Yesterday’s column led with PIT +21.3% AGREE — the largest morning edge in column history — with Paul Skenes on the mound. The framing was explicit: “Either I’m right and this game is over, or I’m broken on aces. No middle-ground outcome — this game is signal.”

The result was signal. LAD won 12-3. Skenes gave up 8 earned runs across his start. The column-record edge lost by 9 runs. 🩸

That’s the lead. We don’t bury bad outcomes here. The Skenes back was framed as the cleanest possible test of whether the model’s elite-pitcher composite was actually reading aces correctly. The answer the game produced: not on this one.

One game doesn’t invalidate the model. A +21.3% morning edge is built on a 71.9% model rate, which means the model gave Skenes a 28.1% chance of losing — and the upside scenario (LAD scoring 12) lives within that band. But a column-record edge losing by 9 runs is the kind of data point the calibration loop on /stats is built to absorb. The (AGREE × LOCK) cell will reflect this. The shrinkage factor will update tonight.

Subscribe now

📊 But Yesterday Was Still a Massive Day: 11-4 (73.3%)

Five of six locks cashed. The Skenes loss was the outlier. 🎯

Locks: 5-1. Strong: 3-1. The slate as a whole was massively profitable — even with the column-record edge melting. That’s exactly the point of running a portfolio of edges instead of single-game conviction plays. ⚖️

Leave a comment

🔬 The CWS Lineup-Aware Find — Day 2 Of The Infrastructure Paying For Itself

Yesterday’s column had ATL @ CWS as a solid AGREE play at +3.7%. That’s what the morning team-level wOBA produced.

By close (after the 5 PM ET lineup-aware re-run), it was CWS +13.6% FLIP LOCK. The edge direction flipped entirely and the magnitude tripled. CWS cashed 6-5.

This is the second consecutive day the lineup-aware re-run found a LOCK the morning slate didn’t have. Monday it was SEA @ BAL +20.6%. Tuesday it was ATL @ CWS +13.6% (FLIP, direction-changing). Both cashed. 📡

The two cases are slightly different:

• Monday SEA @ BAL — the morning team-level wOBA had a smaller edge that grew once lineup-specific wOBA filled in. Same direction, bigger magnitude.

• Tuesday ATL @ CWS — the morning team-level wOBA had ATL favored. The lineup-aware version flipped direction to CWS and made it a lock. The lineup-aware feature didn’t just refine the edge — it inverted the conclusion.

The second case is the more interesting validation. Lineup-specific wOBA with L/R splits and lineup-position weighting can produce a directionally different read than team-level wOBA. That’s the entire point of the 5 PM ET re-run shipping. Two days, two locks. Both cashed. The infrastructure earned its keep. 🛠️

🌟 The Ace Narrative, Reckoned With

Yesterday’s slate had four ace plays. Three worked, one (Skenes) blew up.

Ace-BACK plays (the new archetype, last 4 days):

• 🟢 deGrom +7.9% AGREE → 10-0 shutout

• 🟢 Buehler +4.0% AGREE → SD 6-2 win

• 🔴 Skenes +21.3% AGREE → LAD 12, PIT 3 (loss)

• 🟢 Cole +6.9% AGREE → NYY 3-2 win

Ace-FADE plays (yesterday’s two):

• 🟢 Gallen FADE → MIA +13.5% LOCK cashed 10-6 (Gallen gave up 10 R)

• 🟢 Wheeler FADE → TOR +7.3% cashed 3-2

The combined record across last week’s ace involvement: 5-1. The five wins came from a mix of backs (deGrom, Buehler, Cole) and fades (Gallen, Wheeler). The one loss was Skenes — which happened to be the biggest edge.

The clean read: the model has gotten meaningfully better at calling ace games. The historical 1-7 ace-fade streak doesn’t represent the current state. What it does represent is training data the calibration loop has absorbed — segments that previously underperformed are now being shrunk on slates where they appear. The Skenes outcome will become another data point in that loop.

What this doesn’t mean: that the column should “stop covering aces” or do anything reactive. What it does mean: that the model’s magnitude-confidence isn’t perfectly calibrated yet. A +21.3% morning edge implies near-certainty. The actual outcome distribution clearly has more variance than that. The methodology page note about “edges are model output, not promises” applies here as much as anywhere. 🧘

Subscribe now

🆕 Bad Bets Just Got A Lot Bigger (Site Updates)

I poked around the site this morning and a lot has shipped since Monday’s launch. This is no longer a single-purpose MLB column platform — it’s becoming a real sports hub:

🏈 NFL section — /nfl, /nfl-today, /nfl-live, /nfl-history, /nfl-stats. NFL picks coming at Week 1 kickoff. Fantasy leagues open now — you can start a league with friends before the season. No DFS, no upsells, no ads. Just classic season-long fantasy. That positioning matters — fantasy football is one of the few “betting-adjacent” verticals that’s genuinely social and free.

🎯 MLB Player Props v0.1 — /props. Statcast-driven Poisson model for HR / Hits / K’s, OVER + UNDER, on DraftKings closing lines. Refreshes every 30 minutes. Tier thresholds: Locks ≥12%, Strong ≥7%, Solid ≥4% — tighter than ML edges because props markets are tighter. And critically: position sizing is built in. 1 unit = 1% bankroll, 0.25× Kelly cap. That’s a real bankroll-managed model, not a “here’s a play” content product.

🔬 Model Health page — /model-health. Live diagnostics for the props model: what’s working, what’s broken, daily hit-rate trend, Market × Side breakdown (”If Hits OVER hits 55%, lean into it. If HR OVER hits 20%, kill it.”), calibration plot, by-park breakdown. The same calibration philosophy as /stats but for props. Auto-refreshes off the Supabase player_props_graded table. Built to be useful in under 60 seconds.

🌐 Hub homepage — live scoreboard, trending picks 24h, news feed (All / MLB / NFL filterable), community chatter, “The Receipts MLB 2026” ledger preview, today’s top edges, recent results. One landing page that surfaces everything.

📋 MLB props ledger — /props-history. Track record on every prop pick the model has ever made.

This is what a “transparent quant platform” actually looks like. Receipts on every claim, model-health diagnostics that anyone can audit, a real position-sizing framework, and now a second sport coming online. The “what is Bad Bets” answer is no longer “a daily MLB column on Substack.” It’s the column + the model + the live infrastructure + the receipts + the fantasy social layer. 🌐

Subscribe now

📈 Running Ledger Through 17 Days — Best 3-Day Stretch In Column History

Three-day stretch: 31-7 (81.6%). That’s the best three-day run in column history. 211 picks at 61.6% — comfortably above breakeven, comfortably above the 60% line, trajectory restored.

Full ledger queryable at bad-bets.netlify.app/history. Live (direction × bucket) calibration grid at /stats. 📊

🎯 Today’s Slate — 3 Locks And Ohtani

Headline numbers:

• ⚾ 13 games — slightly lighter Wednesday card

• 🔒 3 LOCKS (≥10%) — MIA, TB, OAK FLIP

• 🎯 9 actionable (≥4%) — 3 locks + 3 strong + 3 solid

• 🌟 ★ Ohtani backed for LAD at +8.2% — the next “back the ace” test after Skenes

• 🏔️ Coors slugfest: CHC @ COL projects 11.67 r at PF 112 — PASS (only +0.8%)

Subscribe now

🔒 Expanded Lock Breakdowns

🥇 #1 · MIA +17.7% AGREE — TOP LOCK, ARI/MIA DAY 2

Model: MIA 67.7% · Market: MIA 50.0% · Total est: 7.72 r · PF: 96 · WX: DOME Matchup: Ryne Nelson (ARI) vs Eury Pérez Gusto (MIA)

Back-to-back lock days for MIA against ARI. Yesterday MIA +13.5% LOCK (Gallen FADE) cashed 10-6 — Gallen got torched. Today the model goes even harder: +17.7% AGREE, the day’s top edge. Market still has it as a 50/50 coin flip; model has MIA at 67.7%. The dome environment + neutral park keeps weather noise out of the equation.

The interesting wrinkle: MIA is now in the column for two straight days with the model pushing the edge UP after yesterday’s lock cashed. Same archetype as the OAK 4-day arc — the model isn’t anchored, it’s recomputing each day independently. Series-day analytics in action. 🔄

🥈 #2 · TB +16.1% AGREE — 74.8% MODEL CONVICTION

Model: TB 74.8% · Market: TB 58.6% · Total est: 9.27 r · PF: 96 · WX: DOME Matchup: Connor Bennett (BOS) vs Drew Rasmussen (TB)

The highest single-team model rate of the slate at 74.8%. TB is already a 58.6% market favorite at home in the dome, but the model thinks they’re a ¾-conviction win. Rasmussen has been a steady mid-rotation arm grading well in the composite; Bennett is a back-end starter the model probably reads as below replacement.

The 9.27 r projection means the model isn’t pricing this as a pitching gap so much as a roster gap — TB’s offense vs Bennett, BOS’s offense vs Rasmussen. The size of the edge is what TB-as-a-team is worth at home against a fringe BOS rotation arm. 🎯

🥉 #3 · OAK +10.1% FLIP — MIL/OAK SERIES GAME 3

Model: OAK 60.0% · Market: MIL 50.1% · Total est: 8.49 r · PF: 91 · WX: 82°F Matchup: James Sproat (MIL) vs Blake Perkins (OAK)

Game 3 of the MIL @ OAK series. The model has now flipped 180 degrees over three days.

• Mon (G1): MIL +11.9% LOCK AGREE → MIL 15, OAK 14 (MIL covered in absurd 29-run game)

• Tue (G2): OAK +1.7% AGREE → OAK 7, MIL 5 (OAK covered)

• Wed (G3): OAK +10.1% LOCK FLIP → TBD

This is the recompute-from-scratch thesis demonstrated as cleanly as it gets. Three games in three days, same two teams, three completely different reads. The model is not anchored to yesterday — or to Monday’s lock cashing. Today is its own evaluation: Sproat (back-end MIL arm) vs Perkins (mediocre OAK arm but at home, in a cooler dome-adjacent environment).

FLIP-LOCK is historically the column’s most volatile bucket. The calibration loop should already be applying shrinkage if /stats has it underperforming — meaning today’s +10.1% number is post-calibration. 🛡️

🌟 #4 · LAD +8.2% AGREE — OHTANI BACKED · BACK-THE-ACE TEST #5

Model: LAD 73.2% · Market: LAD 65.0% · Total est: 9.18 r · PF: 98 · WX: 91°F Matchup: Shohei Ohtani (LAD) vs Joey Jones (PIT)

This is the column’s headline story today. One day after Skenes was backed at the column-record edge and LOST 12-3, the model is once again backing an ace — this time Shohei Ohtani against the same Pittsburgh team that just produced LAD’s 12-3 outcome.

The parallels matter. Yesterday: PIT’s ace (Skenes) at home, +21.3% AGREE, blew up 12-3. Today: LAD’s ace (Ohtani) on the road at PIT, +8.2% AGREE. The model is much less confident on Ohtani than it was on Skenes — appropriate humility after the previous day’s outcome.

The “back the ace” tracker (last 5 attempts):

• ✅ deGrom (Sun): W 10-0

• ✅ Buehler (Mon): W 6-2

• ✅ Cole (Tue): W 3-2

• ❌ Skenes (Tue): L 12-3

• 🟡 Ohtani (Wed): TBD

3-1 on the archetype in the last 5 plays. If Ohtani cashes, it’s 4-1. If LAD loses, the back-the-ace archetype has its second loss in two days and the calibration loop will pull the shrinkage factor harder. The model knows what’s at stake here — the +8.2% number is calibrated against everything that came before. Strong tier, not lock — the magnitude tells you the model isn’t asking for blind conviction. 🎭

What to watch: Ohtani’s hitting matters too. He’s slotted in as the DH on his pitching days. A 9.18 r projected total with LAD’s offense vs Joey Jones means the model expects LAD to score 5+ runs. Ohtani himself contributes to that. If LAD wins by a comfortable margin, this play looks great. If it’s a close one-run game either direction, that’s noise.

Leave a comment

📋 Strong & Solid Tier

The PHI/TOR same-matchup flip is interesting. Yesterday: TOR favored (model fading Wheeler). Today: PHI favored (model fading Scherzer). The model is anti-ace on this series specifically — TOR’s #1 was faded successfully, now PHI’s veteran is being faded. Two consecutive ace fades on the same series. Watch the (FADE × STRONG) cell tonight. 🧮

🚫 Lean / Pass Tier

WSH @ SF (WSH +2.4% FLIP) — Massive edge shrink from yesterday (+14.2% LOCK FLIP cashed 6-3). Today: just +2.4%. Ray (the SF ace) is going. Lean tier, no oversize.

