The Lab · open research record

We kill our own angles in public.

Most systems are noise. We kill ours in the open to prove the ones we keep are real. Edge Labs backtests every angle against an efficient market · most don't beat it, and we retire those here for everyone to see. This is the record: dozens tested so the few that survive are worth your attention. A small edge is real, and you keep it by paying a better price. Killing the noise is what makes the signal trustworthy.

How we test

Every angle is backtested on a strict out-of-sample split · we fit on older seasons and measure on a held-out season the model never saw. A frequency that only shows up in-sample is noise, and we label it as such.

A real edge is small, rare, and perishable, so most of what we test does not survive. That is the point. An angle only becomes a staked, tracked pick after it clears a hard gate. Until then it lives on the slate as descriptive context · a real historical frequency with its sample size, never a win probability.

200+ out-of-sample bets

positive median CLV

holds across seasons

Angles tested

Killed out-of-sample

Survived · validated

Still in testing

Models deprecated

These are counts of research, not bets. Roughly 2% of the angles we have tried have cleared the out-of-sample gate · the rest are public in the graveyard below. A signal that reaches the board has passed through this filter · that is what it means.

For the live, angle-by-angle version of this same ledger (every angle, its real sample size, and its closing-line value, re-graded nightly), see Angle Accuracy.

Validated A survivor of the filter

In FBS non-conference regular-season games (weeks 1-2, 2015-2025 excluding 2020), when the preseason SP+ rating differential (home_sp - away_sp + 2.5 HFA) diverges from the Bovada closing spread by >= 7 points in direction favoring one team, that team covers the closing no-vig spread at >= 62% of qualifying games (n >= 500 in the full 2015-2025 window), because preseason SP+ ratings systematically incorporate team-quality signals that early-season betting lines underweight due to narrative anchoring and public-sentiment pricing.

Mechanism confirmed via data analysis (run_id=2063, note_117): (1) Divergence filter adds massive signal beyond naive strategy — baseline "bet SP+ better team" = 54.3% ATS (n=893); at divergence >= 7 = 73.4% ATS (n=369). (2) Signal is NOT a P4-vs-G5 artifact: G5-vs-G5 = 81.6% (n=…

2 survivors out of 90 tried is not a thin result · it's the honest one. Validated angles keep getting re-tested as new seasons land, and any that stop holding go straight back to the graveyard, in public.

The number we keep · closing-line value

The forward metric we build around is closing-line value (CLV): did the market move toward a number after we flagged it? CLV is the honest forward measure because it proves an edge long before a win rate possibly could · a bettor who consistently beats the closing line is, by definition, finding value the market hadn't priced yet.

CLV runs alongside the win-loss record, it does not replace it. We stand behind our strongest convictions publicly (the running record is at /record, wins and losses both), and we pair it with CLV because CLV can't be faked and it shows up fast, long before a win rate settles. Members get CLV applied to their own bets: the Vault grades every logged bet against the de-vigged close.

We will never post a number we haven't earned. The public conviction record and every forward log appear graded at the close, misses included.

Case study · from the graveyard

Our best angle hit 56.7% over 598 games. We killed it anyway.

The getaway-day UNDER: discovered June 21, validated June 22, blocked by our own adversarial review June 23, and killed July 1 when 2026 out-of-sample data reversed to 41.4% (n=70). The full autopsy, every number from the research database, is public.

Read the autopsy →

Rejected hypotheses

Rejected CFB N>=80 Jul 29, 2026

FBS regular-season home teams at outdoor venues >= 4,500 feet above sea level (Wyoming War Memorial Stadium 7,220 ft, Air Force Falcon Stadium 6,920 ft, Colorado Folsom Field 5,427 ft, Colorado State Canvas Stadium 5,003 ft, BYU LaVell Edwards Stadium 4,549 ft) cover the closing no-vig ATS spread at >= 56% vs. visiting teams whose home city elevation is below 1,500 feet, 2015-2025 regular season (excl. 2020), minimum n >= 80 qualifying game-sides, because aerobic capacity decreases ~2.4% per 1,000 feet above sea level for unacclimated visitors (Fulco et al. 1998), compressing fourth-quarter competitive output in ways the standard 2.5-3.0 point home-field-advantage adjustment does not capture.

Altitude physiology literature (Levine and Stray-Gundersen 1997, Fulco et al.

Rejected CFB Jul 28, 2026

In FBS non-conference regular-season games (weeks 1-3, 2022-2025, excluding 2020) where (a) the home team is a Power 4 program [ACC, Big 12, Big Ten, SEC] and (b) the closing spread is between -2.5 and -14 (competitive home favorite), the home P4 team covers the no-vig closing spread at >=56%, because early-season home crowd intensity in first competitive non-conference home matchups exceeds the standard 2.5-3.0 point HFA adjustment applied uniformly by bookmakers across all game weeks.

Context: All coaching-change CFB hypotheses are exhausted and rejected (H9 OC/DC turnover, H13/H14 new-HC spread buckets).

Rejected MLB N>=150 Jul 24, 2026

MLB full-game totals close OVER the no-vig closing line at >=54% in outdoor regular-season getaway-day games (final game of a road or home series where at least one team's next scheduled game is at a different venue within 24 hours, excluding domed venues, Oracle Park, and Petco Park), 2021-2025 regular season (n>=150), because managers limit starting pitcher pitch counts on getaway days to protect arms before travel, increasing early-inning bullpen exposure above what ERA/WHIP-anchored totals imply.

