Model accuracy + full history, in public

Our probabilities, measured in public.

We don't sell a betting edge — we sell accurate analysis. So we measure what actually matters for a decision-support tool: how well our fair-value probabilities match real outcomes (calibration), shown below in full. Our earlier +EV-pick experiment is kept underneath as a labeled archive — including the honest finding that the playable market is too thin to beat. Nothing hidden, nothing cherry-picked.

Sharp-anchor calibration

Are the sharp-anchor fair prices accurate?

✓ Beats the base-rate baseline (significant)

Predicted (avg)

45.1%

Actual win rate

41.9%

95% CI 37.9%–45.9%

Brier score

0.215

lower is better

Base-rate baseline

0.243

beat this = informative

Across 583 settled win/loss predictions (counted once per match-market, pushes & voids excluded — so this is smaller than the total settled-pick count) the de-vigged Pinnacle (sharp-anchor) fair value said 45.1% on average and 41.9% actually happened — the sharp-anchor probabilities track real outcomes, and its Brier score beats the base-rate baseline beyond statistical noise — it adds real information. This measures accuracy, not profit: we're a decision-support tool, not a tipster.

PredictedPredicted vs actualn

0.0%–10.0%

pred 7.1%

actual 12.5%

10.0%–20.0%

pred 15.5%

actual 16.1%

20.0%–30.0%

pred 25.3%

actual 20.9%

30.0%–40.0%

pred 34.6%

actual 27.9%

40.0%–50.0%

pred 45.1%

actual 41.8%

50.0%–60.0%

pred 53.6%

actual 49.5%

200

60.0%–70.0%

pred 64.6%

actual 66.0%

70.0%–80.0%

pred 74.7%

actual 73.9%

80.0%–90.0%

pred 84.3%

actual 87.5%

90.0%–100.0%

pred 92.5%

actual 75.0%

Research-phase archive (retired) — our earlier +EV-pick experiment

Below is the full result history from when we tested publishing +EV betting picks. We've since found the playable market for our users (~2 GGL-licensed sportsbooks) is too thin to sustain a betting edge, so we repositioned to analytics. We keep every result here — wins, losses, and the finding itself — because deleting losers is what tipsters do. It is NOT how we measure the tool (see model accuracy above), and it is reported as independent decisions, not per-book rows.

n=643· early-stage data

Equity curve · All time

-12.24u cumulative

643 settled · May 17 → Jun 21

Each point is a settled pick's contribution to cumulative net profit (stake units). Wins lift the line, losses drop it. No deletions, no smoothing, no cherry-picking.

Last updated: 2026-06-21 20:01:26 UTC

n=643· early-stage dataEvery settled result published — wins and losses

Track record builds with each settled pick — current depth: n=643.

Lock-in significance threshold: 1500 picks. We publish every result regardless of sample size — the receipts are the product.

Returns · live from the ledger

n=643· early-stage data

7-day ROI

−13.7%

218 settled picks

30-day ROI

−5.1%

560 settled picks

All-time ROI

−1.6%

643 settled picks

Longest win run

5 settled picks

Longest lose run

8 settled picks

Units staked

757

Profit (u)

−12.24u

Hit rate

41.9%

Avg odds

2.98

Avg CLV vs close

−7.6%

our honest edge metric · n=549 · 9% beat close

Per-sport models

ROI by sport — every model in public.

Each row is the cumulative ROI of every settled pick for that sport. No cherry-picking — losing models are visible too.

Football (Soccer)

n=305 · 35.2% win

+6.9%

Baseball

n=195 · 44.9% win

−10.5%

Basketball

n=134 · 51.2% win

−10.1%

Ice Hockey

n=9 · 44.4% win

−18.6%

Archive note: rows are logged per (match × selection × bookmaker), so the same decision often appears at several books. The headline stats above count 643 independent decisions; this table lists all 1753 per-book rows in full (most of these book-prices were never playable for a DE user in the first place). Kept complete for transparency — we count decisions, not rows.

