Where Multiple Signals Converge
How Optiqal scans every active sports and prediction market simultaneously, assembles objective evidence across six independent criteria, and only surfaces picks where the case is airtight on multiple fronts.
The Premise
Markets Are Hard to Beat, and That Is the Point
Modern sports betting markets, particularly lines anchored to sharp books like Pinnacle, absorb enormous amounts of information. The average game, on the average night, is priced efficiently. Single-factor models that look at one angle of a matchup and call it an edge will be wrong more often than the market.
Durable edges arise differently: when several independent, objectively measurable signals all point the same way at the same time. Sharp line movement saying one thing. A cross-market price gap saying the same thing. A demonstrable situational or personnel advantage on top. When evidence from genuinely separate sources converges, the probability of a mispricing is much higher than any single signal alone could establish.
Universal Scope
Every active league from The Odds API is scanned each run. Edges do not announce themselves by sport — the system finds them wherever they exist.
Multi-Criteria Gate
A pick requires at least three independently confirmed criteria. The threshold is non-negotiable: two signals, no matter how strong, do not clear the bar.
Honest Absence
Criteria the system cannot objectively establish are reported as absent — never constructed or fabricated. A thin evidence base correctly produces no pick.
The bar is intentionally high: on an efficient slate where lines are well-priced, the model correctly declines to bet. Volume is not the goal. Picks with a genuine confluence of evidence are.
The Six Criteria
Independent Signals, Not Correlated Noise
Think of the criteria as six completely different windows onto the same matchup. A sharp-money signal and a cross-market price gap both come from the price feed — they are not independent. Sharp money and an injury burden asymmetry come from entirely different data sources, with no shared inputs. The system is designed to assemble evidence from genuinely separate categories so that confirming all three is harder to do by chance than confirming three signals that share an underlying data source.
Each criterion has a state: confirmed (strong, directly evidenced), probable (weaker or indirect support), or absent (no objective signal available). Only confirmed criteria count toward the three-signal gate.
Sharp Money
ActiveTracks Pinnacle's opening-to-current line movement toward the selection and the offset between Pinnacle's fair price and the median soft-book price. Line movement at a sharp book reflects informed money flowing in — it is one of the strongest objective signals the public price feed can provide.
Line Value
ActiveComputes the price gap between the sharp sportsbook and any tradeable prediction-market price for the same event. When the sharp book and a liquid prediction market price the same outcome differently, that divergence is an objective signal — not a narrative.
Injury and Rest Edge
ActiveMeasures asymmetric injury burden between the two sides by key position (quarterback, starting pitcher, primary goalkeeper). Both teams must be covered by the live injury feed for this criterion to compute; when coverage is missing, it is reported absent rather than assumed neutral.
Head-to-Head Situational
Active (NBA / NHL / MLB / NFL)Looks up recent decided meetings between the two exact teams, excluding ties and spring-training noise. A matchup must have at least five historical meetings and an 80% or higher win rate for one side before this criterion confirms — a single-season sample does not qualify.
Starter Confirmation
Active (MLB)Confirms the starting pitcher from the official MLB Stats API and evaluates the ERA differential between the two probable starters. For this criterion to confirm, the starter must be officially announced and the ERA edge must clear a minimum threshold — a marginal differential does not count.
Contrarian Value
Honest AbsentWould measure public betting-percentage distribution to identify overcrowded sides. No US-accessible public betting percentage feed currently provides this data reliably. Rather than approximate it from a secondary proxy, this criterion is reported absent until a verifiable data source exists.
The Confluence Engine
Evidence First, Pick Decision Second
Every criterion is computed before any pick decision is made. The system assembles the full evidence packet — prices, criteria states, edge calculations, liquidity — and only then reasons over it to determine whether the evidence supports a bet and on which side.
This sequencing matters. A system that picks a winner first and then finds supporting criteria has chosen its conclusion before examining the evidence. The confluence engine does the opposite: it computes every criterion independently, hands the assembled evidence to a reasoning layer that has no prior preference, and lets the weight of confirmed signals drive the outcome. A judgment of “no bet” on an efficient slate is the correct output, not a failure.
The reasoning layer can only select a pick whose selection, bet type, and venue all appear in the assembled data packet. Any response that references data not present in the packet is rejected automatically — the pick never publishes.
