
Opening price. Anchor to de-vigged external odds (auto-block if |p₀ − reference| > 5 pts); no anchor → 24–48h book-only price discovery; always a 10–20% probe tranche first; news scan within 15 minutes of seeding.
Dead time. Hurdle math uses duration + oracle window (2–5 d ≈ 30% ROI haircut on a 7-day market). Stagger seeding so redemptions land daily; keep a 15–20% liquid buffer.
03Risk Rules
Ranked by what actually kills operators, not by how often the textbook mentions them.
Adverse selection. Listing screen: outcome generated by a process, not a < 20-person decision; no one plausibly knows > 24h early; a public updating signal exists; p̄ ≤ 0.9 at 2%; projected turnover ≥ 2× breakeven. Any "no" → book-only or reject. Personnel decisions, single-company announcements, niche esports: zero AMM seed, ever.
Correlation. Same league + week / same election / same asset = one cluster. Max 10% of bankroll per cluster, 25% per category; diversification counts clusters, not markets.
Resolution. Mandatory wording template (source + timestamp + tiebreak + postponement); oracle bond + gas < 5% of seed; dispute reserve = 2× bonds outstanding; propose resolution ourselves at T+0.
Bankroll. Max 50% deployed; per-market cap 2% of bankroll; sized to survive three consecutive zero-upset months (≈ −22%).
Ops. Fee asserted ∈ {100, 200, 300} bps at deploy; read back on-chain config before announcing; never hike fees mid-life (53:1 self-harm ratio).
Abuse. Rewards ≤ 50% of net fees paid per address per epoch; exclude incentive volume from the turnover stats used for seeding — wash flow looks exactly like profitable noise.
04The Business Model
"Points and token value aren't revenue — they're deferred dilution buying turnover."
Revenue lines, ranked. ① Book taker fees (50–100 bps, zero inventory risk — the scalable line) ② AMM LP margin (real only in the contested high-churn band) ③ withdrawal skim ④ B2B sublicensing once the book prints ⑤ sponsorships ⑥ points / token value — not revenue.
KPIs. List only at projected turnover ≥ 1.5× breakeven; fee capture ≥ 0.6·f·T̂; annualized ROI ≥ 30% incl. settlement gap; seed / daily-volume ≤ 1×; category loss ratio < 0.6.
Year-one calls. Order-book-first with the AMM as feeder; 2% AMM flat at creation / sub-1% book taker; sports for count + politics for money, skip crypto price targets (breakeven mean, undiversifiable variance).
Pick a scenario, then move the sliders. Everything from the playbook — fees, convergence loss, breakeven turnover — recalculates live.
选一个场景,然后拖动滑块。手册里讲的一切——手续费、收敛损失、保本换手率——实时重算。
Annualized ROI on locked capital
—
—
Parameters
Advanced
Results
Net profit / loss
Fees earned
Loss at settlement
Breakeven daily turnover
Cumulative P&L over the trading window
Cumulative feesAccrued convergence lossNet P&L
Net P&L vs terminal price
Net P&L at resolutionFee revenueResolution loss
Reference
Glossary
术语表
Every term the playbook and calculator lean on, in one place.
手册与计算器用到的全部术语,集中在这里。
A
AMM (Automated Market Maker)AMM(自动做市商)
A pricing contract that quotes both sides of a market from a liquidity pool, so trades execute instantly against the pool instead of waiting for a counterparty.用流动性池对市场双边自动报价的合约,交易直接与池子成交,无需等待对手方。
Adverse selection逆向选择
Trading against counterparties who know more than the pool. In insider-prone markets, informed flow steadily extracts pool value.与比池子知道更多的人交易。在内幕型市场中,知情资金会持续抽走池子的价值。
Annualized ROI年化回报
Net profit scaled to a yearly rate on locked capital, including the oracle settlement window.按锁定资金折算成年化的净收益,计算时包含预言机结算窗口。
B
Bankroll总资金
The operator's total capital across all markets. At most 50% is ever deployed into pools at once.运营者在所有市场上的全部可用资金。同时入池的部分永远不超过 50%。
Barbell catalog杠铃目录
A few heavily-seeded head markets plus a long tail of cheap board filler — and very little in between.少数重仓的头部盘,加上大量低成本的长尾充实盘——中间地带很少。
Basis points (bps)基点(bps)
One hundredth of a percent — the unit fees are quoted in. 100 bps = 1%.百分之一的百分之一,即 0.01%;费率的计价单位,100 bps = 1%。
Breakeven turnover保本换手率
The daily noise volume, as a share of pool size, at which fees exactly cover convergence loss plus ops cost.手续费恰好覆盖收敛损失与运营成本时的日换手率(占池子规模的比例)。