NYY @ CLE (NYY +1.9% FLIP) — Day 2. Yesterday Cole back cashed 3-2. Today Rodón is going for NYY, but the edge is microscopic. Lean tier.

STL @ NYM (STL +1.5% AGREE) — Day 2. Yesterday STL +9.8% FLIP cashed 7-0 (shutout). Today the magnitude is tiny. Dog value play but lean tier.

CHC @ COL (CHC +0.8% AGREE) — The Coors slugfest of the slate, two days running. Yesterday COL +1.2% won 7-3 (model was right on direction but no real edge); today CHC +0.8% AGREE — both teams roughly priced fairly. PASS tier with Imanaga going for CHC at 11.67 r projected total. The model refuses to engage at Coors without a real edge. 🏔️

Share

⚠️ Caveats

1. ❌ Not betting advice. Model output. Same caveat, every day.

2. 🔄 Three consecutive days at 70%+ is far above the model’s true rate. Expectation reverts. Today’s 60-65% range is the long-run baseline, not 80%.

3. 🩸 The Skenes loss should temper any “back the ace” enthusiasm. Ohtani today is not an automatic +EV play — it’s a +8.2% strong-tier edge, magnitudes smaller than Skenes was, and the calibration loop has presumably tightened the cell. Smaller magnitude = smaller conviction, full stop.

4. 🔒 3 locks is a meaningful conviction day even after yesterday’s 5. The variance band is still wide, especially on the FLIP lock (OAK) which historically is the column’s most volatile bucket.

5. 🧮 Tier reminder: Column uses Lock ≥10%; site methodology says ≥15%. By strict cut today, zero ultra-locks (none clear +15%). All three locks today live in the 10-18% range.

6. 🌐 The site has expanded a lot. If you’ve only been reading the Substack — check out the new /props and /model-health pages.

✅ Where The Column Stands At 17 Days

Three days at 70%+ in a row, including a 13-2 Sunday, a 7-1 Monday, and yesterday’s 11-4 even with the column-record edge losing 12-3. The single biggest morning edge in column history blew up — and the slate around it still went 11-4. That’s portfolio thinking. That’s why the column doesn’t tout single locks. That’s why the methodology page describes a probabilistic model, not a guarantee generator.

And in parallel: Bad Bets has quietly become a real sports platform. NFL fantasy leagues open. MLB props model live. Model Health diagnostics for both. Three-month-old column hand-coded in Substack footnotes has turned into a full transparent quant ecosystem.

Site: bad-bets.netlify.app. ⚾🏈

See you tomorrow.

— T&G

Subscribe now

The model is back to 5 LOCKS (≥10% edges) on a 15-game Tuesday. Top edge is PIT +21.3% AGREE with Paul Skenes on the mound — the largest morning edge in column history.

Yesterday SEA @ BAL +20.6% broke the previous record. Today PIT @ LAD breaks it again. The column has set a new record in two consecutive days. Either the market is asleep, the model is hot, or we’re about to find out where the regression lives. Probably all three. 🎢

The 5 locks: PIT +21.3% AGREE · CIN +16.5% AGREE · WSH +15.4% FLIP · MIA +13.8% AGREE · SEA +12.2% AGREE (Day 2 in a row on the SEA @ BAL line — yesterday’s +20.6% bigger version cashed 6-3).

Strict methodology threshold (≥15%): 3 ultra-locks (PIT, CIN, WSH). Column threshold (≥10%): 5 locks. Either way, this is the heaviest conviction day on the board. ⚖️

📊 From The Ledger: Monday Was Also A Massive Day

Model picks: 7-1 (87.5%). Edges: 7-1 (87.5%). Two consecutive days at ≥85%.

Across the weekend + Monday: 20-3 (87%) on model picks over the last two days. The “going downhill” question from last week now feels like one of the column’s stranger artifacts — written exactly one day before the model produced its best 48-hour stretch of the run.

The Monday slate I covered in yesterday’s morning article was 7 games. The full graded slate was 8. The 8th was SEA @ BAL +20.6% AGREE LOCK — discovered by the 5 PM ET lineup-aware re-run after the column went out — and it cashed SEA 6, BAL 3. The biggest morning edge in column history was found by the new infrastructure that shipped two days ago. The column missed it. The site caught it. The site won the day. ✅

I’m flagging this transparently because it’s exactly the kind of methodological gap the new lineup-aware re-run is built to close. Team-level wOBA blind spots get filled in once lineups post. The +20.6% number wouldn’t have existed in the morning slate even if I’d covered all 8 games — it required the lineup-aware corrected wOBA to materialize. That’s a real validation of the v0.7 audit work that auto-shipped last week.

The MIL lock deserves its own paragraph for sheer absurdity. The model projected the total at 7.33 runs. The teams scored 29. A LOCK at +14.2% AGREE projected for the lowest-scoring environment of the slate (Coliseum, 59°F cold air, neutral park factor) and the game ended 15-14. Lock still cashed by one run. The path was unpredictable. The outcome wasn’t. That’s variance. That’s also why the column says “don’t oversize on locks” every single day. 🎲

Subscribe now

🌟 The Ace Story: 3rd “Back The Ace” Play In 3 Days

The ace archetype has dominated the last 72 hours:

• 🟢 Sun: TEX backed deGrom +7.9% AGREE → TEX 10, CLE 0 (SHUTOUT)

• 🟢 Mon: SD backed Buehler +4.0% AGREE → SD 6, CIN 2 (WIN)

• 🟡 Today: PIT backs Paul Skenes +21.3% AGREE → TBD

The model went 1-7 fading aces in the first three weeks. Skenes was specifically faded twice and lost both times. Today’s slate flips the script: the model BACKS Skenes at the column’s largest edge ever. PIT at home (PF 98) facing Eric Lauer in an 85°F game. Model has PIT at 71.9% — that’s an ace-driven number, the kind only justified when the composite reads the SP gap as enormous.

Updated ace-back / ace-fade record across the run:

Total ace-related plays: 3-7. The two recent backs are the only wins. Skenes today is the cleanest test yet — if PIT cashes, the back-the-ace archetype goes 3-0 in three days and the column has its second clear ace-pattern validation after the deGrom 10-0. If PIT loses, the segment cell on /stats will reflect it tonight and the calibration loop will pick up the data point.

The calibration loop on the site is built precisely for this. The (direction × bucket) shrinkage will eventually sort the ace pattern out automatically. No more hand-writing audit columns about ace exposure — the system handles it now. ⚙️

Subscribe now

📈 Running Ledger Through 16 Days

196 picks at 60.7%. Back above the 60% line for the first time in a week, on the strength of two consecutive 85%+ days. Trajectory restored. 📊

For audit-curious readers: the full pick-by-pick ledger lives at bad-bets.netlify.app/history — filter by direction, bucket, date. The live calibration grid (n / hit% / ROI / Wilson CI / shrinkage factor per cell) lives at /stats.

🎯 Today’s Slate — Five Locks, Loud Disagreements

Headline numbers:

• ⚾ 15 games — full Tuesday card

• 🔒 5 LOCKS (≥10%) — biggest conviction day on the board to date

• 🎯 9 actionable plays (≥4%) — 5 locks + 4 strong/solid

• 📈 Top edge: PIT +21.3% (Skenes backed) — column record

• 🔥 Coors slugfest: CHC @ COL projects 11.65 r at PF 112 — PASS (only +0.3%)

• 🪑 Pitchers duel: PHI @ TOR projects 6.73 r at Rogers Centre (dome) — TOR strong edge

Subscribe now

🔒 Expanded Lock Breakdowns

🥇 #1 · PIT +21.3% AGREE — TOP EDGE, BACK THE ACE

Model: PIT 71.9% · Market: PIT 50.6% · Total est: 9.72 r · PF: 98 · WX: 85°F Matchup: Eric Lauer (LAD) vs Paul Skenes (PIT)

The column’s largest morning edge in 16 days of publishing. The pricing tells you why: market has PIT as a 50/50 coin flip, model has PIT at 72%. That’s a 21-point spread on a moneyline, which the market only allows when the SP composite is screaming. Skenes’ real FIP, K-BB%, and recent-form bonus are all elite. Lauer is a back-end starter the model probably reads as borderline replacement-level.

The historical ace-back context matters here: PIT lost both prior Skenes starts the model engaged with — both were FADES of Skenes that lost. This is the first Skenes-BACK the model has shown. After deGrom (10-0 shutout backed Sunday) and Buehler (6-2 backed Monday), today’s play is the cleanest ace-back test yet by edge size. +21.3% is a number that essentially says “either I’m right and this game is over, or I’m broken on aces.” Either outcome is informative. 🎯

What to watch: If PIT cashes, the ace-back archetype goes 3-0 in 3 days and the column gets its second loud ace validation. If PIT loses despite Skenes pitching, the model has a serious problem with the SP composite that goes beyond noise. No middle-ground outcome — this game is signal.

🥈 #2 · CIN +16.5% AGREE — PETCO PITCHER ENV

Model: CIN 69.1% · Market: CIN 52.6% · Total est: 8.06 r · PF: 91 · WX: 68°F Matchup: Andrew Abbott (CIN) vs Lucas Giolito (SD)

Petco at 68°F is a meaningful pitcher’s park. Model thinks CIN is way more favorite than the market is pricing — 69.1% vs 52.6%, a 16-point spread. Abbott has been quietly grading well in the composite, Giolito has been inconsistent. The cool San Diego marine layer + Petco’s spacious dimensions suppress run scoring → the better arm tends to dominate.

Direction-AGREE on a road favorite at a pitcher’s park — this is the same archetype as MIL’s lock yesterday (which cashed in a 15-14 game because the model was right about the SP gap even though variance produced an absurd outcome). 🌊

🥉 #3 · WSH +15.4% FLIP — ROAD DOG WITH OUTDOOR WIND

Model: WSH 64.2% · Market: SF 51.1% · Total est: 9.08 r · PF: 94 · WX: OUT 13MPH 64°F Matchup: Patrick Alvarez (WSH) vs Adrian Houser (SF)

The lone FLIP lock today and the model’s biggest direction-disagreement of the slate. Market has SF as a slight favorite (51.1%); model has WSH at 64.2% — flipping the favorite outright. The 13 mph wind at Oracle is the wildcard — Oracle Park is a pitcher’s park, but heavy outdoor wind can suppress or amplify depending on direction (the data only gives speed, not vector).

FLIP-lock history is the column’s weakest segment per the calibration loop framework — the cells with the worst Wilson CIs tend to be in this exact bucket. The model’s automatic shrinkage will be applied if /stats has it underperforming. Whatever shows up in today’s edge is already post-calibration. That’s the protection the v0.7 work bought us. 🛡️

🏅 #4 · MIA +13.8% AGREE — GALLEN FADE

Model: MIA 68.1% · Market: MIA 54.3% · Total est: 7.79 r · PF: 96 · WX: DOME Matchup: Zac Gallen (ARI) vs Max Meyer (MIA)

The model is FADING Gallen — this is the ace-FADE play on the slate, and after 1-7 fading aces, this is exactly the archetype the column has been most cautious about. But it’s also the archetype with the most data points. If the model has been consistently wrong fading aces, the calibration shrinkage on /stats should be hammering this cell — meaning the +13.8% number is what survives after the model already discounted its ace-fade tendency.

MIA backing Max Meyer at home (dome, neutral PF) against a Gallen who the model thinks is overvalued — this is the kind of edge that either confirms the SP composite is correctly reading Gallen’s recent form OR confirms the ace-fade pattern is structural. Worth a paragraph in tomorrow’s recap either way. ⚙️

🎖️ #5 · SEA +12.2% AGREE — DAY 2 IN A ROW

Model: SEA 63.3% · Market: SEA 51.1% · Total est: 9.68 r · PF: 100 · WX: 83°F Matchup: Logan Gilbert (SEA) vs Kyle Gibson (BAL)

SEA @ BAL is on the slate for the second consecutive day. Yesterday’s version (Hancock vs Gibson, lineup-aware corrected to +20.6%) cashed 6-3. Today’s version (Gilbert vs Gibson, +12.2% as a morning edge) is a smaller but still-lock play. The model has now flagged this matchup twice in 48 hours.