Canary Verification
Every response must echo a unique identifier embedded in the data packet. A mismatch — indicating the response was not grounded in the correct matchup — discards the pick entirely, regardless of how compelling the reasoning looks.
Three-Confirmed Gate
Only confirmed criteria count. Probable signals do not contribute to the gate. Three confirmed criteria is the hard floor; picks with two confirmed and several probable are declined.
Price Edge Prerequisite
The selection must show at least a 3% price edge over the de-vigged fair probability before the confluence check even runs. Criteria are evaluated only for matchups that already show a meaningful price discrepancy.
Invalidators on Record
Each pick records the conditions that would most likely invalidate the thesis — early pitcher pull, a late lineup change, sharp reversal on the line — so the pick's own fragility is documented alongside its case.
Universal Coverage
Every Active League, Every Day
Earlier sport-specific models are locked to a fixed list of leagues and become dormant when those leagues go out of season. Optiqal operates differently: each run discovers every currently active match league from the odds feed — including FIFA World Cup, WNBA, KBO, NPB, tennis, boxing, and any league that opened since the last run.
This matters because edges do not cluster by sport. A sharp line discrepancy might exist in an international soccer league during MLB's midseason while every NBA and NHL market is quiet. A sport-locked model never sees it.
One known exclusion: eSports (CS2, League of Legends). The odds feed does not include a Pinnacle sharp-anchor for esports, so there is no reliable baseline for de-vigging. Esports is covered by dedicated models that use the correct data sources for those markets.
Multi-Venue Routing
Best Price, Best Venue
When a qualifying pick is identified, the system selects the best-priced executable venue across both sportsbooks and prediction markets. The price comparison is fee-adjusted: prediction markets charge trading fees that reduce the net payout, so a 62-cent Polymarket price is worth less than it looks versus a sportsbook offering the same outcome at -162.
Pinnacle is used as the sharp-book anchor for de-vigging and edge calculation. The system pulls odds from multiple US-legal sportsbooks and routes the pick to the best available line among those clearing the liquidity minimum.
Polymarket and Kalshi are queried for the same event. Both are subject to a liquidity floor (minimum open interest) before they qualify as an execution venue. A thin prediction market with a favorable price is not a valid route — it cannot be exited at size.
The Math Layer
De-Vigging From a Sharp Anchor
Most public sportsbooks build 5–10% margin into their lines. Taking odds at face value as implied probabilities overstates the probability on both sides. De-vigging strips that margin out to arrive at the market's actual fair price for each outcome.
Pinnacle, which runs a low-margin book and attracts a disproportionate share of sharp money, is used as the de-vig anchor. Its lines provide the most reliable available estimate of where an efficient market values each side.
Three-way markets: Soccer 1X2 is a three-outcome market with roughly 25-30% implied probability on the draw. De-vigging only the two team prices as a two-way market would ignore the draw mass entirely and artificially inflate both team probabilities. Optiqal de-vigs across all three outcomes proportionally for any market that includes a draw outcome.
Quality Filters Before Criteria
Before the criteria engine is consulted, each potential selection must clear objective pre-filters. These run on the data and numbers alone, with no judgment required:
Odds data older than 30 minutes is not used. Stale prices may no longer reflect the market.
Games more than 30 hours away are excluded. Picks are for same-day or next-morning events with settled lineups.
3% minimum edge over the de-vigged fair probability. Smaller edges are below the noise threshold and absorbed by the vig.
Fee-adjusted EV must be positive above a minimum. An edge exists on paper but the venue's fees eliminate the economic case.
Maximum five picks per day across the entire run. Volume is not a goal; pick quality is.
A per-day cost ceiling prevents runaway inference costs on very large slates. The cap is tuned against typical active-league counts.
Unit Sizing
Conviction Determines Size
A pick that clears the three-confirmed gate with exactly three signals is less compelling than one where five independent criteria all point the same direction. Unit sizing reflects this: the more criteria confirmed, the larger the position. This is not arbitrary — more confirmed signals from genuinely independent sources means the probability of a real mispricing is higher, and a larger bet is mathematically justified.
The minimum qualifying pick. Strong enough to surface, sized conservatively to respect that the gate was cleared by one signal of margin.
A clear case where evidence from a fourth independent category has aligned. Sized up proportionally.
Maximum confluence. Five or more independent signals all pointing the same direction represents the strongest case the system can produce.