Direction reversal of H74 (getaway-day UNDER, rejected 2026-07-11).

Rejected MLB Jul 23, 2026

The OVER hit rate advantage in 2023 outdoor MLB games (vs. 2021-2022 outdoor-game baseline, dome-excluded) is >= 3 percentage points higher in April-May 2023 (weeks 1-7) than in June-September 2023, because sportsbooks explicitly did not pre-adjust closing totals for the 2023 rule changes but stated they would monitor and adjust within 3-4 days to 2-3 weeks of regular-season data — meaning any underpricing of the shift-ban and pitch-clock scoring increase was concentrated in the opening weeks, not distributed uniformly across 2023-2025.

Two-source mechanism: (1) Kennedy-Shaffer (2025, note_149) establishes via DiD that the shift ban causally increased LHB BABIP and OBP by 9 points each.

Rejected MLB Jul 23, 2026

MLB outdoor regular-season full-game OVER hit rate against the no-vig closing total is >= 2 percentage points higher in 2023-2025 (post-infield-shift-ban) than in 2021-2022 (pre-ban), dome-excluded, at p < 0.10, indicating that books underadjusted closing totals to the ~0.1 runs/game scoring increase from the shift ban and creating a measurable regime break in pooled 2021-2025 training windows.

Markes et al. (2024) [note_145] established that the infield shift suppressed left-handed batter expected runs by 0.01211 per PA (IPTW, 95% CI: -0.01770 to -0.00651), wit…

Rejected MLB N=200 Jul 11, 2026

MLB full-game totals close UNDER the no-vig closing line at >=54% in outdoor regular-season getaway-day games (final game of a road series immediately before both clubs travel to their next destination) in 2021-2025, minimum gate n=200.

Note_97 (EL-RF-268A, getaway-day UNDER seed research) documents 56.69% UNDER in outdoor getaway-day games over 2023-2025 (n=598), a +7pp delta over standard outdoor day games (49.74%).

Rejected MLB N>=50 Jul 11, 2026

MLB full-game totals exceed the no-vig closing line at >=54% in regular-season games (2021-2025) where the visiting team's starting pitcher's most recent prior start was played at a stadium located >=3 time zones WEST of the current game's stadium (pitcher traveled eastward to reach the current venue within 48 hours of first pitch), because Song et al. (2017) documented that eastward transmeridian travel impairs pitcher performance by elevating home-run-allowed rates by approximately 0.023 HR/game — an OVER signal — and ERA/WHIP-anchored closing totals do not capture circadian phase-advance disruption on a per-start basis. Minimum feasibility gate: n>=50 qualifying starts.

Song, Severini & Allada (2017) [PNAS 114(6)] analyzed 4,919 MLB games and found the dominant measurable jet lag effect was on HOME RUN RATES ALLOWED by the traveling pitcher, not on batting output.

Rejected MLB N>=200 Jul 7, 2026

MLB full-game totals close UNDER the no-vig closing line at >=54% in regular-season day games (first_pitch_utc < 18:00 UTC) where the away team's most recent prior game started at >=22:00 UTC on the immediately preceding calendar day, 2023-2025 effective window (n >= 200 gate).

Internal data: 57.12% UNDER rate ex-push, n=555, z=3.36, consistent across 2023 (56.48%), 2024 (53.98%), and 2025 (60.75%).

Rejected CFB N=200 Jul 6, 2026

FBS regular-season road teams on their 2nd or more consecutive road game (both the current and immediately preceding games are designated away contests) cover the closing no-vig spread at <=46% in 2015-2025 regular season (excl. 2020), minimum gate n=200.

Standard CFB spread pricing incorporates a fixed HFA adjustment per game (~3-4 pts) but does not explicitly discount for cumulative scheduling burden across consecutive road weeks.

Rejected MLB N>=80 Jul 5, 2026

MLB full-game totals close above the no-vig closing total in game 2 of day-night doubleheaders (>=4h gap between game 1 and game 2 first pitches, same venue, same two teams, 2021-2025 regular season, n>=80) at >=54%, because extended rest enables full pitching staff availability and lineup adjustments that bookmakers do not fully price into game 2 totals anchored to game 1 starter ERA/WHIP.

Signal isolated from exp_57 (H65 rejection).

Rejected CFB N>=2,500 Jul 5, 2026

FBS teams on 6-day rest (played prior Saturday, current game on Friday night) cover the closing no-vig spread at <= 47% against opponents on standard 7-day rest, FBS regular season 2015-2025, n >= 2,500 qualifying sides.

Data feasibility audit (note_119, run_id=2098) found that the <=5-day rest threshold in H71 yields only n=70 qualifying sides — underpowered.

Rejected CFB N>=150 Jul 5, 2026

In FBS non-conference regular-season games (weeks 1-3, 2015-2025, excluding 2020) where the closing spread is >= 24 points and the underdog has a closing moneyline of >= +600 (implied win probability <= 14.3%): the underdog covers the closing no-vig ATS spread at >= 54% of qualifying games (n >= 150), because cognitive probability overweighting inflates the underdog's moneyline action, prompting bookmakers to shade the ATS spread wider than fair value to balance book — leaving the underdog's ATS price more favorable than the true talent gap warrants.