The interesting wrinkle: the morning team-level wOBA gave it +12.2%. Yesterday the lineup-aware re-run at 5 PM bumped it to +20.6%. Watch the LINEUPS IN badge on the today page at 5 PM ET — if the lineup-aware version moves the edge meaningfully (≥2pp delta), that’s another data point on the new infrastructure working as designed. 📡

Subscribe now

📋 Strong & Solid Tier

The Wheeler fade is the second ace-FADE on the slate (alongside MIA’s Gallen fade). After the column’s 1-7 record fading aces, two FADE plays on one slate is a stress test for the SP composite. The (direction × bucket) cell for “FADE the ace” on /stats is the cell to watch tonight.

The Cole back is the smallest ace-back yet (+4.2% vs Skenes +21.3%, deGrom +7.9%, Buehler +4.0%) — the model is much less confident on Cole than on Skenes today. That’s appropriate magnitude calibration: Cole’s composite scores well but not at Skenes’ level.

🚫 Pass / Lean Tier

MIL @ OAK (OAK +3.2% FLIP) — Day 3 of MIL @ OAK on the slate. Yesterday MIL +14.2% LOCK cashed in a 15-14 game. Today the model flips direction to OAK. The recompute-from-scratch thesis at work: model isn’t anchored to yesterday’s lock win, it’s reading today’s matchup independently (Gasser vs Ginn, both back-end starters, Coliseum cold). Lean tier — no oversize.

HOU @ LAA (LAA +2.2% AGREE) — Dog value play. Model has HOU at 51.1% vs market 53.4%. Both agree HOU wins; model thinks the magnitude is closer than priced, so LAA gets dog value. Lean tier discipline.

CHC @ COL (CHC +0.3% AGREE) — The Coors slugfest of the slate. PF 112, 11.65 r projected, 88°F game. Model has CHC at essentially 50/50 vs market — no edge. Pass tier. Coors is the park where the column historically refuses to engage on small edges because the variance overwhelms the math. The model agrees today. 🏔️

Share

⚠️ Caveats — Read These

1. ❌ Not betting advice. Model output. Same caveat, every day.

2. 🔢 5 locks is unusual. The column has shown 3 locks at most across the run. Variance band is wide on locks specifically — yesterday’s MIL lock won 15-14, the column has had locks lose by blowout, and “lock” doesn’t mean “guarantee.”

3. 📉 After two consecutive 85%+ days, regression is more likely than continuation. Two-day stretches at 87% are inside the variance band of a 58-60% true-rate model — but they’re at the rare positive tail. The expectation for today is something closer to the long-run rate, which is 60%-ish.

4. 🎯 Don’t oversize on the 5-lock day. The column has already broken the morning-edge record twice in 48 hours. Markets that mispriced this badly two days running are not zero-information events. The model could be hot or the market could be slow on injury/lineup news.

5. 🧮 Tier definitions: Column uses Lock ≥10%. Site methodology uses Lock ≥15%. By the stricter definition: 3 ultra-locks today (PIT, CIN, WSH). Either threshold, this slate is loud.

✅ The Column At 16 Days

A daily transparent read on where the MLB betting model and Vegas disagree. Two straight 85%+ days. The largest morning edge in column history, two days running, both records. The “back the ace” archetype testing positive three days in a row. The audit infrastructure auto-finding edges (+20.6% SEA) that the morning column missed.

The site is live: bad-bets.netlify.app. Methodology, ledger, live calibration. Audit it yourself.

See you tomorrow. ⚾

— T&G

Subscribe now

bad-bets.netlify.app is live.

Sixteen days of Substack columns and slate graphics, and now there’s a real platform behind it. The site has:

• 📊 /today — the live slate, refreshed in the morning and again at 5 PM ET after lineups post

• 📜 /history (Ledger) — every pick the model has ever made, with results, filterable by direction (AGREE / FLIP), bucket (lock / strong / solid / lean), and date range

• 📈 /stats — the rolling track record AND the model’s live (direction × bucket) calibration grid with Wilson 95% confidence intervals and the current shrinkage factors

• 🔬 /methodology — the model end-to-end. Inputs, market read, edge interpretation, the AGREE/FLIP thesis, the afternoon lineup-aware re-run, the calibration learning loop. Nothing hidden.

• ⚡ /live — in-game momentum while the action is happening

• 📰 /news and 📅 /schedule for daily context

Bad Bets is no longer just a Substack column. It’s a transparent quant platform with a public receipt for every claim. That’s the whole point — the average tells you whether to trust this, not the highlights. The site lets anyone audit the average. 🛠️

Subscribe now

🔬 The Audit Became Infrastructure

Two big shipped features deserve their own paragraph because they are exactly the v0.7 priorities I was hand-writing in the column last week:

1. The 5 PM ET lineup-aware re-run. Morning slate uses team-level wOBA (lineups don’t post until 2-5 PM). At 5 PM a second pass replaces team wOBA with lineup-specific wOBA — actual hitters, L/R splits vs the opposing starter, weighted by lineup position (1-3 at 1.20×, 4-6 at 1.00×, 7-9 at 0.85×). Today cards show a LINEUPS IN badge once the re-run covers them, with a delta indicator if the lineup-aware edge changed by ≥2pp. The team-level wOBA blind spot I was flagging is now closed.

2. The calibration learning loop. Every night after games are graded, a recalibration script reads the full log and computes hit-rate per (direction × bucket) segment with Wilson 95% CI. If any segment’s CI ceiling sits below the −110 breakeven (52.4%), the script writes a shrinkage factor < 1.00 to a config file the model reads on the next slate. It can only shrink (1.00 → 0.50 minimum). It refuses to act on segments with fewer than five graded picks. It requires the CI ceiling — not the point estimate — to be below breakeven. That combination means the model can’t overfit a single bad week. As the log grows, the calibration sharpens.

The state of every cell is live on /stats. Readers can see the model correcting itself in real time. The audit is no longer a column section. It’s the system. 🧠

The remaining v0.7 work I was outlining (v0.5 vs v0.6 backtest on the data source, explicit ace-dampener filter): the second is now redundant if the calibration loop catches the underperforming archetype. The first is still useful diagnostic work but no longer urgent. The biggest lever is shipped.

Subscribe now

📊 From The Ledger: Sunday Was The Biggest Comeback Day Of The Run

After five straight losing days, the model produced its single best day of the entire 15-day run.

Model picks: 13-2 (86.7%). Edges: 8-7 (53.3%). Actionable plays: 6-3 (66.7%). LOCKS WENT 3-0. 🎯

The three locks, all cashed:

OAK was the biggest morning edge in column history. After losing as a lock on Friday (1-5) and getting blown out as a lock on Saturday (2-13), the model bumped the edge to +17.9% on Sunday — a new high on a team that had just been blown out twice. And OAK won 5-0 in a SHUTOUT. The “recompute from scratch” thesis got its loudest validation possible. Four straight days on OAK: 1-1 lock record before yesterday, now 2-2. The bigger-edge day cashed in dominant fashion.

The other actionable plays:

6-3 on the actionable tier. The losses were all reasonable variance (SEA one-run loss, LAA blown out by LAD’s offensive eruption, WSH road struggles continue).

Subscribe now

🌟 The deGrom Validation

The “back the ace” test passed in shutout fashion. Yesterday’s article framed deGrom as the cleanest possible test of whether the model’s elite-pitcher composite was broken on fades only (the 1-7 archetype) or broken on both sides. deGrom delivered a 10-0 shutout. TEX +7.9% AGREE cashed in dominant fashion.

Updated ace record across the run:

• Cole: 1-1 (faded)

• Skenes: 0-2 (backed both times, lost)

• Yamamoto: 0-1 (faded, ace won)

• Strider: 0-3 (faded all three times, ace won all three)

• deGrom: 1-0 (backed, won 10-0) ✅

• Combined: 2-7 across 9 attempts — still concerning, but the deGrom data point matters because it’s the first BACK that cashed since Skenes.

The pattern looks like: when the model FADES a true elite arm, it’s gotten burned (0-7 across Skenes/Yamamoto/Strider attempts). When the model BACKS a true elite arm, the sample is tiny but the one win is significant. The calibration loop will sort this out automatically over the next 30-50 picks. This is exactly what the (direction × bucket) shrinkage is built for.

🤔 What Yesterday’s Result Means For Last Week’s “Going Downhill” Question

Friday’s column had the explicit framing:

“My honest read: The model isn’t broken. It’s having a -1.5σ stretch over 4 days within a 13-day above-breakeven sample. Don’t structural-overhaul on 5 days of picks. Wait another 7-10 days.”

Day 6 produced a 13-2 result. The 5-day cold stretch turned out to be exactly what it looked like statistically — a -2σ noise window inside a model that’s true rate sits in the high 50s. The discipline call (don’t act on small samples) was the correct one. If the column had announced a structural overhaul on Friday, the 13-2 day would have come from a worse model. 🧘

This is the column’s most important methodological lesson to date: don’t overfit to recent variance. The calibration loop on the site is the codification of that principle — it requires 5+ picks per segment AND a CI ceiling below breakeven before it’ll move any shrinkage factor. Same logic, automated.

Subscribe now

📈 Running Ledger Through 15 Days

188 picks at 59.6%. Bounced from yesterday’s 57.2% back up to 59.6% on the strength of Sunday’s 13-2. The trajectory looks exactly like a 58-60% true-rate model with normal variance bands around it. 7+ points above breakeven across 188 picks. Closing the gap on what the live calibration on the site will eventually pin down. 📊

For the audit-curious: the full ledger is at /history. Filter by direction or bucket and see the cells yourself.

Subscribe now

🎯 Today’s Slate — Light Monday Card, MIL Returns As Top Lock

Headline numbers:

• ⚾ 7 games (Monday slate — most teams off)

• 🔒 1 LOCK — MIL +11.9% AGREE @ OAK

• 🎯 5 actionable plays (≥4%) — 1 lock + 4 strong/solid edges

• 🔥 Slugfest: HOU @ LAA projected 10.57 r at Angel Stadium

• ❄️ Pitchers’ duel: NYY @ CLE projected 7.09 r (Warren vs Williams)

🔒 The Lone Lock — Expanded Breakdown

🎯 MIL +11.9% @ OAK · AGREE (🔁 MIL BACK)

Model: MIL 71.7% · Market: MIL 59.7% · Total est: 7.33 r · PF: 91 (Coliseum, cold) Matchup: Kyle Harrison (MIL) vs Jeffrey Springs (OAK)

MIL on the slate three straight days, the edge persisting after a blowout win:

• 🟦 Sat Jun 6: not actionable for MIL (COL series day 2)

• 🟩 Sun Jun 7: +10.7% LOCK AGREE @ COL → MIL 12, COL 4 BLOWOUT

• 🟦 Today: +11.9% LOCK AGREE @ OAK → TBD

The edge GREW slightly (10.7% → 11.9%) after the blowout win. The model isn’t anchored to yesterday’s 12-4 — it’s pricing today’s matchup independently. MIL ships Kyle Harrison out (strong recent grades on K-BB%), OAK counters with Jeffrey Springs (decent but not elite). The Coliseum at 59°F with cold air pulls runs out of the equation. Pitcher-friendly environment that amplifies the pitcher-quality gap the model is seeing.

The total projection (7.33 r) is on the low end of today’s slate — exactly the kind of game where the better arm tends to win the one-mistake battle. Same archetype as yesterday’s MIL lock (+10.7% at Coors, which despite the high run environment converted into a 12-4 blowout because the pitching gap was that wide).

What to watch: The OAK Coliseum has been a graveyard for road favorites historically — the cold, the foul ground, the night game on West Coast. But the v0.6 composite has already priced those factors. The model thinks 71.7% is still meaningfully better than the market’s 59.7%. If MIL cashes, that’s two straight lock AGREE wins on the team — confirmation that the calibration is reading them correctly. If MIL loses, the lock-AGREE-MIL archetype gets one data point against, and the segment cell on /stats will reflect it tonight.

Share

📋 The Strong & Solid Edges

Note the deGrom symmetry: yesterday the model backed deGrom +7.9% and cashed 10-0. Today it’s backing Buehler +4.0% — a similar archetype (veteran ace at a pitcher-friendly park) but with a meaningfully smaller edge. The model is appropriately less confident on Buehler than it was on deGrom. That’s the kind of magnitude calibration the v0.6 composite is supposed to produce.

🚫 Pass / Lean Tier Notable Spots

BOS @ TB (TB +3.1% FLIP) — Lean tier. Brayan Bello vs Drew Rasmussen at Tropicana Field (dome). Pure direction-flip with small magnitude. Pass-tier discipline says don’t publish.