Note: Even maximum-confluence picks lose. The unit sizes reflect expected value over a large sample, not certainty on a specific event. Any individual outcome is noisy regardless of how many signals aligned.
How We Measure Success
Closing Line Value (CLV)
Every pick is locked at a snapshot price and tracked against the closing line. Consistently beating the closing line is the most objective measure of edge — if the market moves toward your pick after you take it, the market confirmed your thesis.
Criteria-Level Accuracy
Each confirmed criterion is tracked against actual outcomes separately. A criterion that consistently confirms on the losing side is a candidate for recalibration. The six criteria are expected to show different edge profiles over time.
Automated Settlement
Results are graded automatically using verified results from official sources. Every pick has a documented paper trail: data packet, confirmed criteria, pick odds, closing odds, result, and profit or loss.
Honest Decline Rate
The fraction of scanned matchups that produce no pick is a signal of model health, not failure. A model that declines to bet on 95% of the slate on an efficient day is working correctly. Artificially high pick volume would be the problem.
Technical Breakdown
Data Sources
| Source | What It Provides |
|---|---|
| The Odds API | All active match sports + odds from 30+ US sportsbooks including Pinnacle. Used for league discovery, de-vig anchor, and line-movement tracking. |
| Polymarket | Prediction-market prices for games covered by the platform. Matched to sportsbook events by team name + game date. |
| Kalshi | Prediction-market prices and open interest. Liquidity gate applied before the venue qualifies for routing. |
| Ball Don't Lie (BDL) | Live injury reports, head-to-head game history, and player statistics across NBA, MLB, NFL, and NHL. |
| MLB Stats API | Official probable-pitcher announcements for starter confirmation. Used directly from the league's public endpoint, not inferred. |
Projection Pipeline
| Step | Process |
|---|---|
| 1. League Discovery | Pull every active match sport from the odds feed. Exclude futures and outrights. Apply any operator-configured exclusions. |
| 2. Odds Ingestion | Fetch moneyline odds across all discovered leagues. Extract Pinnacle anchor, soft-book markets, and prediction-market prices. |
| 3. Pre-Filtering | Apply staleness check, lookahead window, liquidity floors, edge floor, and EV floor. Discard anything that fails. |
| 4. Criteria Assembly | Run all six criteria providers independently for each qualifying matchup. Record confirmed / probable / absent with cited evidence. |
| 5. Packet Build | Assemble the full data packet: prices, fair probs, edge, EV, criteria states, and a unique canary identifier. |
| 6. Confluence Reasoning | Pass the assembled packet to the reasoning layer. It selects a grounded pick or declines. No numbers are computed at this step — only judgment is applied. |
| 7. Validation | Verify canary echo, selection traceability, and confirmed-criteria count. Any failure discards the pick. |
| 8. Write + Size | Code attaches all numbers (edge, EV, size by confirmed count). Pick is written as shadow until the operator flips live. |
Daily Schedule
| Timing | Job |
|---|---|
| Morning (8–10 AM) | Main projection run: discover active leagues, build packets, run confluence engine, write picks for the day's slate |
| Pre-game (rolling) | Lock picks as game times approach; snapshot odds at lock time for CLV tracking |
| Hourly | Odds sync: update closing odds on locked picks for CLV calculation |
| Post-result | Settlement: grade locked picks against verified results from official sources |
The Bottom Line
Optiqal is built on one premise: a bet is only worth taking when multiple independent signals agree. The philosophy in practice:
- We scan every active league simultaneously — edges do not organize themselves by sport
- We compute all six criteria independently before any pick decision is made
- We require three or more confirmed signals from genuinely separate categories
- We route picks to the best fee-adjusted price across sportsbooks and prediction markets
- We de-vig from the sharpest available anchor, using three-way math for soccer markets
- We verify every pick is traceable to the exact data packet it was reasoned from
- We size positions by the number of confirmed signals: more confluence, larger bet
- We track closing line value and per-criterion accuracy to validate every assumption
Most days, the model declines. That is not a bug. On an efficient slate, the correct answer is no bet, and the system is designed to say so.
Full Access
- Less than $1 a day — one winning pick can pay for months of access
- +111 units of tracked profit — that's +$11,100 at $100/unit, verified in public
- Zero research needed — the exact pick, odds, and bet size handed to you every day
- Everything included — every sport, every model, every market. No tiers, no upsells