When the implied moneyline probability falls below ~14% (>= +600), bettors enter the deep-misperception zone of the Snowberg & Wolfers (2010) probability-weighting functi…

Rejected CFB N=2,115 Jul 3, 2026

FBS regular-season post-bye teams cover the closing no-vig spread at ≥54% vs. non-bye opponents in 2015–2025 (n=2,115 games where exactly one team had a bye the prior week), because books anchor on the most recent game result rather than schedule-adjusted team strength, systematically underpricing the preparation and rest advantage the bye team carries into the matchup.

Glickman and Stern (1998) model NFL team strength as a Gaussian random walk with estimated weekly-change SD of approximately 2.2 points, implying that a team coming off a…

Rejected CFB N=300 Jul 3, 2026

In FBS regular-season games (2015-2025) where one team played their prior game <= 5 days before the current game (short-rest team, typically a Thursday-game team returning Saturday) and the opposing team played their prior game >= 7 days before the current game, the short-rest team covers the closing no-vig spread at <= 46% of qualifying games (i.e., the rested opponent covers >= 54%), because CFB closing lines anchor on prior-game results and talent differentials rather than schedule structure, leaving the 1.5-2 day rest disadvantage systematically underpriced. Minimum n=300 qualifying game-sides.

Literature convergence (note_118): (1) Miller & Davidow (2019) document NFL short-rest teams underperforming closing spread by 0.8-1.2 points — equivalent to ATS cover rate below 47% for short-rest sides.

Rejected Jul 2, 2026

SP+ rating differential is under-weighted by Vegas in non-conference games (especially early-season) because books rely on preseason ratings that lag SP+.

SP+ is a strong predictive rating but the public sees it less than power-conference matchups.

Rejected Jul 2, 2026

Programs whose actual CPR (College Production Rating) systematically diverges from their recruiting our composite rating (our composite rating) create persistent Vegas mispricing in their direction of divergence.

CPR measures actual college performance.

Rejected MLB N=598 Jul 1, 2026

MLB full-game totals close under the no-vig closing line at >=54% in outdoor day games (excluding marine-layer venues Oracle Park and Petco Park) where either team's next scheduled game is in a different city within 24 hours ("getaway day"), 2023-2025 regular season (observed 56.69%, n=598 in backtest sub-type EL-exp39).

H53 v2 sub-type segmentation (experiment_id=39, run_id=1407) found that getaway-day games drive the entire outdoor day-game UNDER signal.

Rejected MLB N>=80 Jun 29, 2026

MLB full-game totals close above the no-vig closing total in game 2 of traditional same-day doubleheaders (both games played on the same calendar date at the same venue by the same two teams, 2021-2025 regular season, n>=80) at >=54%, because spot starters and same-day bullpen depletion from game 1 inflate expected run-scoring above what the total line implies when set using seasonal pitcher ERA/WHIP anchors before game 1 resolves.

Three structural factors inflate game 2 run-scoring above line expectations: (1) Spot starters -- replacement-level or stretched-reliever arms not in normal rotation -- a…

Rejected MLB Jun 28, 2026

MLB full-game totals exceed the closing no-vig total at >=54% in regular-season games (2021-2025) where both starting pitchers average >=5.5 innings-per-start in their current season (>=5 starts to qualify), because deep-usage starters are significantly more likely to face the lineup a third time (TTO 3), incurring the ~.040 wOBA / ~0.4 expected-run TTO-3 penalty that ERA-anchored line-setting systematically underestimates.

Lichtman (2013, SABR, note_id=104) documented a cumulative wOBA penalty of ~.040 points by TTO 3 versus TTO 1 across 469,721 PAs (1999-2002).

Rejected MLB N>=60 Jun 27, 2026

MLB full-game totals close under the no-vig closing line at >=54% in games where the visiting team traveled eastward across >=3 time zones within 24 hours of first pitch, 2021-2025 regular season (n>=60), because eastward transmeridian travel imposes a circadian phase-advance penalty on visiting offensive output that is not captured in the ERA/WHIP anchors used to set the closing total.

Recht, Lew, and Schwartz (1995) analyzed 1991-1993 MLB outcomes and found teams that traveled east 3+ time zones won only 37.1% of qualifying games vs.

Rejected CFB Jun 26, 2026

CFB: when the opening total is 60 or higher, the UNDER beats the CLOSING total across multiple seasons; the edge is high-total-specific.

Sourced from claude edge-hunt 2026-06-22 + adversarial filter.

Rejected MLB Jun 26, 2026

MLB K-prop totals (strikeouts over/under) move at least 0.25 strikeouts higher from open to close in games where the starting pitcher ranks in the top third of starters by 30-day K/9 AND the opposing lineup ranks in the top third by 30-day team K%, at a rate exceeding 55% of qualifying games.

Opening K-prop lines likely anchor to pitcher K/9 alone without fully compounding opposing lineup K-tendency.

Rejected MLB Jun 26, 2026

MLB run line: the -1.5 favorite no-vig implied probability drifts up from open to close (price erosion), a one-directional CLV-capture flow play.

Edge-hunt 2026-06-22 + filter.

Rejected MLB Jun 24, 2026

MLB totals: when one book closing total sits >= 0.5 run ABOVE the 8+ book consensus median, betting the UNDER at that stale book captures value (UNDER leg only).