WSH @ SF (SF +2.9% AGREE) — Webb going for SF. Model has SF at 62.6% vs market 59.7%. Small AGREE edge in lean tier.

Leave a comment

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Same caveat, every day.

2. 🎰 1 lock on a 7-game slate. Light Monday card by design. Don’t oversize on the limited menu.

3. 📊 188 picks at 59.6%. True rate guess: 55-60%. The site’s calibration grid will pin this down more precisely as the sample grows.

4. 🔬 Audit infrastructure is shipped. /stats shows the model correcting itself in real time. No more hand-wringing in this column about v0.7 changes — the system handles it now.

✅ What This Column Is, Fifteen Days In

A daily transparent read on where the MLB betting model and Vegas disagree. Locks that go 3-0 on a comeback Sunday after five days of losing. The biggest morning edge in column history cashing as a shutout. The “back the ace” test passing 10-0. The audit thesis getting vindicated by exactly the model that “seemed to be going downhill.”

And now: a website with the receipts. bad-bets.netlify.app. Methodology, ledger, stats, live calibration. The work that used to live in column footnotes now lives in a public platform that anyone can audit.

If you’re new to Bad Bets:

• 🌐 Site: bad-bets.netlify.app

• 📩 Substack: subscribe for the daily column

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for JSON outputs and v0.6+ internals

See you tomorrow. ⚾

— T&G

Subscribe now

Locks: 2-2. Edge plays: 3-5. Model picks: 5-7 (41.7%). The shutouts went in both directions.

Direction tracker now stretched to four days: AGREE 13-13 (50%), FLIP 5-10 (33%) across the v0.6 window. The AGREE-edge thesis is back to a coin flip, FLIP is still struggling. N=41 still inconclusive. 📊

The headline wins:

• 🎯 CWS +15% top lock cashed 6-3 — the column’s pet team back on top

• 🎯 SEA +11.9% AGREE lock CASHED 4-0 SHUTOUT — yesterday’s article framed SEA as “if the AGREE thesis has any predictive power at all, this is the lock that should reflect it.” Friday’s SEA lock lost, Saturday’s SEA lock cashed in a SHUTOUT. The audit test produced a recovery data point. ✅

The headline losses:

• 🚫 OAK +12.8% LOCK got BLOWN OUT 2-13 — 4 runs ago on Friday, today by 11 runs. Two straight blowout losses for the OAK FLIP archetype.

• 🚫 TEX +5.6% solid SHUTOUT 0-6 by Cleveland — boring AGREE tier delivered nothing

• 🚫 PIT +8.6% Strider fade lost again — that’s now 1-7 across 8 ace-fade attempts.

Leave a comment

🤔 Let’s Talk About The Last Four Days

The cleanest thing I can do here is show you the math and call it what it is.

The last 5 days the model has been below breakeven (52.4%). Edge plays specifically over that window: 24-33 (42%). That’s not nothing.

But here’s the statistical context:

A model with a true 58% hit rate will produce 5-day windows at 45% roughly 6-8% of the time, purely by variance. One 5-day cold stretch in 14 days isn’t proof the model degraded. It’s also not proof it didn’t.

The complication: two things changed right when the cold stretch began.

1. 📉 Data source switched from FanGraphs wRC+ to MLB API wOBA proxy on/around June 4 due to rate-limiting. If the proxy is noisier, the offense input degraded exactly when the picks got worse.

2. 🎯 The ace-fade pattern went from 1-3 to 1-7 across the same window. Strider 0-3 alone is brutal. Cole 1-1, Skenes 0-2, Yamamoto 0-1, Strider 0-3 — that’s 1-7 on a specific archetype the model keeps publishing.

My honest read: The model isn’t broken. But it has TWO specific items that look more like signal than noise:

🔬 v0.7 Audit Priorities — Concrete

1. Run a v0.5 vs v0.6 backtest on the last 5 days. Same input games, both models, see which would have called the slate better. 30 minutes of work. If FanGraphs wRC+ would have outperformed MLB API wOBA proxy, restoring the FanGraphs feed (caching layer, paid tier, or rotating IPs) is the single highest-ROI fix on the board.

2. Ace-dampener filter. The ace-fade archetype is 1-7. Two practical options:

• Conservative filter: Don’t publish ≥4% edges where the side you’re fading has a starter with K/9 ≥ 10 over their last 5 starts. Just filter them out of the published slate.

• Soft adjustment: Reduce the model’s pitcher-quality penalty by 15-20% for top-decile K/9 starters.

Conservative filter is easier to A/B test. Strider, deGrom, Skenes, Yamamoto, Cole — all would have been removed from yesterday’s slate as fades.

3. Hold off on lock-threshold and lineup-data changes until v0.5 vs v0.6 backtest comes back. One change at a time.

4. Continue logging direction data. N=41 is not enough. Need 100+ direction-tagged picks before any change to the edge filter. We’re 41% of the way there.

The discipline call is the same it’s been all week: Two days of bad data is variance. One specific archetype going 1-7 across eight tries IS data. Build the filter. Run the backtest. Don’t structural-overhaul on 5 days of picks. 🧘

Subscribe now

📈 Running Ledger Through 14 Days

173 picks. 57.2% model accuracy. Down 1.2 pts from yesterday. Still 4.8 points above the 52.4% breakeven. True rate likely in the 54-60% range with current sample. Tight enough that the next two weeks will narrow the band meaningfully. 📊

🎯 Today’s Slate — Three Locks, And A 4th-Straight Day On OAK

Headline numbers:

• ⚾ 15 games on the card

• 🔒🔒🔒 3 LOCKS — OAK, SF, MIL (1 AGREE / 2 FLIP)

• 🎯 9 actionable plays (≥4%) — 3 locks + 6 strong/solid

• 🔥 Slugfest: MIL @ COL projected 10.90 r at Coors (3rd straight series day)

• ❄️ Pitchers’ duel: BOS @ NYY projected 6.83 r (Suarez vs Schlittler)

• 🌟 Aces today: Jacob deGrom (TEX), Luis Castillo (SEA), Aaron Nola (PHI), Kevin Gausman (TOR)

Leave a comment

🔒 The Three Locks — Expanded Breakdowns

🎯 #1 — OAK +17.9% @ HOU · FLIP (BIGGEST EDGE IN COLUMN HISTORY)

Model: OAK 67.3% · Market: HOU 50.6% · Total est: 9.45 r · PF: 101 (Minute Maid) Matchup: Gage Jump (OAK) vs Mike Burrows (HOU)

This is the largest single morning edge the model has published across all 14 days. Prior peak: MIA +16.9% on Friday — which got shut out 0-6.

OAK has now appeared on the slate FOUR straight days:

• 🟩 Tue Jun 2: +16.0% LOCK FLIP → won 2-1 (cashed)

• 🟥 Fri Jun 5: +10.3% LOCK FLIP → lost 1-5

• 🟥 Sat Jun 6: +12.8% LOCK FLIP → lost 2-13 BLOWN OUT

• 🟦 Today: +17.9% LOCK FLIP → TBD

The model went 16% → 10.3% → 12.8% → 17.9% across the four days. The edge keeps moving as new pitching matchups arrive. It’s now bigger than the morning OAK lock that cashed on Tuesday.

Why the model still loves it: Gage Jump’s v0.6 composite grades meaningfully ahead of Mike Burrows. Jump has been one of OAK’s better arms; Burrows is a back-end HOU rotation piece. The model thinks the market is way undervaluing the gap.

The audit angle: This is the cleanest test yet of the “recompute from scratch” thesis. The model isn’t anchored to two straight blowout losses. The edge actually GREW today vs yesterday. Either the model has identified something real (the matchup math) or it’s repeatedly making the same mistake on OAK road games. Today’s outcome is N=1 either way, but a meaningful one.

What to watch: If OAK wins, the recompute thesis has its strongest validation to date — the model correctly identified that a 2-13 loss yesterday doesn’t change today’s math. If OAK gets blown out again, three straight FLIP-lock blowouts on the same team becomes a structural pattern worth investigating, not variance.

🎯 #2 — SF +12.1% @ CHC · FLIP

Model: SF 58.9% · Market: CHC 53.1% · Total est: 9.36 r · PF: 102 (Wrigley) Matchup: Trevor McDonough (SF) vs Jameson Taillon (CHC)

SF column history:

• 🟥 Sun May 31: +1.1% pass → no bet (SF lost in extras)

• 🟥 Tue Jun 2: +3.8% lean → SF lost 12-2 (got crushed by MIL)

• 🟥 Wed Jun 3: not on slate → SF blew out CHC 19-6

• 🟥 Fri Jun 5: not on slate → SF blew out CHC 18-3

• 🟦 Today: +12.1% LOCK FLIP → TBD

SF has been in some absurd blowouts on both sides of the ledger. The model now has them as a direction-flip lock against CHC.

Why the model loves it: Trevor McDonough vs Jameson Taillon. Taillon has the veteran name brand but his real-FIP composite is weaker than market pricing. McDonough is a young arm grading well. SF’s offense has been streaky-hot — they’ve put up 18 and 19 runs in their last two games at this park.

The risk: McDonough is a lefty rookie facing a Cubs lineup that just got blown out by Adrian Houser of all people. The variance band on SF games is currently enormous. Either side could win 12-2.

🎯 #3 — MIL +10.7% @ COL · AGREE (COORS PART 3)

Model: MIL 73.4% · Market: MIL 62.8% · Total est: 10.90 r · PF: 112 (Coors Field) Matchup: Shane Drohan (MIL) vs Kyle Freeland (COL)

COL series at Coors, day 3:

• 🟩 Fri Jun 5: COL +5.9% solid AGREE → MIL 9, COL 7 (model picked MIL ✅, edge play COL ❌)

• 🟩 Sat Jun 6: MIL won decisively (no actionable edge published)

• 🟦 Today: MIL +10.7% LOCK AGREE → TBD

The model has flipped from “COL underdog edge” Friday to “MIL favorite lock” today. The market caught up to MIL’s superiority across the series, and the model now thinks the market is still undervaluing MIL even at -180-ish odds.

The Coors warning: At PF 112 with a 10.90 r projection, this is the kind of game that can swing on bullpen variance alone. The MIL 73% probability is the model’s read, but the variance band on Coors games is the widest in baseball. A 10% edge on a high-variance game is structurally different from a 10% edge on a Petco game (NYM lost yesterday).

📋 The Strong & Solid Edges

The deGrom play is the most interesting on this list. The model’s ace-related record is currently 1-7 across 8 attempts. Six of those attempts were fades. Today’s deGrom play is a BACK — same archetype as Skenes (0-2). If deGrom cashes, that’s a real data point in the model’s favor on backing aces. If deGrom loses, the model’s elite-pitcher composite has a 1-8 record and needs structural review. 📊

Leave a comment

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Same caveat, every day.

2. 🎰 3 locks today. Lock tier is 5-10 across the run (33.3%). Don’t oversize the lock tier. Variance is the dominant factor.

3. 📊 173 picks at 57.2%. Down 1.2 pts from yesterday. Still meaningfully above breakeven but the lead is shrinking.

4. 🔬 v0.7 audit priorities are now concrete: v0.5 vs v0.6 backtest, ace-dampener filter, continue direction logging. Structural changes wait for data.

✅ What This Column Is, Fourteen Days In

A daily transparent read on where the MLB betting model and Vegas disagree. Locks that cash 6-3 and locks that get blown out 2-13. AGREE thesis that flipped back and forth four times in five days. The biggest morning edge of the entire run published the same week the model went 1-7 on ace-fade plays.

The model isn’t broken. The model needs targeted maintenance. That’s a different problem than “make new model.” Today’s slate is published. Tomorrow’s audit work starts.

If you’re new to Bad Bets:

• 📩 Subscribe for daily MLB picks at 10am ET

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for JSON outputs, v0.6 internals, and v0.7 audit progress

See you tomorrow. ⚾

— T&G

Subscribe now

Edge picks: 7-7 (50%). Model picks: 6-8 (42.9%). Two big locks went two opposite directions.

The lock results, sorted by morning edge:

Locks went 1-3. The biggest morning edge the column has ever surfaced (MIA +16.9%) got shut out by Tampa Bay 0-6. The second-biggest (WSH +14.9%) blew Arizona out 14-1. Same morning, same direction tier (both FLIP), opposite outcomes. That’s a clean illustration of variance in the high-conviction tier. ⚾

The full edge ledger Friday:

Subscribe now

🔬 The Audit Tracker: Three Days In

Day 3’s data nudges AGREE back ahead, but both directions are below breakeven over the three-day window. This is exactly the noise I committed to waiting through. v0.6’s tracker will need 100+ direction-tagged picks before we can say anything confidently. We’re at 26.

