THE NUMBERS · FUND METRICS

The record, with the intervals attached.

Closing-line value, the flat-stake return stream, and calibration — recomputed from the journals every cycle, with the sample size and a confidence interval on every figure. Below thirty of a type, a number is labeled forming — shown, never concluded. The rules are in The Standard.

CLOSING LINE VALUE

Did our price beat the final market price?

CLV = our published price vs the price at lock, in points. Consistently positive means the model knows something the market doesn't — yet. The market moving toward our side is positive.

OVERALLFORMING · n<30
no CLV rows yet from any sport
GOLFFORMING · n<30
no frozen board yet — CLV opens at the first R1-lock freeze (the market moving toward our side is positive)
MLBFORMING · n<30
no qualifying edges yet — the open rides each day file as first_market (first write wins), so the curve starts with the first qualifying edge. Close = the day file’s last retained market snapshot, an approximate close (the board can rewrite after first pitch)
THE RETURN STREAM

One unit per pick. Every pick graded.

Flat-stake units, wins and misses at full size. Sharpe is per bet — average return per unit of bet-to-bet swing — deliberately not annualized at this sample.

COMBINEDFORMING · n<30
no qualifying picks yet — the record starts with the first qualifying edge
GOLFFORMING · n<30
no graded golf picks yet — the first journaled event grades after it completes
MLBFORMING · n<30
no qualifying picks yet — the record starts with the first qualifying edge
CALIBRATION

Our 60%s should happen ~60% of the time.

Predicted probability vs observed frequency across every graded game — picks and non-picks alike. Points on the diagonal mean the probabilities mean what they say. Dot size is the bucket count.

RELIABILITY CURVEn=272 GRADED
PREDICTED PROBABILITYOBSERVED
● OUR PUBLISHED NUMBER● THE MARKET, SAME GAMES
every graded probabilistic row, not just picks: MLB published blend on market-priced games + graded golf picks at their model probability. The market curve uses the de-vig market probability on the same rows.
PER-BUCKET COUNTS · OUR NUMBER vs THE MARKET
BUCKETnOUR PREDOBSERVEDMARKET PREDOBSERVED
35–45%2942.3%41.4%41.5%43.3%
45–55%15250.8%46.7%50.7%43.9%
55–65%8158.4%61.7%58.3%67.1%
65–75%1068.6%50.0%68.2%54.5%
Sources: 272 market-priced MLB games at the published blend, 0 graded golf picks at their model probability.
SIGNIFICANCE

Small samples, said out loud.

Every metric above carries its n. Under 30 of a type, the figure renders muted with a forming label — a number with an interval, never a conclusion. This week the pick record is thin by design: the thresholds are pre-registered and the record starts with the first qualifying edge. The calibration base is the deeper sample — every graded game counts there, not just the ones we called.

CLV
CLV per pick, in percentage points, one convention across sports: the market moving toward our published side is positive (we beat the close). Golf = edge at first journal − edge at R1-lock freeze (edge = published model probability − de-vig market consensus; with a fixed model that equals consensus at freeze − consensus at first journal). MLB = de-vig market probability of our side at the close − at the open, where the open is the game’s first_market (the first market numbers written into the day file, first-write-wins) and the close is the file’s last retained market snapshot — an approximate close, as the board can rewrite after first pitch. Qualifying picks are re-derived from the journaled numbers via the pre-registered floors (dfs-diff/lib/edge.mjs: ≥3pp golf outright / ≥5pp golf placement, real two-sided markets only; ≥55% published side confidence AND ≥5pp for MLB sides) — stored flags are never trusted.
UNITS
Flat 1u per pick. Golf settles at the journaled best available price on the frozen board (joined back from the graded row by event+market+player — the graded row carries no price). MLB settles at the de-vig market-implied fair price of the open (first_market); the board journals no ticket price, and qualifying edges whose day file predates first_market settle at the close instead.
ROI INTERVAL
95% bootstrap interval on ROI per unit staked: 10,000 resamples of the per-bet units vector, seeded RNG (mulberry32, seed 42) so identical inputs reproduce identical intervals.
PCT POSITIVE INTERVAL
Wilson score interval (z = 1.96) on the share of picks with positive CLV.
SHARPE
Per-bet Sharpe = mean ÷ sample sd of flat-stake per-bet units. Deliberately not annualized — at this sample an annualization assumption would fabricate precision.
CALIBRATION
Reliability buckets over every graded probabilistic row: the published blend vs observed frequency, and the de-vig market curve on the same rows (MLB market probs + golf frozen consensus). Fixed bucket edges; every bucket carries its count.
MLB PICKS
MLB qualifying edges are re-derived from the journaled board (both legs: ≥55% published side confidence AND ≥5pp edge over book-implied). Totals carry no pick flags in v1, so the MLB record is sided only.
FORMING
Any metric with n < 30 ships forming:true and renders muted — a number with an interval, never a conclusion.
THE RECORD

Every graded pick. Misses at full size.

JOURNALED PRE-LOCK · NOT YET GRADED · 10
Xander Schauffele top 20 · +8.7 pts · +110J.J. Spaun top 20 · +7.7 pts · +170Patrick Cantlay top 20 · +7.4 pts · +155Ben Griffin top 20 · +7.2 pts · +170Si Woo Kim top 20 · +6.9 pts · +125Xander Schauffele top 10 · +6.2 pts · +230Hideki Matsuyama top 20 · +6.2 pts · +175Ryan Gerard top 20 · +6.0 pts · +200Nicolai Hojgaard top 20 · +5.3 pts · +225Keegan Bradley top 20 · +5.1 pts · +275
Rocket Classic · journaled Jul 28 · board freezes at R1 lock Jul 30 — these grade when the event completes, win or lose.
SORT4 GRADED
Colombia 70% · +7pp
Colombia v Ghana
—
SOCJul 3HIT
OUR PRICE—
CLOSE—
CLV—
RESULTFT 1-0
Croatia 59% · +5pp
Croatia v Ghana
—
SOCJun 27HIT
OUR PRICE—
CLOSE—
CLV—
RESULTFT 2-1
Canada 61% · +8pp
Canada v Bosnia and Herzegovina
—
SOCJun 12MISS
OUR PRICE—
CLOSE—
CLV—
RESULTFT 1-1
Mexico 75% · +7pp
Mexico v South Africa
—
SOCJun 11HIT
OUR PRICE—
CLOSE—
CLV—
RESULTFT 2-0
archived soccer rows are shown in the record for completeness — no price was journaled, so they carry no CLV or units and stay out of every aggregate.