Edge-hunt 2026-06-22 + filter.

Rejected CFB N>=400 Jun 23, 2026

CFB full-game totals close under the no-vig closing total at >=54% in games where the closing spread is >=21 points, 2015-2025 regular season (n>=400), because game totals are set with an implicit garbage-time scoring premium that partially fails to materialize when the leading team's starter is pulled after halftime and both teams' scoring efficiency compresses in the fourth quarter.

Practitioner evidence (Smith/"Mass" 2018-2024, note_81) documents 54-56% under rates in FBS games with spread >=21, rising to ~57-58% at >=28.

Rejected MLB N>=100 Jun 21, 2026

MLB first-inning YRFI bets (yes-run-first-inning) close above the no-vig YRFI line at >=54% in home games where the home team batting positions 1-3 average combined OBP >=.360 (approximately top-20th percentile for a qualifying lineup), 2021-2025 regular season (n>=100), because the documented structural home first-inning scoring advantage (~0.1 additional expected runs, accounting for 58% of total HFA per Smith 2006) is compounded by elite OBP in the guaranteed-to-bat top-3 positions, and NRFI/YRFI lines anchor on aggregate starter ERA rather than home-split first-inning scoring rates driven by lineup-specific OBP.

Two independent findings converge on this angle.

Rejected MLB N=1,797, Jun 21, 2026

MLB full-game totals close under the no-vig closing line at >=54% in day games (first pitch before 5:00pm local park time) at outdoor venues excluding marine-layer parks (Oracle Park, Petco Park), 2023-2025 regular season (n=1,797, observed: 54.36%).

Afternoon sun angles create documented batter visibility problems, particularly with high sun positions in the late afternoon.

Rejected MLB N=64 Jun 19, 2026

MLB full-game totals exceed no-vig closing line at >=54% when wind blows out (Out To CF/RF/LF) at >=15 mph at outdoor parks excluding marine-layer venues (Oracle Park, Petco Park), 2021-2025 regular season.

H51 backtest (experiment_id=35) found full-sample 47.93% over failed the 54% gate, driven by Oracle Park (33.9% over, 36% of sample, n=64).

Rejected MLB N=150 Jun 18, 2026

MLB full-game totals exceed no-vig closing total at >=54% when wind blows out (Out To CF/RF/LF) at >=15 mph at outdoor-only parks (no retractable roofs or domes), 2021-2025 regular season.

Sub-hypothesis of rejected H19 (coarse direction flag across all parks, >=10mph threshold).

Rejected MLB N>=150 Jun 18, 2026

MLB full-game totals exceed the no-vig closing total line at >=54% in games where either starting pitcher's most recent prior start occurred <=4 calendar days before the current game date, in 2021-2025 regular season (n>=150), because oddsmakers anchor the closing total on each pitcher's seasonal ERA and WHIP rather than rest-adjusted performance, and starting pitchers on 4-day rest show a documented ERA premium of +0.2 to +0.4 runs versus 5+ day rest across modern pitch-count management regimes.

Practitioner consensus (MGL/Tom Tango, Baseball Prospectus) documents a measurable ERA penalty for starting pitchers working on 4 days rest versus the standard 5+ days, d…

Rejected CFB N>=100 Jun 17, 2026

FBS regular-season games (2015-2025, n>=100) where a team won their prior game by >=28 points show opening-to-closing spread movement of >=1.5 points in that team's favor in >=58% of qualifying matchups, and backing the opponent at open (before public sentiment inflates the blowout winner's line) yields positive CLV of >=0.5 points vs. the closing spread.

Avery and Chevalier (1999) [note_80/EL-RF080] document that NFL bettors systematically overreact to prior-week blowout wins, pushing opening spreads past fair value before sharp money corrects by game time.

Rejected MLB Jun 17, 2026

MLB full-game totals close under the no-vig closing line at ≥54% in games where either team scored 10+ runs in their prior game, 2020–2025 regular season (n≥150), because bettors and book-setters overreact to extreme prior run-scoring events — anchoring the next-game total above the team's sustainable mean run rate in a pattern analogous to the stock-market overreaction documented by De Bondt & Thaler (1985).

De Bondt & Thaler (1985, note_79) show extreme prior outcomes inflate forward expectations; Tversky & Kahneman (1974, note_78) provide the cognitive mechanism (anchoring with insufficient adjustment).

Rejected CFB Jun 16, 2026

CFB full-game totals close under the no-vig closing total at ≥54% in games where either team's prior game had a final margin ≥28 points, 2015–2025 regular season (n≥200), because book-setters and public bettors anchor the next-game total on the extreme prior outcome, inflating it above the team's sustainable scoring rate.

Stern (1991) established that the spread equals the market''s unbiased expectation of the final margin.

Rejected Jun 16, 2026

Big Ten, MAC, or Mountain West home teams cover ATS at ≥54% vs same-conference road opponents when game-time temperature ≤40°F in regular-season games October–December (2021–2025, 2020 excluded, n≥60 qualifying games pooled), because cold-weather home-field advantage persists in intra-conference play through venue familiarity, weekly practice conditions, and geographic sub-division temperature variance even when both programs are cold-weather-accustomed.