The honest read: if you had to bet today on whether the direction effect is real, I’d put it at 60/40 yes. Not enough to act on. Enough to keep watching. 📊

🌟 The Bright Spots Worth Circling

🎯 WSH +14.9% LOCK FLIP → 14-1 BLOWOUT at Chase Field. The WSH road team that the market historically fades got all the run support and all the variance. Foster Griffi outpitched Merrill Kelly by a wide margin. FLIP-direction lock cashed in dominant fashion.

🎯 NYM +6.3% STRONG FLIP → 5-0 SHUTOUT at Petco. Pitcher’s duel projection (6.19 r) delivered exactly — except NYM did all the scoring. Christian Scott vs Michael King in the projected 6-run game, Scott went deep with zeros. FLIP-direction strong cashed in shutout fashion.

🎯 STL +3.5% LEAN AGREE → 10-3 WIN. The boring AGREE lean tier kept doing its job.

Leave a comment

📉 The Hurts Worth Owning

🚫 MIA SHUTOUT 0-6. Biggest morning edge of the column’s run, complete dud. The model thought MIA’s Ryan Gusto had a meaningful edge over Drew Rasmussen. The score line says the opposite happened — Rasmussen carved up MIA for a shutout. Pure variance or model error? N=1 doesn’t tell us. Pattern to monitor.

🚫 SEA LOST 3-7 as the only AGREE lock. Yesterday’s article literally framed this as “if the AGREE thesis has any predictive power at all, this is the lock that should reflect it. If it doesn’t, the thesis is officially dead.” The thesis isn’t dead yet — it survived on Wednesday’s data — but Friday’s SEA loss is a real data point against.

🚫 OAK LOST 1-5. Direction-flip lock #3 of the day, lost in a non-competitive game. OAK’s road struggles continue.

🚫 SF blew out CHC 18-3. Pass-tier game, not actionable, but it’s been a SF theme — they’re either getting smoked (MIL 16-2 vs them May 31) or doing the smoking. Variance.

📈 Running Ledger Through 13 Days

161 picks. 58.4% model accuracy. Down from 60.4% yesterday. The trajectory continues the regression-to-mean pattern — the early-week noise narrowing toward a true rate somewhere in the high 50s. The lead over the 52.4% breakeven is shrinking but persistent. 📊

Subscribe now

🎯 Today’s Slate — Four Locks, Three Reruns

Headline numbers:

• ⚾ 13 games on the card

• 🔒🔒🔒🔒 4 LOCKS — CWS, OAK, SEA, NYM (2 AGREE / 2 FLIP)

• 🎯 8 actionable plays (≥4%) — 4 locks + 4 strong/solid edges

• 🔥 Slugfest: CIN @ STL projected 11.03 r at Busch

• ❄️ Pitchers’ duel: NYM @ SD projected 6.97 r at Petco (round 2!)

• 🌟 Aces today: Spencer Strider (ATL), Yoshinobu Yamamoto (LAD), Bryce Miller (SEA), Joe Ryan (MIN)

• 🔁 Reruns: 3 of 4 locks are repeat appearances (OAK day-2, SEA day-2 audit rerun, NYM promoted from yesterday’s strong)

Two of the four locks lost yesterday and are back today (OAK, SEA). One won yesterday and gets promoted from strong to lock (NYM). One is brand new (CWS, the column’s pet team returning).

Let me break each one down properly.

Subscribe now

🔒 The Four Locks — Expanded Breakdowns

🎯 #1 — CWS +15.0% @ PHI · FLIP (TOP EDGE)

Model: CWS 60.6% · Market: PHI 54.4% · Total est: 9.12 r · PF: 105 (Citizens Bank) Matchup: Brandon Eiser (CWS) vs Andrew Painter (PHI)

The column’s defining pet team is back. CWS has appeared on the slate as an actionable edge five times across 13 days:

• 🟩 Tue May 26: +12% strong → won 15-2 blowout

• 🟩 Thu May 28: +6% solid → won 6-2

• 🟥 Sat May 31: pass tier → no bet

• 🟥 Tue Jun 2: +12.6% LOCK → lost 4-6

• 🟩 Wed Jun 3: pass tier → CWS won 8-0 shutout (no edge published)

• 🟦 Today: +15.0% LOCK FLIP → TBD

Record on CWS actionable plays: 2-1. The 2-1 includes one blowout win and one lock loss.

Why the model loves it: Brandon Eiser’s v0.6 composite grades meaningfully ahead of Andrew Painter’s for this matchup. Painter is the rookie name brand, but his early-career FIP and K-BB% reads worse than the market’s pricing. Citizens Bank Park (PF 105) is also a hitter-friendly environment where CWS’s offense vs RHP grades better than market is crediting.

What to watch: Direction-FLIP play. Market has PHI as 54% favorite. Model has CWS winning outright. Same archetype as Tuesday’s CWS lock (lost 4-6). If the bigger-edge (+15% vs +12.6%) play shows up the same way, the model’s CWS thesis needs a longer cooling period.

🎯 #2 — OAK +12.8% @ HOU · FLIP (🔁 2ND STRAIGHT)

Model: OAK 61.8% · Market: HOU 51.0% · Total est: 10.39 r · PF: 101 (Minute Maid) Matchup: Kade Morris (OAK) vs Tatsuya Imai (HOU)

OAK on the slate for the third time this week, second straight day:

• 🟩 Tue Jun 2: +16% LOCK FLIP → won 2-1 (biggest morning edge of that day, cashed)

• 🟥 Fri Jun 5: +10.3% LOCK FLIP → lost 1-5 (yesterday)

• 🟦 Today: +12.8% LOCK FLIP → TBD

Record on OAK actionable plays: 1-1.

Why the model loves it (again): Same series, second straight day. Yesterday OAK lost 1-5. Today the model is back at a bigger edge (+10.3% → +12.8%) because the pitching matchup changed — Kade Morris is a different (and better-grading) arm than yesterday’s Jack Perkins. Imai is a respectable veteran but the gap to Morris is wider than the market is pricing.

The recompute pattern: Same series, lost yesterday, edge increased today. We’ve seen this pattern with LAA (Day 4 streak ended in a 14-3 win at the end), KC (back-to-back lock attempt failed), and now OAK Day 2. The model isn’t anchored. The question isn’t whether it’s right today — it’s whether the recompute approach pays out over many tries. 🔄

What to watch: Slugfest projection (10.39 r) at hitter-friendly Minute Maid. If OAK can score, the model’s pitcher edge translates. If they get shut down again (yesterday: 1 run), the issue is OAK’s offense vs HOU’s specific arms, not the model’s pitcher reads.

🎯 #3 — SEA +11.9% @ DET · AGREE (🔁 AUDIT RERUN)

Model: SEA 67.2% · Market: SEA 55.3% · Total est: 7.86 r · PF: 99 (Comerica) Matchup: Bryce Miller (SEA) vs Keider Montero (DET)

This is the audit’s most important test case of the week. Yesterday’s article explicitly framed SEA’s AGREE lock (+14.8%) as the cleanest possible test of “direction-agreement plays outperform direction-flip plays.” It LOST 3-7. Today the model has SEA back as an AGREE lock at a smaller edge (+11.9%).

Why the smaller edge today: Yesterday’s Bryan Woo got the start (elite K/9). Today’s Bryce Miller is also good but grades slightly lower in the v0.6 composite. The model trimmed conviction from 70.1% to 67.2%, which moved the edge from +14.8% to +11.9%. Recompute from scratch, edge adjusted appropriately.

Why the model still loves it: Miller’s FIP and K-BB% still grade meaningfully ahead of Montero’s. DET’s offense vs RHP rates below average in the v0.6 splits. Same archetype as yesterday’s setup — both model and market favor SEA, model thinks the favorite is more dominant.

What to watch: This is the second straight AGREE lock at +11-15% on the same team. If SEA cashes today, the audit’s AGREE-tier signal gets a recovery data point. If they lose again, two straight AGREE-lock losses on the same team is a meaningful pattern — either the model is overrating Miller/Woo’s pitcher quality, or the AGREE-lock archetype isn’t actually as reliable as Wednesday’s data suggested.

🎯 #4 — NYM +10.7% @ SD · AGREE (🔁 PROMOTED FROM STRONG)

Model: NYM 63.0% · Market: NYM 52.3% · Total est: 6.97 r · PF: 91 (Petco) Matchup: Nolan McLean (NYM) vs Griffin Canning (SD)

NYM on the slate for the second time in three days, and the edge GREW:

• 🟩 Fri Jun 5: +6.3% strong FLIP → won 5-0 shutout at Petco

• 🟦 Today: +10.7% LOCK AGREE → TBD

Note the direction change! Yesterday: model NYM, market SD → FLIP. Today: model NYM, market NYM → AGREE. The market caught up to the NYM read after yesterday’s shutout win. Model still thinks the market is under-pricing NYM by a wider margin than the new line reflects.

Why the model loves it: Petco Park is the most pitcher-friendly venue in baseball (PF 91). The model has McLean grading well; Canning is the home favorite but the gap in real-FIP composite is meaningful. NYM’s road offense vs RHP has more upside than the market is pricing.

Park-pitcher caveat: At PF 91 with a 6.97 r projected total, this is a one-mistake game. The team that scores 3+ runs wins. If NYM’s offense gets shut down by Canning the way SD’s was shut down by Scott yesterday, the model’s pitcher edge doesn’t translate to wins. Reverse-symmetry risk. 🔄

Leave a comment

📋 The Strong & Solid Edges

The STRIDER trend is now a real audit item. The column has now faded Strider twice (May 31 ATL +5.3%, today PIT +8.6%) and backed him zero times. Cole 1-1, Skenes 0-2, Yamamoto 0-1, Strider 0-2 (last attempt). The “fade name-brand starters” thesis is currently 1-5 across 6 published attempts. That’s either real signal (the v0.6 composite over-rates elite arms) or one bad sample.

🚫 Pass-Tier Notable Spots

LAA @ LAD (+1.3% LAA, AGREE) — Yamamoto on the mound for LAD. Model and market both have LAD at 74-76%. Essentially no edge. The model is no longer trying to fade Yamamoto. After Yamamoto threw a 7-inning gem against PHI on Wed (LAD 9-1), the composite recalibrated. He’s now priced correctly.

BOS @ NYY (+1.1% BOS, AGREE) — Ranger Suarez going for BOS. Model has NYY 53.3%, market has 54.4%. Basically a coin flip. Pass tier discipline.

Three locks lost yesterday plus the MIA shutout. Today the model published four more. Conviction tier is doing what conviction tiers do — high variance, sometimes huge wins (WSH 14-1), sometimes brutal losses (MIA 0-6). The math doesn’t change based on yesterday’s outcome. 🧘

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Same caveat, every day.

2. 🎰 4 locks for the second time in three days. Wednesday: 1-2 on locks. Friday: 1-3 on locks. Today: TBD. Don’t get aggressive on lock-tier sizing — the variance band is too wide.

3. 📊 161 picks at 58.4%. The rate keeps moderating but stays above breakeven. True rate guess: 55-60%.

4. 🔬 v0.6 audit signal still inconclusive at N=26. Need 100+ direction-tagged picks. We’re 26% of the way there.

✅ What This Column Is, Thirteen Days In

A daily transparent read on where the MLB betting model and Vegas disagree. Locks that cash 14-1 in shutout-style blowouts. Locks that get shut out 0-6 the same day. Pet teams returning at bigger edges. Strider fades reaching their 5th attempt without a single win. The audit tracker dutifully logging direction data for another 174 picks before any change to the edge filter.

If you’re new to Bad Bets:

• 📩 Subscribe for daily MLB picks at 10am ET

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for JSON outputs, v0.6 internals, and audit progress

See you tomorrow. ⚾

— T&G

Subscribe now

Quick note before the recap: the v0.6 model shipped this morning. Three notable changes:

1. 📐 Direction tracker (AGREE/FLIP) — every pick now labeled by whether model and market favor the same team or opposite teams. This was the #1 audit priority from yesterday’s column and it’s now baked into the daily output.

2. ⚾ Offense input switched to wOBA proxy (MLB API) — the FanGraphs wRC+ feed was rate-limited the last few days. The MLB API’s wOBA-derived proxy gives similar signal with more reliable uptime.

3. 📊 Report card cross-tab — daily grader now outputs bucket × direction (e.g., “solid AGREE 0-1, solid FLIP —”). Lets the audit measure both dimensions simultaneously.