Date	Match	Pick	Odds	Stake	Result
2026-05-17	Inter Milan vs Hellas Verona Football (Soccer)	away Smarkets	12.50	3u	lost-3.00u
2026-05-17	Inter Milan vs Hellas Verona Football (Soccer)	away Matchbook	13.00	3u	lost-3.00u
2026-05-17	Inter Milan vs Hellas Verona Football (Soccer)	away Betsson	12.50	3u	lost-3.00u
2026-05-17	Inter Milan vs Hellas Verona Football (Soccer)	away Tipico	12.00	3u	lost-3.00u
2026-05-17	Inter Milan vs Hellas Verona Football (Soccer)	away Betfair	13.00	3u	lost-3.00u
2026-05-17	Pisa vs Napoli Football (Soccer)	home Tipico	8.50	1u	lost-1.00u
2026-05-17	Como vs Parma Football (Soccer)	away Tipico	12.00	3u	lost-3.00u
2026-05-17	Como vs Parma Football (Soccer)	draw Tipico	6.50	2u	lost-2.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	draw Smarkets	4.50	1u	lost-1.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	draw Matchbook	4.50	1u	lost-1.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	draw Betfair	4.50	1u	lost-1.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	away Marathonbet	5.10	1u	lost-1.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	draw Marathonbet	4.50	1u	lost-1.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	away FanDuel	5.20	2u	lost-2.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	away Betfair	5.60	3u	lost-3.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	away Smarkets	5.60	3u	lost-3.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	away Matchbook	5.70	3u	lost-3.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	away NordicBet	5.20	2u	lost-2.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	away Betsson	5.20	2u	lost-2.00u
2026-05-17	Manchester United vs Nottingham Forest Football (Soccer)	away Tipico	5.20	2u	lost-2.00u

Showing 1734–1753 of 1753 book-rows (newest first)

NewerOlder

Methodology

How we measure accuracy (calibration + CLV)

Our headline metric is calibration: across our settled predictions, does an outcome we put at probability p actually happen about p of the time? We publish the calibration curve (predicted vs actual) and a Brier score tested against the base-rate baseline, with a statistical-significance gate. It measures accuracy, not profit.

We also show Closing Line Value (CLV) honestly — how our published odds compared to the market's de-vigged close, including when it is negative. Formula: CLV% = (published_odds / closing_odds − 1) × 100

CLV is a transparency receipt from our retired +EV experiment, not a current edge claim: the playable market proved too thin to beat the closing line, so we dropped the profit framing and kept the numbers visible anyway.

Settlement rules

Each pick is settled against one of four outcomes: Won, Lost, Push, or Void.

Won: Selection correct. Profit = (odds − 1) × stake.
Lost: Selection incorrect. Profit = −stake (full loss).
Push: Bet cancelled by bookmaker (e.g. Asian handicap half). Stake refunded. Profit = 0 units.
Void: Match abandoned or selection suspended. Stake refunded. Profit = 0 units.

The pick_results table is database-enforced append-only — UPDATE and DELETE are blocked at the database level, so a settled result cannot be silently amended or removed after the fact.

All profit/loss figures use the tax-neutral formula. If a jurisdiction-specific adjustment applies it appears in the pick's audit page under “Edge (post-tax).”

Sample-size caveat

Statistical significance in sports betting requires a large sample. The conventional threshold for drawing firm conclusions is ~1,500 settled picks — at that depth, a consistently positive ROI is extremely unlikely to be variance alone.

BetEdge launched in May 2026 with a baseline of ~293 free picks. Every result is published regardless of sample depth — the receipts are the product. The n= · early-stage data chip on the dashboard is the honest signal: “real numbers, but interpret ROI with an appropriate confidence interval.”

Lock-in threshold: 1,500 picks. The chip disappears once that milestone is crossed. We do not cherry-pick which results to show.

FAQ

About the track record

How is ROI calculated?

ROI is cumulative profit divided by total stake, expressed as a percentage. Only settled picks are included. Example: if 100 units are staked and 8 units net profit results, ROI is 8%. We report both positive and negative ROI — the ledger is public regardless.

What is CLV?

Closing-Line Value (CLV) is the difference between the odds at which we placed the bet and the closing line at kickoff. Positive CLV means we secured a price the market later agreed was underpriced. It is the sharpest long-term indicator of whether an edge-finding method actually works.

Why does the chart show negative returns sometimes?

Because they happened. We publish every settled pick — winners and losers alike. Variance is real and hiding losing stretches would be dishonest. Full transparency means you see everything, including the drawdowns.

What counts as a settled pick?

A pick where the match has finished and the result was determined by the governing body. Pushes (draws on spread markets) and voids (cancelled matches, walkover results) are excluded from ROI calculations but remain visible in the ledger for completeness.

How do I know these numbers aren't cherry-picked?

Every settled pick is published here with its odds, result, CLV, and profit — winners and losers alike. Results are stored append-only: once a match settles, the row cannot be edited or deleted. We never remove losing picks or hide drawdowns. The complete ledger is on this page exactly as recorded.

BetEdge is an analytics and decision-support tool — not a bookmaker and not a tipster service. We don't accept bets or hold funds. For educational and informational purposes only. 18+.