H38 original (hypothesis_id=38) was untestable: non-conference scheduling clusters in September, leaving 0 warm-weather visitors in October–December games.

Rejected MLB N>=100 Jun 15, 2026

First-inning totals close under the no-vig first-inning line at >= 54% in MLB games where a right-handed starting pitcher faces a batting lineup with >= 6 right-handed batters in positions 1-5, 2022-2025 regular season (n >= 100), because first-inning lines anchor on aggregate starter ERA rather than platoon-adjusted ERA and the documented RHB-vs-RHP wOBA penalty (~25-30 points) is large enough to compress first-inning scoring below the implied total.

Platoon effects are among the most replicated findings in sabermetrics: right-handed batters face RHP with roughly 20-30 wOBA points of suppression compared to left-handed batters in the same matchup.

Rejected MLB N>=75 Jun 15, 2026

MLB F5 totals exceed the closing F5 line at >= 54% in starts where the starting pitcher's season-to-date FIP exceeds their ERA by >= 1.0 run AND the pitcher's season-to-date ground-ball rate is less than 45%, 2020-2025 regular season (n >= 75), because extreme ground-ball pitchers legitimately suppress BABIP through contact-type profile and should be excluded from FIP-ERA regression signals — the over edge in H39 is concentrated in fly-ball and neutral-profile starters where the FIP-ERA gap is pure BABIP luck.

McCracken (2001) DIPS theory establishes that BABIP is largely outside pitcher control and reverts to mean (~.300).

Rejected MLB Jun 15, 2026

MLB F5 totals exceed the closing F5 line at ≥54% in starts where the starting pitcher's season-to-date FIP (Fielding Independent Pitching) exceeds their ERA by ≥1.0 run AND the pitcher has thrown ≥40 innings in the season, 2020–2025 regular season (n≥100), because MLB betting lines anchor on ERA as the headline visible metric while FIP — which strips out defense, luck on balls in play, and strand-rate sequencing — better predicts future run-allowing; pitchers outperforming their FIP are statistically likely to allow more runs than ERA-based line-setting anticipates.

Fielding Independent Pitching (FIP) removes defense and sequencing luck from ERA, retaining only strikeouts, walks, HBP, and home runs — the batted-ball outcomes a pitcher directly controls.

Rejected MLB N=199, Jun 14, 2026

MLB full-game totals close under the no-vig closing line at ≥54% in games where either team's bullpen recorded ≥20 outs the prior calendar day, 2021–2025 regular season (n≥150), because prior-day depletion reduces next-game high-leverage arm availability below what aggregate bullpen ERA implies and oddsmakers do not adjust the closing total for day-specific arm unavailability.

Post-hoc reversal of rejected H40 (over direction).

Rejected CFB Jun 14, 2026

CFB totals on home P4 teams ranked AP Top 10 at game time produce under win rates ≥54% vs. closing total in 2021–2025 regular season (n≥100), because books shade totals upward to capture public over-betting on marquee programs and the resulting inflated total is above sharp-priced fair value.

Paul and Weinbach (2007, note_69) documented statistically significant over-bias in CFB totals: books shade closing totals upward in response to heavy public over-betting…

Rejected CFB Jun 13, 2026

Late-season CFB home teams at outdoor Big Ten North, Mountain West, or MAC venues (game-date October 15 – December 15) cover the spread at ≥56% vs. visiting teams from warm-weather conferences (Sun Belt, Conference USA, SEC West non-conference) when game-time temperature ≤40°F, in 2013–2025 regular season (n≥60), because the cold-weather home-field advantage is underpriced when the visitor has no cold-weather practice history.

Borghesi (2007, note_70) found NFL cold-weather home teams covered at approximately 57–60% ATS in games with kickoff temperature ≤40°F, with the effect amplified vs.

Rejected MLB N>=150 Jun 13, 2026

MLB full-game totals exceed the closing line at >= 54% in games where either team's bullpen recorded >= 20 outs the prior calendar day, in 2021-2025 regular season (n >= 150), because prior-day depletion of high-leverage relievers degrades next-game run prevention below what aggregate bullpen ERA implies and oddsmakers do not fully adjust the closing total for day-specific arm availability.

Oddsmakers set game-total lines using each team's season ERA for starters and bullpen.

Rejected MLB N<100 Jun 12, 2026

MLB F5 totals exceed the closing line at ≥54% in games where the starting pitcher's mean pitch velocity (all pitch types combined) in their most recent prior start was ≥0.8 mph below their season-average velocity AND their prior start lasted fewer than 15 outs (fewer than 5 innings pitched), signaling an early removal due to performance degradation, in 2023-2025 regular season (n≥100).

SPORT: MLB | MARKET: F5 totals (over) MECHANISM: A starter who was knocked out early (fewer than 5 innings) AND is showing velocity decline is a double-flagged degradation signal.

Rejected MLB Jun 11, 2026

MLB games where the starting pitcher threw a single dominant pitch type at ≥65% usage rate (e.g., sinker-heavy or sweeper-specialist) produce full-game totals that exceed the closing line at ≥54% in 2023-2025 regular season (n≥200), because opposing lineups decode a predictable pitch mix over 5+ innings and generate incremental run-scoring above what ERA/WHIP predicts.

SPORT: MLB | MARKET: full-game totals (over) MECHANISM: A starting pitcher leaning on one pitch type at ≥65% becomes predictable as an at-bat sequence.