C
Cash-out提前结算
Closing a position before resolution. Turns one-way bets into fee-paying round trips — a core demand mechanic.在结算前平仓。把单程下注变成缴两次费的往返交易——核心的需求侧设计。
Cluster (correlation)相关性 cluster
Markets whose outcomes hang on the same underlying event — same league week, same election, same asset. Capped at 10% of bankroll each.结果由同一底层事件决定的一组市场——同联赛同周、同一选举、同一标的。单个 cluster 上限为总资金的 10%。
Convergence loss收敛损失
The pool's loss as price moves from the opening price toward the true outcome — paid out to informed traders along the way.价格从开盘价向真实结果收敛的过程中池子产生的亏损——沿途付给了知情交易者。
D
Dead time死区时间
The days between market close and oracle payout, when capital is locked but earning nothing. 2–5 days cuts a 7-day market's ROI by ~30%.市场关闭到预言机赔付之间的天数——资金被锁定但零收益。2–5 天能让 7 天盘的 ROI 打七折。
De-vigged odds去水赔率
External bookmaker odds with the margin stripped out — the reference anchor for every opening price.去掉庄家抽水后的外部赔率——所有开盘定价的参考锚。
F
Fee capture费用捕获
Realized fees as a share of the theoretical fee × turnover. Listing KPI: at least 0.6.实际收到的手续费占理论值(费率 × 换手)的比例。上架 KPI 要求不低于 0.6。
Fee switch费率切换
A scheduled change to the trading fee partway through a market's life. Buy-side fees convert at (1 − price), so a late hike mostly lands on sellers.在市场生命周期中按预定时点切换费率。买入侧手续费按 (1 − 价格) 折算,后期提费主要落在卖出端。
H
Head market头部盘
The 8–10 markets that decide the whole P&L — marquee sports, contested politics, crypto monthlies.决定整体损益的 8–10 个市场——焦点赛事、胶着政治盘、加密月度盘。
Hurdle rate门槛收益率
The minimum acceptable APR on locked capital. A market can be profitable and still fail the hurdle.锁定资金可接受的最低年化收益率。一个市场可以盈利但仍达不到门槛。
I
Informed flow知情流
Trades made on real information. Fee-immune by construction, and the source of all inventory loss.基于真实信息的交易。按机制几乎不缴费,且是全部库存亏损的来源。
K
Kill criteria关停标准
Pre-committed delisting rules checked at days 30, 60 and 90 — turnover floors and one-directional-flow caps.在第 30、60、90 天检查的预承诺下架规则——换手率下限与单向流占比上限。
L
Long tail长尾
The 70–80 filler markets that exist to fill the board. Median lifetime volume of a tail sports event: about $7.4k.用来充实目录的 70–80 个市场。体育长尾盘的中位生命周期交易量仅约 $7.4k。
N
Noise flow噪声流
Retail round-trip trading driven by entertainment rather than information — the only flow the fee actually monetizes.出于娱乐而非信息的散户往返交易——手续费唯一真正能赚到的流。
O
Opening price (p₀)开盘价(p₀)
The probability the pool quotes at seeding. The most-resourced function in the operation: a 0.30 open resolving toward 0.85 burns 53.5% of seed.铺仓时池子报出的初始概率。整个运营中投入最重的职能:0.30 开盘、0.85 收敛会烧掉 53.5% 的种子。
Oracle预言机
The mechanism that reports the real-world outcome on-chain and triggers settlement.把真实世界结果上报到链上并触发结算的机制。
Order book订单簿
A matching engine where traders quote each other directly. Zero inventory risk for the operator — the scalable revenue line.交易者直接互相报价撮合的引擎。运营方零库存风险——可规模化的收入线。
P
Probe tranche探针仓
The first 10–20% of seed deployed to test a new market before committing full size.全量铺仓前先放入的 10–20% 测试仓位。
R
Resolution结算
The final determination of the outcome and redemption of winning shares. Wording templates and dispute reserves live here.对结果的最终判定与中奖份额的兑付。措辞模板与争议准备金都属于这一环节。
Round trip往返交易
A buy plus a later sell — pays the fee twice. The unit of monetizable volume.一买一卖的完整来回——缴两次手续费。可变现交易量的基本单位。
S
Seed liquidity种子流动性
Capital deposited to open the pool, sized at 0.5–1× expected daily volume — accounted for as paid marketing with a measurable cost.开池注入的资金,规模为预期日交易量的 0.5–1 倍——按"有可计量成本的付费获客"记账。
T
Terminal price (p̄)终价(p̄)
Where the price ends up just before resolution — how certain the outcome becomes. The single biggest driver of convergence loss.结算前价格的最终位置——结果变得多确定。收敛损失最大的单一驱动因素。
Turnover换手率
Daily volume divided by pool size. The variable that decides whether fees beat losses.日交易量除以池子规模。决定手续费能否跑赢亏损的核心变量。
W
Wash trading刷量
Self-dealing volume farmed for incentives. Statistically identical to profitable noise — which is why incentive volume never counts toward seeding stats.为薅激励而自买自卖的交易量。在数据上与盈利噪声完全相同——所以激励量永远不计入铺仓统计。
Withdrawal trigger撤池触发
An automated liquidity pull when the forecast-vs-price gap exceeds 15 points. It only stops future convergence — the accrued loss is sunk.预测终价与现价差超过 15 点时自动撤池。它只能阻止未来的收敛——已发生的亏损是沉没成本。