Methodology · How we build and grade decisions
Show the math. Show the doubt.
This is how Willy Walters, our fictional AI analyst, turns market data into a small number of decisions — independent projections, no-vig prices, a contrarian review, conservative stakes and a public audit. It describes our process and its limits. It does not describe results, because there are none yet.
Educational software — not a promise of profit. Most bettors lose money over time and no outcome is guaranteed. Adults 21+ only. Responsible betting.
Document status
Principles
Six rules the process cannot bend.
They come from our product blueprint and from publicly documented ideas about disciplined, value-based betting.
Independent numbers first
We build our own projection before looking at the market narrative, so we are not just echoing the price.
Price beats opinion
A good opinion at a bad price is a pass. The price, the edge and the playable range come before the team story.
Challenge every edge
A contrarian analyst argues the case against each wager, and an uncertainty haircut is applied before anything is sized.
Protect the bankroll
Stakes are small, capped and correlation-aware. Fractional Kelly may guide; hard caps override it.
Audit everything
Every recommendation is graded at the published line, compared with the close, and reviewed for process errors.
Be honest about uncertainty
Facts, estimates, assumptions and unknowns are labeled differently. A missing number stays missing.
The math
Formulas we use, in plain sight.
Every calculation behind a decision card can be reproduced. Two of them are live as free calculators.
Implied probability
Negative odds: |odds| ÷ (|odds| + 100)
Positive odds: 100 ÷ (odds + 100)Example: -110 implies 110 ÷ 210 = 52.38%. That is the win rate the price needs to break even.
No-vig market probability
p1 = implied(side 1), p2 = implied(side 2)
no-vig = p ÷ (p1 + p2)Removes the sportsbook margin so the market can be compared with our number. It is an estimate of the market.
Raw edge vs. true market edge
raw edge = model prob − implied prob
true edge = model prob − no-vig probWe judge edge against the no-vig probability, because the raw price includes the margin.
Expected value
EV = (win prob × profit) − (loss prob × stake)The average result per wager if the probability estimate is right. A theory, never a promise.
Fractional Kelly
stake = k × (b·p − q) ÷ bp = win probability, q = 1 − p, b = net decimal odds, k = a fraction such as one half. Hard caps and uncertainty adjustments override the result.
Closing-line value
compare the price we published with the closing priceA process measure: did we beat the final market? It is read with ROI, calibration and drawdown, not instead of them.
Factors
What goes into a projection.
The football factor library from our blueprint. Other sports get their own sport-specific libraries, and the first live market is chosen by what is in season.
Ratings & personnel
- Team and player power ratings
- Starting and backup quarterback value
- Opponent strength
Efficiency & pace
- Offensive and defensive efficiency
- Pace and turnovers
- Special teams
Injuries & availability
- Injury, illness and limitations
- Expected availability
- Replacement quality
Venue & travel
- Venue-specific home advantage
- Altitude, surface, crowd
- Travel and time zones
Rest & schedule
- Rest, short weeks, byes
- Consecutive road games
- Divisional familiarity
Weather
- Wind first
- Then precipitation and temperature
- Field conditions
Coaching & game script
- Coaching and scheme
- Personnel mismatches
- Expected game script
Market movement
- Open, current and close
- Whether information is already priced
- Liquidity and timing
Data flow
From a licensed feed to a logged decision.
The target architecture. Today some of these steps are manual; see How it works for what runs in the current phase.
Data governance rules
- Every commercial data source is licensed for its intended display.
- Source, timestamp, freshness and transformation history are stored.
- Model inputs and outputs are versioned and reproducible.
- The language model never invents a missing price, injury or statistic.
- A freshness threshold must pass before “bet now.”
- Every published change is logged; the originally published line is preserved.
- Model probability is structured data, kept separate from the written narrative.
Grading & audit
Judged on process, not streaks.
The ledger will be graded at the published line and audited after every settlement. No results exist yet, so none are shown.
Recordkeeping rules
- Timestamp every recommendation and its exact available price
- Never delete a published losing pick; corrections stay visible
- Grade pushes, voids and partial settlements consistently
- Show results at the recommended line, not a later favorable number
- Separate free, premium, test and unpublished research picks
- Never combine backtests with live results
- Use an independent grading or data service when practical
Postgame audit questions
- Was the input data accurate and current?
- Did we beat the closing line?
- Was the probability calibrated?
- Was the stake appropriate?
- Did variance decide it, or was there a repeatable miss?
- Did the model double-count a factor?
- Was it a good wager despite losing, or a bad one despite winning?
- Does any change have enough evidence to justify a model update?
What the public scoreboard will show
Every figure will carry a date range, sample size and methodology. Pre-launch: no figures are shown.
Model versioning & calibration
A model is a record, not a mood.
Every model has a version, documented inputs and a test history. Changes are logged with a reason, and retired versions stay on the record.
- Calibration asks: when we say 60%, does it happen about 60% of the time over a large sample?
- A model is only changed when there is enough evidence, not after one bad week.
- Backtests are labeled as backtests and never mixed with live results.
Teaching diagram: what “well calibrated” looks like. Not our results.
Model version record
Illustrative format only. No live model version exists yet.
Limitations and uncertainty
- Probabilities are estimates. They can be wrong, and any wager can lose.
- Most bettors lose money over time, and a good process does not remove that risk.
- No live model, record or measured accuracy exists yet. We do not claim one.
- Numeric edge thresholds are not set yet and will be published before launch.
- Backtests can overfit and do not predict live results.
- Data licensing and freshness limits can delay or remove a price.
- Injury and lineup news can change fast and may not be priced in our snapshot.
- The current phase is AI-assisted and human-reviewed, so people can make errors too.
Who is applying the method
A fictional character, a documented process.
Willy Walters presents the verdict. The method behind him is described on this page and is inspired by publicly documented principles, not by any private or proprietary system.
Public background that inspired the principles: the Billy Walters overview on Wikipedia and the book Gambler: Secrets from a Life at Risk. Claims about the real person are to be verified against primary sources before launch.
- Methodology at a glance (PDF · 4 pages)
- Results ledger & grading standard (PDF · 3 pages)
Methodology FAQ
Questions, answered straight.
Short answers about accuracy, data and the persona.
01Do you publish model accuracy?
No. No live model or record exists yet, so we publish no accuracy figure. Once a live record exists, we will show results with a date range, sample size and methodology.
02Is this Billy Walters’ system?
No. Willy Walters is a fictional AI analyst created by Win More Wagers, inspired by publicly documented principles. We are not affiliated with, endorsed by, or operated by Billy Walters, and we do not use any private or proprietary system.
03Why do you remove the vig?
Because the raw price includes the sportsbook margin. Comparing our probability with the no-vig market probability gives a fairer estimate of the edge.
04What happens when the data is stale?
A stale or invalidated price cannot stay labeled “bet now.” Freshness has to pass a threshold before a bet-now verdict can be issued.
Follow along
Get the free betting brief.
Get the education-first brief, and be told when the record opens. Free, and you can unsubscribe at any time.