How Consensus Works
Answer Hub · Concepts
“Consensus” is the foundational concept behind every Otterline pick. This page explains what it means in plain language, why it matters, and how to read the consensus data on the dashboard.
What “Consensus” Means
In sports analytics, consensus means that multiple independent prediction sources have arrived at the same conclusion about a game — specifically, they've picked the same team to win. The keyword is independent: if two models share the same input data or the same methodology, their agreement tells you almost nothing extra. Real consensus requires genuine independence.
High consensus (3–4 models agree)
Multiple independent sources have reached the same conclusion despite different inputs. Historically, high-agreement spots perform better than low-agreement ones across most sports and sample sizes.
Low consensus (1–2 models agree)
Models disagree more than they agree. These spots have no consistent historical edge in Otterline data and are typically filtered from curated buckets unless other factors compensate.
Why Agreement Alone Is Not Enough
Raw agreement percentage is the starting point, not the signal itself. Two critical adjustments are applied before any pick is considered:
1. Combo Win Rate
Not all agreeing pairs are created equal. The Otterline Model + Dunkel Index might have a historical win rate of 62% when they agree on an NHL pick. The CBS + Mladek combination might win at 54% when they agree. Same raw consensus percentage (50%), very different value. Combo win rates are tracked individually for every combination of 2, 3, or 4 models and updated as new results come in.
2. Market Comparison
Even a high-consensus, high-combo-win-rate pick isn't necessarily a good bet if the market already agrees. If the books have priced a team at -280 (implied ~74%) and our models rate them at 72%, we're paying for certainty the market has already absorbed. The edge only exists where the models exceed market implied probability — not where they confirm it.
Reading the Consensus Board
The consensus dashboard (available at /nhl/consensus and /nba/consensus) shows each game on today's slate with a summary of model agreement.
Consensus %
The percentage of models that picked the displayed team. 75% means 3 of 4 models agree. This is the primary sort key on the board — higher is better, but check the combo record before acting.
Combo record
The historical W-L and win rate for the specific combination of models shown agreeing. A 75-consensus pick where the agreeing trio has a 63% historical win rate is more compelling than one at 51%.
Confidence tier badge
The label assigned to the pick (Power Play, Diamond, etc.) based on meeting the published threshold criteria. No badge means the pick doesn't meet any tier's minimum — visible but not curated.
Market implied
The probability the market assigns to this outcome, derived from book lines or Polymarket. Compare against consensus % to judge whether the model edge has been priced in.
Model breakdown
Which specific models are on each side. Visible on the expanded view. Useful for cross-checking combo records and understanding why a pick is or isn't in a curated bucket.
Common Misreads
"100% consensus means bet big."
100% means all four models agree — that's a strong signal. But if the market already prices this team at -250, the edge may already be gone. Always check the implied probability column.
"Low consensus = skip entirely."
Some low-consensus picks have strong combo records for the specific pair of models agreeing. The Mladek Spotlight, for example, often fires on a single model signal where the market edge is the primary filter — not raw agreement.
"The consensus board tells me how many to bet today."
The board shows available picks, not a recommended portfolio. Most users should select 1–3 plays from the curated buckets (Power Plays, Diamonds, etc.) rather than betting the full board. Quality over quantity applies here.
"If consensus drops after I bet, I should hedge."
Consensus can shift as late model updates come in, but the pick was posted based on the data available at posting time. The system is pre-game and static once published. Don't hedge based on intraday consensus drift.
A Note on Sample Size
Consensus and combo win rates only become meaningful over large samples. A combo with a 65% win rate over 20 games is statistically unreliable — sample variance can easily explain that outcome. A combo with a 61% win rate over 300 games is a different story. Otterline enforces minimum sample constraints before publishing combo records, but users should still prefer buckets with longer history when evaluating which tier to focus on.
Evaluate your own results over 50–100+ bets minimum before drawing conclusions. Sports betting systems require patience and discipline to judge correctly.