High-level themes from internal research. Working notes — not peer-reviewed, and not a how-to for reconstructing our systems.
Results are provisional and sample-size dependent. We emphasize out-of-sample evidence and treat reported performance cautiously when many configurations have been tried.
1 · Exposure is a risk dial
Day-to-day return swings often scale with how aggressively capital is deployed. Changing exposure can change volatility without changing the underlying quality of the signal.
Classic framing of staking under uncertainty — exposure is a choice, not the edge itself.
2 · Bookmaker margins matter
Small apparent edges can disappear once the bookmaker’s margin is accounted for. Fair-price comparisons need to clear that hurdle before they mean much.
How posted prices relate to true probabilities in betting markets.
3 · Volume objectives can chase noise
Optimizing for total volume can push selection into thin edges that do not survive costs or selection bias. Risk-adjusted, out-of-sample criteria are a better north star.
Deflate for the number of trials before trusting a “winning” configuration.
4 · Diversification vs concentration
Concentrating into fewer, larger bets can raise drawdowns even when per-bet returns look better. Spreading risk across a broader opportunity set often protects the path of equity.
5 · Venue quality & exact-payoff verification
Not every market that looks “soft” is. Some venues are more adversely selected than others, and apparent cross-venue spreads frequently mask subtle rule divergences (such as first-half moneyline tie handling or player void clauses). Parsing contract rules into canonical state-payoff vectors separates true arbitrage and consensus edges from settlement traps.
Separating genuine inefficiency from variance and selection effects.
6 · Independent agreement vs. crowding
Simple copy-the-filer approaches to public disclosures rarely hold up once costs, taxes and timing are accounted for. Signals where several independent participants reach the same view on their own tend to be more durable, while moments when every popular signal agrees at once are often already crowded.
7 · Prediction markets and options can disagree
Around scheduled events, options markets sometimes keep pricing a lot of uncertainty after prediction markets have largely settled the question. When prediction markets sit near even odds, the uncertainty is usually real. Comparing the two is a useful cross-check on event risk.
8 · Different assets, different tails
In equities, a small share of big winners drives much of the long-run gain, so tight stop-losses can cut off the part that pays. In bounded-payoff contracts and physical-asset flips, there is no big right tail, so any edge has to come from the average outcome, and breadth and turnover matter more.