Rejected MLB Jun 11, 2026

MLB F5 totals exceed the closing F5 line at ≥54% in games where the starting pitcher allowed a barrel-contact rate ≥12% (launch_speed ≥ 98 mph AND launch_angle 8-32°, as a share of in-play batted ball events) in their most recent prior start (2023-2025 regular season, n≥150), because barrel rate is a leading indicator of run-scoring exposure that lags behind ERA/WHIP in same-week line-setting.

SPORT: MLB | MARKET: F5 totals (over) MECHANISM: Expected runs per inning is more precisely predicted by hard-contact rate than ERA, because ERA is diluted by BABIP luck.

Rejected MLB Jun 10, 2026

MLB full-game totals exceed the closing line at ≥54% in games where the starting pitcher's mean four-seam fastball (pitch_type IN ('FF','FA')) velocity in their most recent prior start was ≥1.5 mph below their season-average four-seam velocity (2023-2025 regular season, n≥150), because books price ERA/WHIP rather than real-time stuff deterioration.

SPORT: MLB | MARKET: full-game totals (over) MECHANISM: ERA and WHIP are trailing-season stats that smooth over within-season velocity changes.

Rejected MLB Jun 10, 2026

MLB full-game totals go under the closing line at ≥54% in day games (first_pitch_utc before 18:00 UTC) where both the home team and the away team each played a game on the immediately preceding calendar day (identified via self-join on games by team_id and game_date), in 2023-2025 regular season (n≥200), because player fatigue and managerial rest decisions compress offensive output but the closing total line reflects full-roster statistics.

SPORT: MLB | MARKET: full-game totals (under) MECHANISM: Day-after-night games force managers to rest key starters, deploy backup lineups, and skip normal recovery routines.

Rejected CFB Jun 9, 2026

CFB composite power rating differentials (Sagarin + Massey optimal combination) predict opening-to-closing line movement direction: in FBS games where the composite implies a team ≥4 points stronger than the opening spread favors, that team's line moves ≥1 point in their direction by close in ≥55% of cases, generating positive CLV for composite-informed bettors who enter at open.

Fair & Oster (2005, note_58) prove the FINAL CFB closing line subsumes all published computer ranking information (Fair-Shiller F=0.96, insignificant at 5%).

Rejected CFB Jun 9, 2026

CFB spread bets where the bettor's projection falls within 0.5 points of a key number (3 or 7) and the line is set on the favorable side of that key number (e.g., line −2.5 when projection is −2.9, or line −6.5 when projection is −6.8) have cover probabilities that are understated by ≥1.5 percentage points by the standard Φ(differential/σ) formula, because historical CFB score differentials land on exactly 3 at 2.7× and on exactly 7 at 2.1× the rate predicted by a normal distribution.

Sides, Harvill & Sides (2022) arXiv:2212.08116 show that the unconditional probability of a CFB game ending with a 3-point differential is 9.6% (vs 3.6% from normal), and…

Rejected CFB Jun 9, 2026

CFB spreads on G5 vs G5 games are softer than spreads on P4 vs P4 games because books spend less time grading the matchup.

Lower book interest in G5/non-conference matchups means less line movement and slower correction.

Rejected MLB N=2073 Jun 7, 2026

Hitter props (total bases / hits / home runs) — the softest MLB market — contain an exploitable public-data edge.

REJECTED 2026-06-07. Acquired 15,233 real hitter-prop lines (May 1-21 2025, ~5.6k Odds API historical credits) across batter_total_bases/hits/home_runs; Statcast-projecti…

Rejected MLB N>=100, Jun 6, 2026

Combining 2-3 factors (park, team form, favorite status, weather, etc.) produces an out-of-sample MLB betting edge on totals or sides.

REJECTED 2026-06-06. Exhaustive combo search: 22 leak-free binary factors, ~7,172 factor×target combos, train 2023-24 (4,514 games) -> test 2025 (2,429).

Rejected MLB N=48 Jun 6, 2026

Cross-book +EV line-shopping on MLB totals is exploitable as a betting strategy

MLB 2024-2025 LOO +EV test (n=48 qualifying bets, 2025 only): 21W-24L-3P (43.8% win rate, -$373 on $100 flat).

Rejected N=166 Jun 6, 2026

Pregame total line-movement direction predicts over/under result

Tested 2023-2024 train + 2025 OOS on MLB totals.

Rejected N=50 Jun 6, 2026

Home-plate umpire over/under tendency in prior seasons predicts totals result in current season

OOS test 2023-24 train → 2025 test: Spearman r=0.10 across 75 paired umpires (min 30 train games, 10 test games).

Rejected CFB Jun 6, 2026

CFB team-total under bets on teams with implied score ≤10 points (derived from game total and spread) beat their no-vig probability by ≥3 percentage points because books set team-total lines assuming unconstrained normal score distributions, ignoring the zero-floor censoring effect.

Arscott (2023) documented that the joint system of game total + point spread implies team-level scores that are censored at zero.

Rejected CFB Jun 6, 2026

Weather (wind, precipitation) is under-priced into CFB totals more than NFL totals because CFB has more variance in venue-level weather data quality.

NFL has 30+ stadiums with consistent weather feeds; many CFB venues have spottier weather records, especially outside the SEC/Big Ten.

