The bottom line
A useful result keeps its limits attached.
Log every eligible decision at the moment it is made. Keep the market, price, stake, model or rule version, closing reference, result, and notes in separate fields. Report sample size and missing data beside ROI, hit rate, calibration, and CLV.
Download the research log
Download the WagerProof research log template as CSV. The two included rows are fictional and clearly marked EXAMPLE_ONLY. Delete them before using the file.
The template preserves the fields needed to reproduce a decision rather than only its result.
Record the market as it existed
Save league, event, market, side, period, line, price, sportsbook, timestamp, and time zone. A note that says only "Team A spread" cannot be audited later. Include settlement details when books can grade the same-looking market differently.
Use one row per decision. If the line changes and you make a new decision, add a row rather than overwriting the old state.
Separate research, action, and result
The log has separate fields for the model probability, model version, rule version, research cutoff, decision, stake, closing reference, and graded result. That separation prevents later information from leaking into the original rationale.
Record passes as well as actions when they came from a defined screening rule. Otherwise the record can hide how often a model produced an apparent signal that failed a final check.
Use consistent unit and result rules
A unit is a reporting convention, not a safety mechanism. Define it before the sample and do not enlarge it after losses. Preserve the actual currency stake privately when needed for reconciliation, while public analysis can use units.
Track wins, losses, and pushes separately. Net profit or loss should include the actual price rather than assuming every win returns one unit. Decide whether fees, exchange commissions, and promotions are included, then state that policy.
Calculate common metrics without hiding the denominator
For a stated sample:
Hit rate = wins / (wins + losses)when pushes are excluded and disclosedROI = net profit or loss / total stakedAverage price = a clearly named price representation, not a casual average of American oddsPositive CLV share = records with positive CLV / records with a valid closing reference
Every percentage should travel with its denominator. A 60% hit rate over five resolved records is not equivalent evidence to 60% over five hundred comparable records.
Keep calibration, CLV, and return distinct
Probability calibration asks whether forecasts behave like their stated probabilities across a suitable sample. Closing line value asks whether the recorded entry price compares favorably with a defined later reference. ROI reports realized net return relative to stake.
These measures can disagree. A short sample can have good CLV and negative results, or a winning record with poor price quality. Do not choose only the measure that flatters the period.
Version every process change
When a model, input, filter, agent setting, or selection threshold changes, create a new version. Keep the previous rows intact. Write why the change occurred and whether it was planned before seeing the latest results.
This is especially important for saved historical Systems. Editing the rule after a losing stretch and displaying only the new version erases the evidence needed to judge the process.
Review in fixed windows
Choose a weekly or monthly review cadence. Break results down by sport, market type, price range, model version, and decision rule only when the groups were defined or are labeled exploratory. Report missing closing prices and unresolved events rather than dropping them.
Ask four different questions:
- Was the input data complete at the cutoff?
- Were the probability forecasts calibrated?
- Did the process capture competitive prices?
- What realized result occurred after variance?
The record becomes more useful when it can answer an uncomfortable question without being rewritten.
Protect the person behind the spreadsheet
Tracking can improve transparency, but constant checking can also intensify chasing behavior. Set time and money limits before a session, do not increase stakes to repair a metric, and use the responsible research guide when the process stops feeling controlled.
Plain answers
Frequently asked questions
What fields belong in a sports betting research log?
At minimum record event, market, side, line, price, sportsbook, timestamp, stake, result, net profit or loss, closing reference, model or rule version, research cutoff, and notes. Keep pushes and missing values explicit.
How is ROI calculated?
Divide net profit or loss by total amount staked for the stated sample, then multiply by 100 for a percentage. Always report total staked, sample size, date range, and whether bonuses or fees are included.
Why should model version and rule version be tracked?
A changed model, filter, or selection rule creates a different process. Version fields prevent an updated rule from being mixed invisibly with the record that preceded it.
Evidence ledger
Sources
Sources were opened on the listed date. Product features, prices, platform coverage, fees, and help information can change.
- Strictly Proper Scoring Rules, Prediction, and EstimationJournal of the American Statistical Association · Accessed August 26, 2026
Probability forecast evaluation and separation from realized results
- WagerProof App Store listingApple and WagerProof, LLC · Accessed August 26, 2026
Current first-party automatic grading and visible record claims
- Closing Line ValueOddsJam Betting Education · Accessed August 26, 2026
Common CLV terminology
About the author
Chris Habib
Chris builds WagerProof and writes practical guides about sports research, product methodology, and responsible use of betting data. He focuses on inspectable workflows rather than outcome promises.
Research inside WagerProof
Put the model, market, and record on the same screen.
WagerProof organizes model probabilities, current lines, agents, trends, and graded records. It does not guarantee an outcome.
Have a correction or a newer first-party source? Email the editorial team.