The columns we’ve been hand-classifying for two weeks now ship in the script itself. Bigger samples, faster. 🛠️

Subscribe now

📊 From The Ledger: Thursday Was Tough

Edge picks: 2-6 (25%). Model picks: 4-4 (50%). Two brutal days in three.

The LAD lock loss is what stings most. Yesterday LAD won 7-0 (Ohtani shutout). Today LAD lost 2-3 on a one-run game. Same series, opposite results, no anchor. The model recomputed from scratch, generated the same conviction tier, and got the one-run version. Variance in the conviction tier strikes again.

The other lock-pattern loss: the model’s PIT side-flip didn’t work. Yesterday: PIT +12% lock LOST. Today: model flipped to HOU +6.7% — and PIT won anyway. Same series. The model “recompute from scratch” went the other direction this time and still lost. The HOU pick is now data point #2 in the “model flips sides and loses anyway” file.

🔬 The Audit Thesis Just Got Flipped

This is the part that matters for the v0.6 audit.

Wednesday’s data:

• 🟢 Direction-AGREE actionables: 2-0 (LAD lock cashed 7-0, BOS strong cashed 8-1)

• 🔴 Direction-FLIP actionables: 0-2 (TEX lock lost, COL streak ended)

That was the data point I framed yesterday as confirming the audit thesis: “direction-agreement plays outperform direction-flip plays.”

Thursday’s data:

• 🔴 Direction-AGREE actionables: 1-5 (LAD lock lost, plus four others)

• 🟢 Direction-FLIP actionables: 1-1

Yesterday’s data contradicts the thesis I drew from Wednesday’s data.

Two-day aggregate across the actionable tier:

• AGREE: 3-5 (37.5%)

• FLIP: 1-3 (25%)

Both below breakeven. The direction-effect signal isn’t clean enough to act on yet.

This is the column’s most useful lesson of the week and the entire reason v0.6 ships now rather than after I changed the edge filter:

Two days of consistent data is not data. A 5-data-point hypothesis that gets contradicted by the next 6 data points isn’t a hypothesis — it’s noise. The v0.6 tracker exists to measure the direction effect across 200+ picks before anyone decides whether to act on it.

If I’d “fixed” the edge filter based on Wednesday’s results, today’s picks would all be wrong in a different way. Sample size is the discipline. 🧘

Subscribe now

📈 Running Ledger Through 12 Days

Sliced to 59.9% on 147 graded model picks. Trending down from the early-week 63% but still meaningfully above the 52.4% breakeven. The trajectory is doing exactly what regression to the mean looks like with growing sample size — early-window noise narrowing toward the true rate.

Where I’d guess the true rate lives: 56-62%. That’s the educated read after 12 days and 147 picks. Closer to 60 in the optimistic case, closer to 56 in the pessimistic. Even the low end is meaningfully positive over a fair coin. ⚖️

🎯 Today’s Slate — Four Locks, Three Flip, One Agree

Headline numbers:

• ⚾ 14 games on the card

• 🔒🔒🔒🔒 4 LOCKS — MIA, WSH, SEA, OAK (1 AGREE / 3 FLIP)

• 🎯 7 actionable plays (≥4%) — 4 locks + 3 strong/solid

• 🔥 Slugfest: MIL @ COL projected 11.95 r at Coors (PF 112)

• ❄️ Pitchers’ duel: NYM @ SD projected 6.19 r at Petco (PF 91)

• 🌟 Aces today: Robbie Ray (SF), Jesús Luzardo (PHI), Sonny Gray (BOS), Trey Yesavage (TOR), Michael King (SD)

The structural shape of today is the audit test made manifest. Four locks. Three are direction-FLIP (model says opposite team wins from market). One is direction-AGREE (SEA, where model and market both pick Seattle but model has them as a much bigger favorite).

Under the audit thesis from Wednesday: SEA should outperform the three FLIP locks. Under the thesis as it now stands (Thursday data): that prediction is essentially a coin flip. Today’s results will be data points 13 and 14 in the running tracker.

Share

🔒 The Four Locks

🎯 #1 — MIA +16.9% vs TB (BIGGEST MORNING EDGE) — FLIP

Model: MIA 61.5% · Market: TB 55.3% · Total est: 6.77 r · PF: 96

The biggest morning edge the column has surfaced. Direction-flip play — market has TB winning, model has MIA. Ryan Gusto’s v0.6 composite grades meaningfully ahead of Drew Rasmussen’s for today’s matchup. MIA’s offense vs RHP also rates respectably in the new wOBA-proxy framework.

Note the low projected total (6.77 r) at loanDepot park. This is a pitcher-friendly environment that amplifies pitcher-quality gaps. If Gusto really is the better arm, the gap is more decisive in a low-scoring game.

🎯 #2 — WSH +14.9% @ ARI — FLIP

Model: WSH 59.6% · Market: ARI 55.3% · Total est: 9.81 r · PF: 101

Another direction-flip. Foster Griffi vs Merrill Kelly at Chase Field. The model has the WSH starter grading well; Kelly is the name brand but the composite suggests he’s being overpriced for this matchup.

WSH on the road historically has been a fade spot for the market. The model’s edge here is the market’s anti-WSH bias plus an actual pitcher-quality gap.

🎯 #3 — SEA +14.8% @ DET — AGREE (only AGREE lock today)

Model: SEA 70.1% · Market: SEA 55.3% · Total est: 8.30 r · PF: 99

The audit’s clean test case. Model and market BOTH favor SEA. The model has them as a much bigger favorite (70% vs 55%) — magnitude disagreement, not direction. Bryan Woo vs Framber Valdez.

This is the same archetype that delivered LAD 7-0 on Wednesday (cashed) and lost on Thursday LAD 2-3 (didn’t cash). If the AGREE thesis has any predictive power at all, this is the lock that should reflect it. If it doesn’t, the thesis is officially dead. 📊

🎯 #4 — OAK +10.3% @ HOU — FLIP

Model: OAK 60.3% · Market: HOU 50.0% · Total est: 8.34 r · PF: 101

Direction-flip play. Market has HOU as a coin-flip favorite at home; model has OAK winning outright. Jack Perkins vs Peter Lambert. OAK is back on the actionable slate after being on the lock list Tuesday (won 2-1).

Subscribe now

📋 The Other Actionable Plays (4-10% edges)

The COL play is interesting. Model thinks MIL wins, market thinks MIL wins — they AGREE on direction. But the edge is on the underdog COL because the magnitude gap is small enough that the market is “overpaying” for MIL. This is a different archetype than the AGREE plays where you back the favorite. Worth tracking separately. 🧪

🚫 The Pitchers The Model ISN’T Touching

Robbie Ray (SF @ CHC) — model has SF +2.0% (pass). Direction-agreement, magnitude disagreement, no actionable edge. The model is essentially flat.

Jesús Luzardo (PHI vs CWS) — PHI +0.7%. The model agrees with the market on PHI as a heavy favorite. Nothing to publish.

Sonny Gray (BOS @ NYY) — BOS +2.3%. Lean tier. Direction-agreement (both pick NYY), small magnitude edge on BOS. Not actionable.

Discipline. Three name-brand starters going. None generate an edge worth publishing. This is what calibrated honesty looks like on an ace-light slate. 🧘

Subscribe now

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Same caveat, every day.

2. 🎰 4 locks is meaningful conviction. Across the tracking window, multi-lock days have gone: 4 locks Tuesday (2-3), 5 locks May 25 (0-5), 3 locks Wednesday (1-2). The range is wide.

3. 📊 147 picks at 59.9%. Sample is growing, the rate is moderating. Both true rates of 56% AND 62% are within the noise band.

4. 🔬 v0.6 audit is live but not yet conclusive. Two days of direction data is not enough to change the edge filter. Two months might be.

✅ What This Column Is, Twelve Days In

A daily transparent read on where the MLB betting model and Vegas disagree. Locks that cash 7-0. Locks that lose 2-3 by one run. Hypotheses confirmed by Wednesday and contradicted by Thursday. The audit tooling shipping live so the next 200 picks tell us what the first 147 couldn’t.

If you’re new to Bad Bets:

• 📩 Subscribe for daily MLB picks at 10am ET

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for JSON outputs, v0.6 internals, and ongoing audit progress

See you tomorrow. ⚾

— T&G

Share

Edge picks: 4-10 (28.6%). Worst day since Monday May 25 (which went 0-5).

By edge bucket:

• 🔒 Locks: 1-2 (LAD ✅, PIT ❌, TEX ❌)

• 💪 Strong: 1-2 (BOS ✅, MIN ❌, SEA ❌)

• 🎯 Solid: 1-2 (COL ❌, plus two lean/pass tier hits)

• 🤏 Leans: 0-4

Painful as the headline number is, the pattern inside it is the story. Yesterday’s article literally framed today as a test: “If LAD cashes and TEX doesn’t, the audit thesis gets a data point.”

The test came back with a loud answer.

Subscribe now

🎯 The Audit Thesis Just Got Confirmation

Let me split yesterday’s 7 actionable plays by structure:

🟢 Direction-AGREEMENT plays (model and market both favor the same team):

🔴 Direction-FLIP plays (model and market pick different sides):

Direction-flip plays: 0-2. The TEX lock and the COL streak-ender both lost on the day flagged as the audit test.

Within direction-agreement, the splits get more interesting. The TWO wins (LAD lock, BOS strong) were the two plays where the model thought the market was underpricing the favorite. The three losses (PIT, MIN, SEA) were all plays where the model and market agreed but the model thought the favorite was more dominant than priced — and the favorite either got upset or the variance came out the wrong way.

The cleanest read so far: when the model and market agree on direction AND the model thinks the underdog has more value than priced, that may be a different bucket than “agree on direction, model thinks favorite is bigger lock.” Worth splitting in v0.6’s audit script. 🔬

Subscribe now

🌟 The Big Wins Worth Circling

🎯 LAD +14.1% LOCK (Ohtani) → 7-0 SHUTOUT. Ohtani threw 7 scoreless. The model’s flagship “direction-agreement lock at a magnitude disagreement” played out exactly as the thesis required. This is the cleanest result the column has produced.

🎯 BOS +9.3% STRONG → 8-1 BLOWOUT. Same structural pattern (direction-agreement, magnitude edge), smaller edge, big result. The column’s “boring tier” delivered another headliner.

The losses were structural too. Skenes (PIT +12.0% lock) is now 0-2 across two attempts where the column backed him. That’s a real pattern worth noticing — when the model says “back the ace,” the ace has now lost twice in a row. Either Skenes is in a rough stretch or the model’s elite-arm composite is over-rating top-end pitchers. Logging for the audit.

MIN +7.7% getting SHUT OUT 0-8 by CWS is the bracket-busting one. Yesterday CWS was supposed to win as a +12.6% lock and lost 4-6. Today the model went MIN and they got shut out 8-0. Same series, same teams, opposite picks, both wrong. That’s variance, not signal.

The COL streak ended 11-4. Two wins → one loss in a blowout. The v0.5 wRC+ fix isn’t a perfect solution; just a less-bad one. Still better than 0-3 with locks, but worth not getting overconfident on. ⛏️

Subscribe now

📈 Running Ledger Through 11 Days

60.4% model pick accuracy across 139 graded picks. Slipped from 62.4% after yesterday’s tough day. The trajectory is still above breakeven (52.4%) but the lead is shrinking — exactly what regression to the mean looks like with growing sample.

Edge picks aggregate: Across the 6 days with full edge tracking (Wed 5/27 onward), the bucket-graded record is roughly 35-32 (52.2%). Right at breakeven. The edge filter is currently not adding real value over picking model winners directly — which is the entire reason the v0.6 audit is now top priority.

Subscribe now

🎯 Today’s Slate — Light Thursday, One Big Lock

Headline numbers:

• ⚾ 8 games — light Thursday getaway-day slate

• 🔒 1 LOCK — LAD +14.4% (second straight day)

• 🎯 3 actionable plays (≥4%) — LAD, HOU, PHI

• 🔥 Slugfest: BAL @ BOS projected 10.73 r at Fenway

• ❄️ Pitchers’ duel: SD @ PHI projected 8.94 r (Wheeler going)

• 🌟 Aces today: Wheeler (PHI), Carlos Rodón (NYY), Imanaga (CHC)

The structural pattern of today’s actionable plays is striking — all three are direction-agreement plays where the model thinks the market is mispricing the favorite. That’s the same archetype that delivered LAD 7-0 and BOS 8-1 yesterday. If the audit thesis is right, today should look more like Wednesday than Tuesday.