Rejected N=489 Jun 6, 2026

New OC or DC = 3-6 point Vegas-vs-actual penalty for the first half of season, especially against opposing units with continuity.

Coaching turnover effects are documented anecdotally but priced inconsistently.

Rejected Jun 6, 2026

Park run factor combined with weather (wind direction relative to center field, temperature) is under-weighted in F5 totals.

Anecdotally line-makers adjust for weather close to game time but slowly.

Rejected N=376 Jun 6, 2026

Teams with 5+ new transfer starters underperform spread in weeks 1-3 because chemistry lag is not priced in to preseason lines.

Public over-weights elite portal classes ("Colorado 2023" narrative).

Rejected CFB N=278, Jun 6, 2026

New-HC FBS teams priced as 7-14 point underdogs cover at below-market rates (40.3% ATS, n=278, 2013-2019); fading them — i.e., backing their opponents — yields positive EV in this spread bucket.

EL-RF048 (run_id=370) computed spread-bucket ATS cover rates for new-HC FBS teams across 2013-2019.

Rejected CFB N=186, Jun 6, 2026

Fading new-HC FBS teams priced as 7-14 point favorites in regular season games returns 59.4% ATS for the opponent (n=186, 2020-2025 regular season, FBS only).

When a new head coach's team is priced as a significant favorite (7-14 pts), the market systematically overestimates their execution.

Rejected CFB Jun 5, 2026

CFB FBS score differentials have σ ≥ 15.0 points versus the NFL σ = 13.86 established by Stern (1991), causing the standard Φ(spread/13.86) win-probability formula to overestimate favorite cover probability by ≥ 2 percentage points at spreads of 14+ points in FBS non-conference games.

Stern (1991) measured NFL margin standard deviation at σ ≈ 13.86 using maximum likelihood on regular-season game outcomes.

Rejected MLB Jun 5, 2026

MLB starter strikeout props are systematically over-priced for high-K starters because public squares load up on the over.

Longshot-bias style argument applied to props.

Rejected CFB Jun 5, 2026

Vegas under-weights composite roster talent gap in non-marquee CFB matchups. When Team A composite is 0.05+ above Team B AND Vegas implies A as dog, A covers >55% historically.

The market over-weights recent results + brand-name brand recognition.

Rejected N>=300 Jun 5, 2026

F5 moneyline underdogs priced +130 to +220 beat their Shin-corrected implied probability by 2+ percentage points on average in 2023-2025 regular season games, because the favorite-longshot bias documented by Woodland & Woodland (1994) is amplified in the F5 market where sharp liquidity is thinner than full-game moneyline.

Woodland & Woodland (1994) established FLB in MLB moneylines: public bettors systematically over-back favorites, pushing their price shorter than warranted and leaving un…

Rejected MLB N=270 Jun 4, 2026

Stacking 4+ aligned under-signals yields an under edge on MLB totals

First non-dead result.

Rejected Jun 4, 2026

Finding situational/run-line trends live and riding them (momentum) produces an edge

Tested whether recent streaks persist.

Rejected Jun 4, 2026

Situational trend-mining on our own data produces a run-line betting edge

Generated 787 situational angles (streaks, rest, pitcher experience, hot/cold, opponent state, month) on the MLB run line.

Rejected CFB N=551 Jun 4, 2026

CFB situational ATS edges (HFA overvalue, big-favorite fade, bowl underdogs, G5 softness)

Tested home ATS by favorite/dog bucket (all 0.481-0.508), postseason underdog ATS (0.503, n=551), G5-vs-G5 home dogs (0.488, n=1366), 2014-2025.

Rejected CFB Jun 4, 2026

Preseason power ratings (SP+) beat the early-season CFB opening spread

Tested prior-season final SP+ (proxy for preseason) implied spread vs the opening line, weeks 1-5, 2014-2025, graded ATS.

Rejected MLB Jun 4, 2026

MLB umpire strike-zone size produces exploitable unders vs the closing total

Ranked umpires by called-strike rate from statcast (40+ games each).

Rejected MLB Jun 4, 2026

MLB wind direction produces exploitable totals vs the closing line

Over/under rates by wind direction x speed vs closing total show no coherent signal: wind blowing out 10+ mph -> over 0.484 (documented edge absent); wind in inconsistent (5-9 under 0.508, 10+ under 0.449).

Rejected Jun 4, 2026

Pitcher strikeout props can be beaten with a public-data model (softer market than game totals)

Built expected-K model from pitcher K-rate/workload/whiff form + opponent lineup K-rate, Poisson P(over), EV bets at real closing prices, walk-forward 2024->2025 (4,739 OOS bets).

Rejected Jun 4, 2026

F5 totals (first-5-innings over/under) can be beaten with a public-data fundamentals model

Tested a HistGBM on statcast-derived starter form, reconstructed lineup wOBA, team F5 off/def, park, umpire, and wind/weather across 4,779 out-of-sample games (walk-forward 2024-2025).

Rejected May 27, 2026

F5 (first 5 innings) totals are softer than full-game totals because line-makers and sharp action concentrate on the full-game market.

Soft market thesis. F5 lines do not exist on every book.

