How Fast Do Bookmakers Follow Pinnacle? A Measurement You Can Run Yourself

Everyone says soft books copy Pinnacle's line; almost nobody measures the lag. A reproducible way to time it — and why that lag is the whole edge.

How Fast Do Bookmakers Follow Pinnacle? A Measurement You Can Run Yourself

One claim appears in nearly every piece of writing about sharp betting, ours included: soft bookmakers follow Pinnacle's line. It's repeated so often it feels like a physical constant. But ask the obvious next question — how fast? — and the literature goes quiet. "Seconds to minutes" is the usual hand-wave, which is a strange level of vagueness for a number that determines whether entire strategy classes are viable.

So instead of repeating the folklore, this post does two things: defines the measurement properly, and gives you a harness to run it yourself. It's the companion piece to our latency benchmark, and it exists for the same reason — numbers you can't reproduce are marketing.

Why the lag is the whole ballgame

Mechanically, most sharp-line strategies are the same trade: Pinnacle moves, a soft book hasn't moved yet, and the stale soft price is briefly wrong. Value betting, arbitrage, move-following — all of them live inside the follow window and die at its edge.

Which means the window's width isn't trivia. If soft books reprice in 3 seconds, no human — and no polling bot — is inside the window. If they take 90 seconds on midweek second-division fixtures, the window is wide enough to drive a truck through. Same strategy, opposite outcomes, and the deciding variable is the one nobody publishes.

Defining the measurement (the part everyone fumbles)

"Follow lag" sounds simple until you try to log it. The definition we use:

Lag = t(soft book's first reprice in the same direction) − t(Pinnacle's move), measured per market, counted only when Pinnacle's move exceeds a significance threshold.

Each piece is doing real work:

• A significance threshold on the Pinnacle move (we use a no-vig probability shift, not raw price, so it's comparable across odds ranges — see why no-vig). Without it you're timing noise against noise.
• Same direction — a soft book moving the other way isn't following, it's disagreeing (or rebalancing its own flow).
• First reprice, not full convergence. Books often move partway immediately and drift the rest. The first reprice closes the exploitable window; that's the timestamp that matters.
• Per market, per league. Averaging Premier League moneylines with Finnish third-division totals produces a number that describes nothing.

The harness

Two data streams and a clock you trust:

1. The Pinnacle side needs push delivery — this is the one place polling genuinely can't work, because your poll interval becomes measurement error bigger than the thing measured. We use our own SSE drop stream; the latency benchmark shows the same wiring on the reference side, harness code included.
2. The soft-book side can be polled (an aggregator works), because you only need the reprice bracketed, and its poll interval is honest, quantifiable error you report alongside the result: a 5s poll means every lag is "X ± 5s."
3. One clock. Timestamp both streams on arrival at the same machine. Never trust two vendors' timestamps against each other.

Log every significant Pinnacle move, watch the matched soft market until it responds, write both timestamps. Run it over a real sample — a weekend of fixtures minimum, across at least two league tiers — and you have a distribution, not an anecdote.

What we're seeing (and what we won't claim yet)

We're collecting the sample now — logging every significant Pinnacle move against matched soft markets, across league tiers, exactly as described above. We will publish the distribution on this page once it's a distribution rather than an anecdote: medians and p90s per segment (top-league soccer prematch and live, lower-league soccer, tennis), each with its sample size attached. No numbers until then; a table of invented lags would defeat the entire point of this post.

What we will say before the table fills in, because the mechanism guarantees it: the lag is not one number. Liquid, headline markets reprice fastest — some soft books automate rebalancing off sharp reference feeds precisely there. Niche leagues and secondary markets rely on slower processes, sometimes human ones. The exploitable structure lives in that spread, which is exactly why a single average would mislead you.

The three honest caveats

• Not every soft-book move is a follow. Books also move on their own liability, promos, or copied third-party feeds. The same-direction + threshold rules filter most of this; they don't filter all of it.
• Measuring it and beating it are different problems. Your own stack adds latency on top — that's the other benchmark.
• The window is shrinking where it's most watched. Automation spreads from the top leagues down. A measurement from 2023 is a historical document, which is precisely why the harness matters more than any table of ours.

Takeaway

"Soft books follow Pinnacle" is true, repeated, and useless without a number attached. The lag varies by book, league, and market; it is measurable with two feeds and one clock; and it is the single variable that decides whether a sharp-line strategy is an edge or a story. Measure it for the markets you care about — and be properly suspicious of anyone selling you the strategy without showing you their distribution.

FAQ

Do all bookmakers follow Pinnacle's lines? No — sharp books lead alongside it, exchanges price independently, and soft books follow at very different speeds depending on market and automation. "Follow" is a spectrum, not a rule.

How big is the lag typically? It ranges from near-instant on automated headline markets to minutes on neglected ones. Distrust single-number answers, including from us — measure your target markets.

Why does the Pinnacle side need push delivery for this measurement? Because a polled Pinnacle feed adds your poll interval as error on the reference timestamp — you'd be measuring your own architecture, not the market.

Can I run this without pinnapi? Yes — any Pinnacle source with true push delivery works for the reference side, and the soft side just needs bracketed polling. The method is the point; the vendor is swappable.