Subscribe now

🔒 The Lock — LAD +14.4% (Again)

#1 — Los Angeles Dodgers @ Arizona Diamondbacks

Model: LAD 71.0% · Market: LAD 56.6% · Total est: 9.19 r · PF: 101

Second straight day with LAD as the headline lock. Yesterday Ohtani threw 7 shutout innings in a 7-0 win. Today Justin Wroble is going for LAD vs Ryne Nelson for ARI — different starters, smaller absolute model probability (71% vs yesterday’s 77.6%), same +14% edge.

The model isn’t anchored to yesterday’s win. It’s recomputed from scratch with today’s pitching matchup and still landed at a lock. Wroble grades respectably in the v0.5 composite; Nelson grades meaningfully worse. ARI’s offense vs LHP also rates below average, which compounds the gap.

The direction-agreement structure is preserved: market has LAD favored, model has LAD MORE favored. This is the cleanest archetype the column has identified through 11 days, and it’s the one that pays. ⚾

📋 The Other Actionable Plays

The HOU pick is the most instructive of the slate. The model backed PIT as a lock yesterday, lost in a slugfest 9-11, and today the model is on the other side of the same matchup. Different starters, different read, different conclusion. This is what “recompute from scratch” actually looks like. No emotional attachment to yesterday’s pick. The model went HOU because the new pitching matchup says HOU.

Subscribe now

🚫 The Leans And Passes

CLE @ NYY (+3.3% lean) — Carlos Rodón going for NYY. Model has CLE as a small-edge play (Rodón fade). Not actionable but worth noting the model’s ongoing skepticism on name-brand starters when the composite says they’re being over-priced.

KC @ MIN (+1.5% pass) — MIN back on the slate after yesterday’s 0-8 shutout. Model has them as essentially a coin flip today. Discipline.

OAK @ CHC (+1.3% pass) — Imanaga going for CHC. The model that crushed PIT/Skenes also faded CHC/Imanaga on Tuesday (lost 4-7). Today Imanaga is the home arm and the model is essentially neutral. Same discipline.

Three games where a tout would publish picks and the model is essentially saying “skip.” Light Thursday slates reward discipline. 🧘

Subscribe now

🔬 v0.6 Audit Priorities — Now Sharper

After yesterday’s results, the audit’s most actionable items are:

1. 🎯 Build the direction-agreement vs direction-flip split tracker. The two days (yesterday + reverse last week) of this split data are pointing in one direction. Need 30+ data points before changing the edge filter.

2. ⚾ Backed-ace pattern (Skenes 0-2, Cole 1-1): small sample but worth a specific feature flag. If the model is systematically over-rating elite arms in the v0.5 composite, that’s a calibration issue not a luck issue.

3. 📊 Solid vs strong vs lock tier hit rates aren’t behaving cleanly — sometimes solid outperforms strong, sometimes the reverse. Threshold tuning is likely warranted.

4. ⛏️ COL audit not yet closed. 2-1 across recent road plays. Better than 0-3 but not yet “fixed.”

Audit script work begins this week. 🛠️

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Same caveat, two weeks in.

2. 🎰 Yesterday was -2σ for the actionable tier. Don’t panic. Don’t change anything based on one day.

3. 📊 139 picks at 60.4%. Sample is growing, the rate is moderating. Still above breakeven.

4. 🧘 Light slate = discipline day. Three actionable plays. Resist the urge to fill the slate with lean-tier picks that the model is barely separating from the market.

Share

✅ What This Column Is, 11 Days In

A daily transparent read on where the MLB betting model and Vegas disagree. Locks that cash 7-0. Strong edges that cash 8-1. Direction-flip locks that lose. COL streaks that end. Skenes attempts that go 0-2. Honest accounting that lets the audit work happen in public.

If you’re new to Bad Bets:

• 📩 Subscribe for daily MLB picks at 10am ET

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for JSON outputs, v0.5 internals, and v0.6 audit progress

See you tomorrow. ⚾

— T&G

Subscribe now

Five locks went 2-3. Edge plays went 5-5. Variance ate the rest.

Two of the three lock losses were ONE-RUN GAMES. KC and ARI both lost by a single run in games where the model was directionally correct on the favorite. That’s variance — the model’s reads were close, the games tilted the wrong way on a single play. This is exactly what the “high-variance lock tier” looks like in practice.

The other actionable plays (the 4-10% edges):

Strong/solid edges: 3-2 (60%). The “boring tier” delivered again.

Totals: 🔒 Locks 2-3 · 💪 Strong 3-2 · Edge plays 5-5 (50%) · Model picks 8-6 (57.1%).

🔁 The COL Story Keeps Cashing

Second straight day of COL +5-10% on the road cashing.

• 🟩 Monday: COL +5.7% @ LAA → COL 9, LAA 8 (wild win)

• 🟩 Tuesday: COL +9.8% @ LAA → COL 8, LAA 2 (blowout)

Three weeks ago COL was the column’s biggest model headache — three straight road locks blown out, outscored 29-10. The v0.5 fix (true wRC+ from FanGraphs replacing the OPS proxy) was the suspected solution. Two consecutive cashes is enough data to start saying “the fix held.” Not enough to close the audit, but enough to move COL off the priority list. 📉➡️📈

Leave a comment

📈 Running Ledger Through 10 Days

125 graded picks. 62.4% model accuracy. 10 points above breakeven. Sample size keeps growing, the rate keeps holding. By 200 picks we’ll have something more confident to say. 📊

The lock tier: Now 5-8 across the tracking window (38.5%). Improved from last week’s 28.6% but still below where the model directionally is. The lock variance band is wide — last week 2-5, this week 3-3 across two attempts. Long-run rate likely between 35-55%. Don’t over-bet locks. Treat them as the highest-variance tier they appear to be.

🎯 Today’s Slate — Three Locks, Three Big Names

Headline numbers:

• ⚾ 14 games on the card

• 🔒🔒🔒 3 LOCKS — LAD with Ohtani · PIT with Skenes · TEX direction-flip

• 🎯 7 actionable plays (≥4%) — 3 locks + 4 strong edges

• 🔥 Slugfest: CWS @ MIN projected 10.24 r (highest total)

• ❄️ Pitchers’ duel: CLE @ NYY projected 6.89 r (Gerrit Cole going)

• 🌟 Aces going today: Ohtani (LAD), Skenes (PIT), Cole (NYY), Gallen (ARI), Buehler (SD), Nola (PHI), Gausman (TOR), Kirby (SEA), Peralta (NYM)

Nine big-name starters going today. This is one of those slates where the action is concentrated in the pitcher matchups themselves — and the model is treating it with calibrated aggression. Three locks. None are 16% gaps. None are direction-agreement walkovers. All three need their pitcher to deliver.

Subscribe now

🔒 The Three Locks

🎯 #1 — LAD +14.1% @ ARI (OHTANI ON THE MOUND)

Model: LAD 77.6% · Market: LAD 63.6% · Total est: 8.45 r · PF: 101

Shohei Ohtani vs Zac Gallen. Model and market agree LAD wins; the disagreement is how much. Market has LAD as a 63.6% favorite, the model has them at 77.6%. That 14% gap on a magnitude-only edge is exactly the “direction-agreement” pattern that’s been the column’s most reliable bucket.

The math: Ohtani’s v0.5 real-FIP composite grades as elite (top 5 of any starter pulled this season). Gallen has been good this year — but the model has the gap between them wider than the market’s pricing. ARI’s offense against LHP also grades below average, which compounds the gap.

This is the highest-conviction direction-agreement lock the column has surfaced. If the audit’s emerging thesis is right (”direction-agreement edges outperform direction-flip locks”), this is the day it should show up most clearly. 📊

🎯 #2 — PIT +12.0% @ HOU (SKENES BACK AT IT)

Model: PIT 70.7% · Market: PIT 58.6% · Total est: 8.51 r · PF: 101

Paul Skenes. Round 2. The first time the column backed Skenes (Thursday May 28 — PIT +4.5% solid), he lost to Imanaga and the CHC offense (7-2). The model is backing him again today at a much bigger edge, which is the same recompute-from-scratch pattern we’ve seen on LAA, KC, and TOR.

The matchup math: Skenes vs Spencer Arrighetti. Real-FIP composite has Skenes as one of the three best arms in baseball; Arrighetti is league average. The model thinks the market is under-pricing the gap. If the model is right, Skenes outdueling a back-end Astros rotation is the path to a quiet 4-2 PIT road win.

🎯 #3 — TEX +12.0% @ STL (DIRECTION FLIP)

Model: TEX 60.7% · Market: STL 51.3% · Total est: 8.70 r · PF: 100

MacKenzie Gore vs Andre Pallante. The third lock and the only direction-flip of the three. Market has St. Louis as a slight home favorite. The model has Texas winning outright. The gap comes from Gore’s K-BB% grading meaningfully ahead of Pallante’s in the v0.5 composite, plus Texas’s offense vs RHP being respectable.

Caveat: direction-flip locks are the audit’s most suspect category. Last week the bucket-flip tier was the one underperforming most. Today is another data point. Sized smaller than LAD or PIT, despite the same edge magnitude. 🤔

Leave a comment

📋 The Strong Edges (4-10%)

The BOS pick is worth a sentence. Direction-agreement (both model and market favor BOS), magnitude disagreement, at Fenway in a hitter park. This is the same archetype as the LAD +14.1% lock — just smaller. If you’re betting one strong play, this is structurally the cleanest read.

🚫 The Pitchers The Model ISN’T Touching

Aaron Nola for PHI — model has SD +2.6% (lean). Joins the previous Cole / Skenes / Yamamoto fades. The ace-fade thesis went 0-3 last week. Today the model is small-edging Nola but not making it actionable. Discipline. 🧘

Gerrit Cole for NYY — model has NYY +2.9% (lean). Model and market agree NYY wins, but the gap is too small to publish. Cole’s last start cashed for the column on a NYY pick — today the model is essentially neutral.

Kevin Gausman for TOR — model has TOR +2.4% (lean) vs Grant Holmes. No actionable edge.

This is what discipline looks like on a name-brand pitcher slate. Three aces going. Only two of them generate actionable edges (Ohtani as a back, Skenes as a back). The rest get the lean/pass treatment.

Subscribe now

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Same caveat, every day.

2. 🎰 Three locks is meaningful conviction. Locks have gone 2-3 (yesterday), 0-5 (May 25), and 1-0 (Mon Jun 1) on multi-lock days. Range is wide.

3. 📊 125 picks at 62.4%. Sample is growing. The 90-day decision point is closer than the relaunch.

4. 🔬 v0.6 audit script still pending. Today’s slate is a clean test: 1 direction-agreement lock (LAD), 1 direction-agreement-by-pitcher (PIT/Skenes), 1 direction-flip (TEX). If LAD cashes and TEX doesn’t, the audit thesis gets a data point.

✅ What This Column Is, Ten Days In

A daily transparent read on where the MLB betting model and Vegas disagree. Locks that cash 2-1 on the last play. Locks that lose by one run in three different games on the same night. COL road games that finally cash, two in a row, after a brutal opening week. The honest accounting that lets you check the work.

If you’re new to Bad Bets:

• 📩 Subscribe for daily MLB picks at 10am ET

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for JSON outputs, v0.5 internals, and v0.6 audit progress

See you tomorrow. ⚾

— T&G

Subscribe now

Edge picks: 3-1 (75%) on actionable plays. Locks: 1-0. The audit conversation got immediate good news.

Three wins, two of them blowouts, one of them a wild 10-9 lock that swung on the last pitch of the night.

By the post-game bucket grader (the audit version that uses closing line edges):

• 🔒 Locks: 1-0 (100%)

• 💪 Strong: 1-0 (100%)

• 🎯 Solid: 3-0 (100%) — the “boring is winning” thesis continues 🤓

• 🤏 Leans: 1-2 (33.3%)

• Total graded: 6-3 (66.7%)

The COL edge cashing is the data point I’d circle. Three straight days of COL road losses opened the column’s first audit item back on May 24-26. The v0.5 fix (true wRC+ from FanGraphs replacing the OPS proxy) was the suspected solution. Monday was the first live test where the model surfaced a meaningful COL road edge AFTER the fix shipped — and COL won 9-8 in a wild one. N=1 doesn’t prove a fix worked. But it doesn’t disprove it either. Logging.

