Analyze the prop.
The model starts with the player, market, line, book, game context, team environment, and recent usage.
Methodology
How the engine thinks.
Propeller converts player context, market movement, matchup data, and recent form into one directional signal score. Historical rows are later graded against final stats, with archive limitations documented separately.
By Scott Olmer, Founder · Reviewed July 13, 2026
Score the direction.
Signals lean the prop toward OVER or UNDER. Confidence measures how far that signal is from neutral, not whether it is an OVER.
Grade the outcome.
After the event, the final stat determines win, loss, or push. The historical archive updates from those graded rows.
Agent system
Separate signals, one research view.
Each agent looks at a different part of the prop. The product experience is intentionally simple, but the model keeps the reasons separate so users can inspect what is driving a pick.
Opponent tendencies, position context, pace, and defensive shape.
Minutes, role, workload, lineup changes, and recent opportunity.
Book lines, movement, implied probability, and available prices.
Recent production, volatility, and trend quality without blindly chasing streaks.
Status updates, replacement usage, and team-level ripple effects.
Total, spread, pace environment, and how the matchup is likely to play.
Why UNDERs matter
High confidence is not the same as high score.
The raw model score is over-perspective. A low raw score can be a very strong UNDER. That is why the public record normalizes confidence before grouping results by range.
Audit path
The record has to reconcile.
Sport totals, signal ranges, OVER/UNDER splits, and the historical totals come from public API endpoints. Raw rows and collapsed ledger entries are kept separate, and neither is presented as a forward-tested ROI record.
Historical archive
Legacy data is research context.
The archive includes repeated snapshots and retrospective records. It does not expose a reliable publication time, model version, or evaluation mode for every row.
Forward ROI standard
Featured public picks use a stricter contract.
Eligible high-confidence picks are frozen before a verified event start with the exact side, line, named book, and American price captured no later than publication. Wins earn price-based profit, losses lose one unit, pushes return zero, and void or unpriced outcomes are excluded. ROI equals net units divided by priced settled picks.
Founder 500
Use the method on today's board.
Founder 500 members get free lifetime core access while launch spots remain.