High Confidence
75+193K raw rows171,108-21,718-10
OVER60.8%1,936
UNDER89.0%190,900
Methodology · Calibration · Grading
How the archive is built.
Historical rows start with a line, directional signal score, market, and side, then are graded against the final stat. The archive includes repeated snapshots and legacy retrospective data, so it must not be read as a uniquely published forward-test.
By Scott Olmer, Founder · Reviewed July 13, 2026
Past performance does not guarantee future results. Propeller is a research and analysis tool, not a sportsbook.
Confidence by range.
Directional signal strength counts strong OVER and strong UNDER signals the same way. These buckets summarize raw historical rows, not a prospective recommendation cohort.
88.7%
Medium Confidence
65-74874K raw rows685,773-188,676-22
OVER61.8%21,708
UNDER78.8%852,763
Slight Confidence
55-641.27M raw rows801,912-463,263-34
OVER56.1%245,662
UNDER65.1%1,019,547
Below 55
<55621K raw rows331,066-290,032-22
OVER49.6%268,404
UNDER56.1%352,716
2,953,636 raw historical rows. Buckets describe the archive and are not a forward-test or ROI claim.
The grading path
From stored analysis to historical result.
The current archive documents its grading path while being explicit about what the legacy data cannot prove.
Store the analysis row
The public feed exposes player, market, line, direction, signal score, sport, game date, and book. It does not expose a reliable publication timestamp for every historical row.
Grade against the final stat
Overs need the final stat to finish above the line. Unders need it below the line. Exact ties are pushes.
Refresh the historical archive
Wins, losses, pushes, sport splits, and signal ranges refresh from the public APIs. Raw summaries and collapsed ledger entries remain separate units.
What the record means.
The goal is not to make the number sound bigger. The goal is to make the number understandable.
Pushes are shown in the record, but they are tracked separately from wins and losses.
A large row count can add research context, but repeated observations do not equal the same number of independent recommendations.
A confidence score is a directional research signal, not a calibrated probability or guarantee unless a page explicitly says otherwise.
New public recommendations will use an immutable pre-event publication contract rather than being backfilled from the mixed historical archive. Read the grading answer and corrections policy.
Founder 500
Use the record, then research today's board.
Founder 500 members get free lifetime core access while launch spots remain.