Deprecated models

Deprecated cfb_h7_talent_gap_baseline · v0.1 · spread

clv_ci_hi: 2.8319clv_ci_lo: 2.3419m0_avg_mae: 12.152m1_avg_mae: 12.186ats_hit_rate: 0.4933

Deprecated cfb_h8_portal_upside_baseline · v0.1 · spread

clv_ci_hi: 4.7444clv_ci_lo: 4.0741ats_hit_rate: 0.4762ats_hr_ci_hi: 0.5635ats_hr_ci_lo: 0.3889

Deprecated cfb_h9_coaching_turnover_baseline · v0.1 · spread

m0_avg_mae: 12.014m1_avg_mae: 12.017population: FBS-vs-FBS completed games with spread, seasons 2021-2025sigma_used: 15.3m0_avg_rmse: 15.229

Deprecated cfb_new_hc_fav714_fade · v1 · ats_spread

clv_n: 64total_n: 385holdout_n: 172training_n: 213breakeven_pct: 52.4

Deprecated cfb_spread_spplus_margin · v2 · spread

clv_ci: 2.938,3.1785ats_hr_ci: 0.4741,0.54m0_avg_mae: 12.016m1_avg_mae: 12.021m2_avg_mae: 12.022

Deprecated f5_totals_hgb · 0.2 · totals_1st_5_innings

n_bets: [object Object]verdict: Deeper features halved the loss; selective bets (>=1 run edge) now ~breakeven but NOT statistically distinguishable from breakeven (52.34% vs 52.38%, SE ~1.2%). Still no demonstrated edge. Direction is working.backtest: walk-forward by season, out-of-sample 2024+2025hit_rate: [object Object]breakeven: 0.5238

Deprecated f5_totals_hgb · 0.1 · totals_1st_5_innings

verdict: NO EDGE. Closing line more accurate than model (MAE). Does not clear phase gate.backtest: walk-forward by season, out-of-sample 2024+2025features: fundamentals only (pitcher form, team F5 off/def, park, ump, weather); market line excludedhit_rate: [object Object]breakeven: 0.5238

Deprecated getaway_day_outdoor_under · v1 · full_game_totals

clv_positive: trueexperiment_id: 41hypothesis_id: 54holdout_window: 2024-2025holdout_n_under: 209

Deprecated h12_f5ml_away_fade · v1 · h2h_1st_5_innings

val_2024: [object Object]test_2025: [object Object]pooled_oos: [object Object]gates_total: 3gates_passed: 0

Deprecated h44_bullpen_fatigue_under · v1 · game_totals

oos_n: 199is_2023: [object Object]is_2024: [object Object]oos_2025: [object Object]gate_n_100: PASS

Deprecated h53_day_game_under_rule · v1 · full_game_totals

by_season: [object Object]gate_passed: truetrain_n_over: 485train_n_under: 570train_seasons: 2023,2024

Deprecated mlb_f5_totals_baseline · v0.1 · totals_f5

fold_stats: [object Object],[object Object]population: MLB regular-season (game_type=R) graded F5 games, 2023-2025final_model: [object Object]n_population: 7287walk_forward: expanding window by season; train<S predict S for S in [2024,2025]; 2023 train-only

Deprecated mlb_f5_totals_market_anchored · v0.2 · totals_f5

RUN_STATUS: DEFINITIVE / FULL BACKFILL (Phase-1 gate run)fold_stats: [object Object],[object Object],[object Object]population: MLB regular-season (game_type=R) graded F5 games, statcast era (>=2023)final_model: [object Object]n_population: 7330

Deprecated mlb_fg_totals_market_anchored · v1 · totals_fullgame

RUN_STATUS: DEFINITIVE / FULL BACKFILL (Phase-1 control measurement)fold_stats: [object Object],[object Object],[object Object]model_role: H1 CONTROL (full-game arm) vs F5 model_id=7 / experiment_id=5population: MLB regular-season (game_type=R) graded full-game games, statcast era (>=2023)final_model: [object Object]

Deprecated mlb_kprops_xk · v1 · pitcher_strikeouts

n_bets: 0odds_note: Zero pitcher_strikeouts rows in odds_snapshots. CLV path built and ready; activate post Robby approval of prop-line backfill.RUN_STATUS: DEFINITIVE / FULL ODDS RUN (Phase-1 gate)fold_stats: [object Object],[object Object]population: MLB regular-season (game_type=R) starter-games, statcast era (>=2023)

Deprecated mlb_outdoor_getaway_under · v1 · totals

note: Situational filter, no ML training. Constant prediction = observed under rate. Model LL beats no-vig baseline LL by 0.0051.n_push: 67z_stat: 3.908n_decided: 1533training_n: 1533

Deprecated mlb_run_line_market_anchored · v1 · run_line

clv: [object Object]RUN_STATUS: DEFINITIVE / FULL BACKFILL (Phase-1 gate run)fold_stats: [object Object],[object Object],[object Object]population: MLB regular-season (game_type=R) graded full-game rows, statcast era (>=2023)final_model: [object Object]

Deprecated pitcher_k_props_hgb · 0.1 · pitcher_strikeouts

verdict: NO EDGE, worse than F5 totals. Model less accurate than the closing line; ROI degrades as EV threshold rises (value signal anti-predictive). Root cause: no informational edge over a sharp market.backtest: walk-forward train 2024 -> test 2025, 4,739 OOS betsmae_model: 1.879mae_market_line: 1.76roi_by_ev_threshold: [object Object]