If you run an arb or +EV scanner long enough, your accounts at the soft books start hitting limits. Bets that used to clear at $500 max get held for "trader approval." Eventually the limit drops to $5. Books call this "stratification" or "limit management"; bettors call it "getting gubbed."
This is a feature of the system, not a bug. Books model their customer mix and use limits to keep informed action small. Understanding which signals they watch is the difference between an account that lasts 18 months and one that gets limited in 6 weeks. This post walks through the seven signals public-facing sportsbooks use, in roughly decreasing order of weight.
The strongest single signal. If your bets consistently land within seconds of a line move at the sharp anchor, that's coordinated sharp action by definition. Books log the latency between sharp-book move and your bet timestamp; tight latency, repeated, is unmistakable.
Mitigation: don't pattern-bet immediately after sharp moves. Sprinkle delays. Bet some lines BEFORE the sharp moves on your own model (which makes the timing look like price discovery, not following).
Books compute, post-game, what fraction of your bets landed on the side that closed sharper. Recreational bettors land on the closing-favored side roughly 50% of the time (noise). Sharp accounts land on the closing-favored side 53-58% of the time. That gap is the bookie's classifier.
This is closing-line value (CLV), and the books grade it the same way you do.
Mitigation: this is the one signal you can't fake without giving up edge. The whole point of +EV is to have positive CLV. The books WILL detect it eventually. You can only delay this by limiting your edge per bet and varying which books you use.
If you bet $200 on the recreational-popular slate-headline NFL game and $200 on a 4 PM ET MLB game between two non-playoff teams, you're being inconsistent. Recreational bettors bet more on big games. Sharp accounts bet the same amount everywhere because their edge is uncorrelated with public attention.
Books look for "flat staking with no recreational tilt" as a sharp signal.
Mitigation: scale stakes with public-game popularity. Bet more on flagship games even if your edge is the same. You'll capture less per bet but stay live longer.
Promos (boosted parlays, "free bets," profit boosts on specific markets) are recreational-targeted. Sharp accounts ignore them because they don't fit a +EV play. Recreational accounts take all of them, badly.
Books compute promo-uptake rate as a feature. Zero uptake = sharp signal.
Mitigation: take some recreational-shaped promos, even if they're slight -EV. The cost is small; the cover is significant.
Books track:
Mitigation: log in casually. Browse markets. Place occasional small recreational-shaped parlays. Don't make every visit a bet.
Sharp accounts withdraw early and often. Recreational accounts re-bet winnings until they lose it back. Books use withdrawal rate as one of the strongest behavioral signals.
Mitigation: this is genuinely hard. The smart play is to withdraw on your own schedule (e.g. monthly) and not perfectly tied to win timing. Don't withdraw 100% of profits within minutes of a winning bet.
VPN traffic, datacenter IPs, residential-proxy patterns, multi-account fingerprint overlap (same browser canvas signature across two "different" accounts), all flagged. Books cross-reference this against their existing flagged-account database. If you fingerprint similar to a previously-limited account, you start limited.
Mitigation: use your real residential IP. Don't share devices across accounts. Don't use known browser-automation user agents.
If you've read our latency budget for a betting agent, you know we ship sub-second WebSocket and we genuinely think latency matters. But here's the honest framing: latency optimization helps with signal #1 (timing). It helps zero on signals 2-7.
Past a certain point, more speed makes signal #1 WORSE, not better. If your bet hits the book 50ms after a sharp move, that's so tight it screams "automated arb bot." Books model maximum-plausible-human-latency as a feature; bets that beat it are sometimes auto-rejected outright.
Three modes of operation, in increasing sophistication:
Bet mostly recreational-shaped stuff (parlays, boosted props, headline games) to build the account profile. Sprinkle in +EV bets at a rate of maybe 20-30% of your total volume. Mix stake sizes. Take promos. Withdraw monthly. Lasts 12-24 months at flat-line limits.
Operate 4-8 books actively. Each individual book sees only a fraction of your action. When one limits you, the impact is bounded. You also gain access to more line-shopping opportunities. Costs: bankroll fragmentation, KYC paperwork on more accounts, more time managing.
Each book gets a different style of account. One is "recreational NFL bettor," one is "MLB props specialist," one is "live-betting hobbyist." Each persona is consistent with itself. Limits stay high because no single book sees the cross-product +EV pattern.
This is what professional sports-investment shops do. It requires infrastructure (you're effectively running multiple model-runners with different profiles) but extends account longevity dramatically.
What we help with:
What we don't help with (and no API can):
Books also use machine-learning classifiers (gradient-boosted trees, mostly) over all the signals above plus dozens more. Hard rules are easier to defeat than ML classifiers because you can probe the rules; ML classifiers expose less. Some books retrain monthly. The right framing isn't "what triggers the rule," it's "what does my account distribution look like compared to a known-good recreational account in the same demographic."
The accounts that last are the ones that look like themselves, consistently, over time.