算法赌场 · AI 博彩产业全景 · 批判性行业分析 · 2026-07The Algorithmic Casino · AI in Gambling · A Critical Map · Jul 2026

庄家的 AI 从不预测比赛,它预测的是你——赔率师这个工种消失了,抽水活得比任何时候都好 The house's AI never predicts the game — it predicts you. The odds compiler is gone as a trade, and the vig has never been more alive

全行业最诚实的一条规则写在风控里:「限注赢家」。一家真在对赌的公司不会驱逐赢家——驱逐这个动作本身,就是「公平博弈」谎言的完整供词。The industry's most honest rule lives in risk control: 'limit the winners.' A house genuinely betting against you would never expel the winners — the expulsion itself is the full confession that 'fair play' was a lie.

当两边都武装到牙齿,AI 对 AI,信息差归零、预测优势归零;牌桌上唯一不归零的,是庄家事先写进赔率里的抽水。When both sides are armed to the teeth, AI against AI, the information gap goes to zero and every edge goes to zero; the only thing on the table that never goes to zero is the vig the house wrote into the odds beforehand.

主脊是「一注」(A Bet)的生命周期七节点(定价→获客→产品→风控→结算→留存→合规),每节点传统赔率师 vs AI 算法工厂双轨对照。两个节点被单独打光:④风控＝杀赢家、⑥留存＝最黑暗的 AI 应用。另有诚实层、抽水率阶梯、龙头竞技场、AI 应用四色地图、防御工具箱与边界扩张。本页是批判性行业结构分析——揭露而非教学,不含任何投注建议。金融博彩化(0DTE / 预测市场)作「边界扩张」论据,正经投资理财见 wealth。 The spine is the seven-node lifecycle of A Bet (price → acquire → product → risk → settle → retain → comply), each node running traditional odds-compiler vs the AI algorithm-factory. Two nodes get their own spotlight: ④ risk = killing winners and ⑥ retention = the darkest AI application. Plus the honesty layer, the hold ladder, the leaders arena, the four-color AI-use map, a defense toolbox, and boundary expansion. This is critical structural analysis — exposition, not instruction; no betting advice whatsoever. The gamblification of finance (0DTE / prediction markets) appears as boundary evidence; real investing lives on wealth.

传统节点·赔率师Traditional · odds compiler

AI 节点·算法AI node · algorithm

抽水·庄家优势Vig · house edge

杀赢家·危害Killing winners · harm

最黑暗·精准剥削Darkest · exploitation

10.2%

2025 美国全国抽水率(hold)——2019 年还是 7.0%。AI 时代抽水不降反升(同场串关驱动),投注者每年为此多交约 53 亿美元;这一个数字就够反驳「AI 让博彩更公平」The 2025 US national hold — it was 7.0% in 2019. In the AI era the vig rose, not fell (driven by same-game parlays), costing bettors an extra ~$5.3B a year; this one number refutes 'AI makes gambling fairer'

78%

FanDuel(Flutter)+ DraftKings 两家吃下的美国合法体育博彩收入——FanDuel 的护城河不是营销,是母公司 Betfair 交易所喂给定价引擎的真实点对点数据The US legal sports-betting revenue held by FanDuel (Flutter) + DraftKings — FanDuel's moat isn't marketing but the real P2P data its parent Betfair exchange feeds the pricing engine

≈50%

被「杀赢家」限注的客户里,被限时实际是净亏损的比例(Fanatics 高管在马萨诸塞博彩委员会承认)——算法限的不是你赢了多少,是你「赢的方式」像不像职业选手The share of 'winner-limited' customers who were actually net losers when limited (a Fanatics exec before the Massachusetts Gaming Commission) — the algorithm limits not how much you won, but whether the way you win looks professional

$759亿

Polymarket + Kalshi 过去一年合计成交(约 362 / 397 亿,约一半是体育)——赌注换上「事件合约」的西装走进了金融区,边界扩张的头号证据Polymarket + Kalshi combined trailing-year volume (~$36.2B / $39.7B, roughly half sports) — the wager put on an 'event contract' suit and walked into the financial district; boundary expansion, exhibit one

口径警告:抽水率(hold/house edge)是长期统计收敛值,不是单次结果——短期方差之下玩家完全可能赢;各玩法 hold 三文档口径并列(老虎机 4–15%、串关 20–30%、SGP 常 30%+、彩票 40–50%);市场规模因 GGR/handle 口径不同分歧极大(全球在线博彩 330 亿~1550 亿美元区间);B2B 厂商性能数字(Alpha Odds +10/15%、Genius +18%、Kambi 49/50%)多为厂商自述——仅 Alpha Odds(ISAE 3402)与 GameScanner(GLI 验证 87%)有独立背书,其余一律标「公司披露」;PHAI 集体诉讼指控未经法院认定(宾州早期案已驳回大部分、上诉中);OpenBet「Neural」定价引擎为未证实传闻,不上图;部分批判是基于商业逻辑的推断。「无感」不等于「准」:英国博彩委员会自 2024 年 8 月试点财务脆弱性检查(用破产记录等公共数据评估重大财务风险),初期约 97% 的检查无感完成,高于白皮书 80% 的目标,但监管方未公开技术方案或所用模型——97% 说明自动化程度高,不说明识别得准,「没打扰到人」和「找对了人」是两个指标;同理某负责任博彩工具自称可识别「至少 87% 高风险与问题赌博案例」,该数字来自公司自述、建立在其自有行为样本上。本页揭露行业结构,不提供任何投注策略。 Basis warning: hold / house edge is a long-run statistical convergence, not a single outcome — under short-term variance players absolutely can win; per-game hold spans three sources' bases (slots 4–15%, parlays 20–30%, SGP often 30%+, lottery 40–50%); market sizing diverges hugely by GGR/handle basis ($33B–155B for online gambling); B2B vendor performance figures (Alpha Odds +10/15%, Genius +18%, Kambi 49/50%) are mostly self-reported — only Alpha Odds (ISAE 3402) and GameScanner (GLI-verified 87%) carry independent backing, the rest flagged 'company-disclosed'; the PHAI class-action allegations are not court-adjudicated (an early Pennsylvania case was mostly dismissed, on appeal); OpenBet's 'Neural' pricing engine is an unverified rumor, excluded; some critiques are inferences from business logic. «Frictionless» is not «accurate»: from August 2024 the UK Gambling Commission piloted financial-vulnerability checks (using public data such as insolvency records), with about 97% completing frictionlessly against the white paper's 80% target — but the regulator has not published the technical approach or models. 97% demonstrates a high degree of automation, not accuracy; «did not disturb anyone» and «identified the right people» are different metrics. Likewise a responsible-gambling tool's claim to catch «at least 87% of high-risk and problem-gambling cases» is vendor-reported, built on its own behavioural sample. This page exposes structure; it offers no betting strategy.

◆ 诚实层 · 地基The Honesty Layer · foundation

娱乐叙事 vs 数学现实The entertainment narrative vs the math

左边一栏是他们打广告时说的话,右边一栏是财报与判例里真实发生的事。两栏之间隔着的不是误会,是一场算力、数据、神经科学与心理学的非对称战争——而你是徒手来的那一方。The left column is what the ads say; the right is what actually happens in filings and court records. What separates them is not a misunderstanding but an asymmetric war of compute, data, neuroscience and psychology — and you are the side that came unarmed.

庄家的 AI 从不预测比赛,它预测的是你——赔率师这个工种消失了,抽水活得比任何时候都好 The house's AI never predicts the game — it predicts you. The odds compiler is gone as a trade, and the vig has never been more alive

从这张图看到的五条规律Five patterns this map makes visible