Paper research · live operations
Tennis pipeline health, decision eligibility, model diagnostics, and paper P&L.
Operations first
Pipeline execution, accepted prediction state, and browser delivery are reported separately.
One primary group per blocked row; open a match for its exact audit reasons.
“Feature complete” means required values are present. It does not certify that upstream match history, duplicate quarantine, or train/serve parity passed an independent audit. Exact lineage remains visible on every match.
Accepted prediction run
Only rows from the last successful prediction-bearing run are considered.
Loading the accepted run…
The promoted production snapshot is complete. Exact-version candidate status is shown separately inside each match.
These are monitored but not actionable. The reason is stated on each row.
Started, suspended, or source-time-elapsed matches retained for audit only. They are never live recommendations.
Paper-only model evidence
Parallel populations—not one sample
Prediction cohorts measure model quality. Counterfactual bets replay a rule on those predictions. Placed-bet outcomes measure the paper ledger; attribution-uncertain recoveries release capital and preserve P&L but never become model evidence.
Open dated full ledger snapshot This report may lag the accepted sync shown above.
Scalar summaries
These are scorecards, not diagnostic curves. Use ROC and calibration below to inspect model behavior.
Version-aware comparison · cohort labels explicit
Choose any model lines below. Live-forward curves use the growing exact-settled sample and show their own n. With sealed replay enabled, every selected candidate, market, and production-reference curve uses the same fixed cohort. Dots are the exact authoritative thresholds; smoothing only rounds the visual guide and never changes AUC or the underlying points.
One curve · every exact threshold
Inspect any available production, exact-version live-forward, or sealed-replay curve and its full threshold table. Use the versioned comparison above to overlay selected lines. The diagonal is random ranking; AUC summarizes the area under the curve.
Reliability diagram
Dots above the diagonal are underconfident; below are overconfident. Empty bins are omitted.
Scalar comparison—not a diagnostic curve
Candidate lane · never actionable
Candidate reliability
Each shadow variant uses one deterministic opening observation per match and model version, joined to the operational opening feature snapshot. Hourly repeats do not increase n. “Actual paper” performance remains in Paper bets; both flat figures here are counterfactual cohort replays. Bovada uses the offered decimal odds and raw break-even hurdle. Kalshi uses each side’s raw ask as the hurdle with no de-vigging, then requires the exact per-market rule envelope, entry-time fee evidence, and exchange terminal value before scoring P&L. Generic match winners and generic filings are not settlement authority. Treat n below 250 as exploratory even when ROI is green.
Strategy research
Candidate comparisons and recorded wagers are separate evidence.
Loading verified replay…
Conservative scenario, not actual wagers: capital is released at 00:00 UTC two calendar dates after the logged settlement date. Unknown results keep stakes locked. This is not the sportsbook’s payout timeline or proof of future profitability.
Starts with the reference capital above. Stakes use the selected fraction of full Kelly, up to 5% of current equity per bet and an 18% new-allocation budget per prediction run. All entries in the same prediction run share one allocation batch, scaled proportionally if cash or exposure limits bind. Equity compounds only realized outcomes; open positions stay valued at cost. No minimum wager, cent rounding, fees, or in-play mark-to-market is modeled.
Qualification remains a 2-point edge over raw price break-even. Each model has its own simulated account on the same audited match opportunity set, including unresolved outcomes. Max drawdown uses the complete event timeline; charts show daily closing points. The slider explores the same historical sample—it does not select an optimal live strategy. Snapshots refresh every six hours when the production pipeline is idle; a busy pipeline retains the previous dated snapshot.
Hypothetical, non-compounding comparison. Open stakes are not reserved here; this is not an executable bankroll replay.
One opening prediction per match and exact model release, on the same settled forward GOLD matches as Performance. Entries require at least a 2-point probability edge above the offered odds’ raw break-even probability. Market fair probability is not the betting hurdle.
Flat stakes are 1% of reference capital per qualifying bet. Published Kelly results use 0.18 of full Kelly, capped at 5% per bet, sized against the initial reference capital. No compounding. A 5% stake cap is not a 5-point maximum edge. These are existing fixed evaluation rules, not settings optimized on these results.
ROI means net P&L divided by total stake; return on reference capital uses the reference amount instead. Pending matches are outside this settled comparison. Profitability, cash availability and realistic drawdown are not established by this view.
This reference ignores reserved cash and compounding. The cash-aware charts above instead include pending opportunities and jointly scale each prediction run to available funds. Exact sportsbook payout times are not recorded, so those charts use the explicitly labeled delayed-release scenario—not reconstructed real account transactions.
Recorded wagers · not candidate simulations
These wagers retain the models that originally placed them, including V1.X.X. They do not describe the selected candidate version. P&L includes recorded accounting outcomes; not every wager qualifies as GOLD model evidence. All pending stakes—including overdue ones—remain reserved.
Realized ROI = settled net P&L ÷ decided stake. It is not the percentage change in your starting bankroll. Historical wagers are never replaced with later candidate predictions.
| Placed | Match / side | Price | Stake | Edge | Status | P&L |
|---|
Settled history
Original operational predictions, with their recorded versions. Probabilities are for the first player listed. Candidate results remain in Performance; missing historical candidate rows are not reconstructed here.
| Match date | Event | Match | Winner | Score | XGB · P1 | NN · P1 | Market fair · P1 | Market path |
|---|
Manifest-pinned registry view
What is promoted, what is only being measured, and what changed across releases.
| Model | Version | Lifecycle | Features | Probability | Released | Artifact | Notes |
|---|
Recorded releases not shown above. Absence from the accepted run does not, by itself, prove a model was withdrawn.
| Model | Version | Lifecycle | Features | Probability | Released | Artifact | Notes |
|---|
Audit surface
Status, errors, and stage counts remain visible even when a run produced no predictions.
| Run | Started | Status | Odds | Features | Predictions | Bets | Settled | Capital / exposure | Post-run stages | Error |
|---|