The lock cashing is the OTHER data point worth circling. Going into Monday the lock tier was 2-5 across the 8-day tracking window (28.6%). DET +14% cashing pushed it to 3-5 (37.5%). KC’s +8.6% strong edge cashing in a 9-2 blowout also feels relevant to the audit. The directional-disagreement plays (where model and market disagree on who wins outright) were largely the winners yesterday. Good day for the edge filter. 📈

The only loss: SF +4.3% at MIL. Brewers won 16-2. When the model is wrong on a direction-flip play, it tends to be wrong loudly. Same pattern we saw Sunday.

Subscribe now

📈 Running Ledger Through 9 Days

Up to 63.1% on model picks across 111 graded games. That’s an 11-game series of 66.7% to close the week. Slightly more meaningful sample size now — still below where we’d need it for confident claims (200+) but the trend is consistently above breakeven. 📊

🎯 Today’s Slate — FIVE LOCKS

Headline numbers:

• ⚾ 14 games on the card

• 🔒🔒🔒🔒🔒 5 LOCKS — the most since Monday May 25 (which famously went 0-5)

• 🎯 10 actionable plays (≥4%) — 5 locks + 5 strong edges

• 🔥 Slugfest: PIT @ HOU projected 10.83 r (highest total of the slate)

• ❄️ Pitchers’ duel: SF @ MIL projected 6.86 r (McDonough vs Harrison)

• 🌟 Big names going: Aaron Nola (PHI), Kevin Gausman (TOR), Dustin May (STL), Eric Lauer (LAD)

This is the most concentrated conviction slate the column has surfaced since the relaunch nine days ago. After a full week of 0-1 locks per day, the model is now finding huge gaps with the market all over the board.

The cautious read: the last time a 5-lock day happened (Monday May 25), all five lost. The column went 4-9 that day. The memory is fresh.

The hopeful read: that 0-5 day was with v0.2’s OPS-derived offense proxy. Today’s v0.5 model has true wRC+ from FanGraphs, real FIP, weather, and Shin de-vig. The math is sharper. The reads should be sharper too.

The honest read: we don’t know. Five locks is a lot. The variance band on a 5-lock day is enormous. Could easily go 5-0 (column makes the front page). Could easily go 0-5 (column eats its second 0-5 day). Most likely outcome: 3-2 or 2-3. Size accordingly.

Subscribe now

🔒 The Five Locks

🎯 #1 — OAK +16.0% @ CHC (BIGGEST MORNING EDGE OF THE RUN)

Model: OAK 62.7% · Market: CHC 53.4% · Total est: 9.36 r · PF: 102

The biggest morning-edge play the column has surfaced. Direction-flip — model has OAK winning outright, market has CHC. Gage Jump vs Jameson Taillon. The model’s real-FIP composite has Jump grading much closer to Taillon than market pricing suggests, and OAK’s offense vs RHP has been better than the headline record indicates.

🎯 #2 — CWS +12.6% @ MIN (THE PET TEAM RETURNS)

Model: CWS 65.2% · Market: CWS 52.6% · Total est: 8.31 r · PF: 103

CWS is the model’s defining team. They were the right call on Tue/Wed/Thu of last week (15-2 blowout, 6-2 win). Today they’re back at +12.6% — Davis Martin vs Connor Prielipp at Target Field. The 12.6% gap is the model saying market still under-prices Davis Martin against MIN’s offense. We’ll see.

🎯 #3 — MIA +12.5% @ WSH

Model: MIA 60.3% · Market: WSH 52.3% · Total est: 9.48 r · PF: 100

Lake Bachar vs Miles Mikolas at Nationals Park. Direction-flip — market has WSH, model has MIA. Bachar’s real-FIP grades favorably; Mikolas’s veteran name brand may be over-priced. Contrarian conviction play.

🎯 #4 — KC +12.0% @ CIN (🔁 BACK-TO-BACK LOCK)

Model: KC 58.5% · Market: CIN 53.5% · Total est: 10.35 r · PF: 109

KC’s second straight day as a strong-or-better play. Yesterday they cashed 9-2 as an +8.6% strong edge. Today they’re a +12.0% LOCK against a different CIN starter (Noah Cameron vs Andrew Abbott). Same series, same park, bigger edge. This is exactly the “the model recomputes from scratch and the number went up” pattern we saw with LAA last week.

🎯 #5 — ARI +11.3% vs LAD (LAUER FADE)

Model: ARI 58.7% · Market: LAD 52.6%

Direction-flip play — market has LAD (Eric Lauer), model has ARI (Michael Soroka). The 5th lock of the day. The model is fading Eric Lauer specifically — his real-FIP and K-BB% in the v0.5 composite read meaningfully worse than the market is pricing.

📋 The Strong Edges (4-10%)

The NYY pick is the most underrated of this group. It’s the only actionable play today where the model and market AGREE on direction. The edge is purely about magnitude (75.8% vs 68.7%). Direction-agreement edges have been the column’s most reliable bucket this week — they’re the “boring solid” tier the column kept flagging Sunday and Monday. If only one of today’s 10 picks were sized larger, the case for NYY is the cleanest.

Share

🔁 COL Update

COL is on the slate AGAIN — same LAA series, second straight day (Monday COL +5.7% cashed 9-8, today COL +9.8%).

Same arguments as Monday’s note apply: this is a real-time test of whether v0.5’s true wRC+ fix solved the COL road-bias issue from the May 24-26 streak. Monday’s win was the first positive data point. A second straight win today would make it two consecutive — small sample, but directionally meaningful. The Sugano vs Rodriguez matchup is one the model rates very differently from the market. ⛏️

Subscribe now

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Same caveat, two weeks in.

2. 🎰 Five locks is a LOT. The last 5-lock day went 0-5. Don’t assume the model has cracked the code — assume variance is the dominant factor and size accordingly.

3. 📊 111 picks. 63.1% model accuracy. Still inside the noise band but with shrinking error bars. By 200 picks the true rate becomes more knowable.

4. 🔬 v0.6 audit still pending. The split-tier backtest script (separating direction-flip vs direction-agreement edges) is the next priority. Today’s slate is a perfect dataset — 5 direction-flip locks + 1 direction-agreement strong (NYY) gives clean signal.

✅ What This Column Is, Nine Days In

A daily transparent read on where the MLB betting model and Vegas disagree. Locks that cash 10-9 in walkoffs. Locks that get shut out 8-0. COL roads that finally win after losing three in a row. Five-lock slates that history says will be a coin flip on the outcome.

The honesty is the product. Today the model surfaced more conviction than any day since the relaunch. We’ll find out by midnight whether that conviction was warranted. ⚾

If you’re new to Bad Bets:

• 📩 Subscribe for daily MLB picks at 10am ET

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for JSON outputs, v0.5 internals, and v0.6 audit progress

See you tomorrow. 🫡

— T&G

Subscribe now

Let’s not dress it up. Edge picks went 3-6 (33%). Model picks went 10-5 (66.7%). The split tells the whole story.

Three blowouts AGAINST our picks. PHI lost 1-9. MIA lost 1-10. MIN lost 3-9. When variance hits, it doesn’t hit politely.

But here’s the interesting part — the model picks were still 10-5. The model identified the direction correctly on two-thirds of the slate. The losses were concentrated almost entirely in the games where the model and market disagreed enough to publish an edge. Same pattern. Different teams. Different week.

Leave a comment

🔬 The Edge Audit: What 30 Edge Plays Are Telling Us

After eight days, the data has a pattern that demands attention:

The model is meaningfully above breakeven (52.4%) on direction. But the edge filter is currently making our pick quality WORSE, not better. That’s the puzzle.

This is v0.6’s audit priority. Here’s the short list of suspects:

1. 🎯 Shin de-vig may be over-correcting on heavy favorites. If de-vigged market prices systematically under-price 60%+ favorites, the model looks like it has edge when it doesn’t.

2. 📊 v0.5’s true wRC+ is using ~50 games of current-season data. Hot starters look better than they are; cold starters worse. A blend of 0.4 × current + 0.6 × prior might calibrate better until July.

3. ⚾ The 58/42 SP/bullpen blend was calibrated for v0.2’s ERA-derived composite. With real FIP now sharper, the SP weight may need to increase (~65/35).

4. 📏 The 4% edge threshold came from intuition, not backtesting. Maybe the true threshold is 5%. Maybe it’s 6%. We’ll know when we run the cutoff sensitivity analysis.

5. 🤔 Directional-disagreement plays may be inherently noisier. When model and market disagree on who wins, the model is taking the variance side by definition.

The fix isn’t to panic. It’s to build the audit script, find the slice that’s leaking, make ONE targeted change, and watch another 30-50 picks. Changing five things at once means we never know which one helped.

v0.6 roadmap is officially open. Audit script first, changes second. 🛠️

Leave a comment

📈 Running Ledger Through 8 Days

Model-pick accuracy holds steady. The leaderboard column to watch going forward is the edge column — and whether the v0.6 audit moves that number above 50%. ⚖️

Share

🎯 Today’s Slate

Monday MLB picks at a glance:

• ⚾ 8 games — light Monday slate (it’s a getaway day for most teams)

• 🔒 1 LOCK — DET +14.0% at TB · first lock in four days

• 🎯 4 actionable plays (≥4%) — DET, KC, COL, SF

• 🔥 Slugfest: COL @ LAA projected 10.12 r (Freeland on the mound)

• ❄️ Pitchers’ duel: NYM @ SEA projected 7.63 r (Hancock at T-Mobile)

• 🌟 Big name watch: deGrom going for TEX, Yamamoto fresh off shutting down PHI Sunday

The thin Monday slate is what it is. But after three straight days with zero locks, the model surfaced something with conviction today — and it’s the biggest edge since the NYY +17.2% lock that cashed on May 26.

Share

🎯 The Lock — DET +14.0% @ TB

#1 — Detroit Tigers @ Tampa Bay Rays

Model: DET 54.5% · Market: TB 59.6% · Total est: 8.32 r · PF: 93 · WX: dome

This is the biggest edge the model has surfaced in 6 days, and the matchup itself is fascinating.

Tampa Bay is priced as a near-60% favorite at home. The model has Detroit at 54.5% — a meaningful direction flip, not just a magnitude disagreement. The market thinks TB wins outright. The model thinks DET wins outright. That’s the kind of gap that either pays huge or reveals a model flaw immediately.

Why the gap, in plain English:

• ⚾ Ty Madden’s real FIP grades meaningfully ahead of Griffin Jax’s in the v0.5 composite. The market is pricing Jax as the better arm by a wider margin than the underlying numbers support.

• 🏟️ Tropicana, PF 93, dome. Low-scoring environment that amplifies pitcher-quality differences. If Madden truly is the better arm by even a half-run per nine, that gap is more decisive in a 7-run game than a 10-run game.

• 📊 DET’s offense vs RHP grades respectable in the splits the model is pulling. Not elite, but not the punchless lineup the market seems to be pricing.

The honest caveat after Sunday: the edge tier has been 0-5 on locks in 6 of 7 attempts since May 24. Six picks before this one have all looked like reasonable reads. Five lost. The seventh attempt today is the biggest gap yet — which means either the model has unique insight, or this is exactly the kind of disagreement that’s been costing us.

Verdict: Today’s only lock. Biggest edge in six days. The cleanest test yet of whether the conviction tier is broken or just unlucky. ⚾

Subscribe now

📋 Honorable Mentions (4-10% Edge Plays)

Share

🔁 The COL Note

After three straight days of COL +10%+ road losses (May 24-26), the model has surfaced COL on the road again — but at a much smaller edge (+5.7%) and against a non-elite opponent.

The trajectory of COL road plays since the relaunch:

• 🟥 Sun-Tue: 0-3 (all locks, all blowouts, outscored 29-10)

• ⚪ Wed-Sat: No actionable COL plays (model correctly priced them as coin flips)

• 🟦 Today: COL +5.7% @ LAA

This is the cleanest test yet of whether v0.5’s switch to true wRC+ from FanGraphs (replacing the OPS proxy) actually fixed the COL road-bias issue. If COL is genuinely undervalued at +5.7%, the model has learned. If they get smoked again, the audit list grows. ⛏️

Subscribe now

🌟 The deGrom Watch

deGrom going for TEX at STL tonight — model has TEX +0.6% (pass tier). The model isn’t generating an edge despite the name brand.

Reading between the lines: the market has correctly priced deGrom this time, and the model agrees. This is actually the correct outcome for the “fade name-brand aces” pattern that lost 0-3 across the week (Cole, Skenes, then today Yamamoto’s domination). When model and market agree, there’s no edge to publish. Discipline. 🧘

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Same caveat, new month.

2. 🔧 v0.6 audit is now the top priority. The edge filter has been net-negative across 30 plays. That’s actionable signal.

3. 🎲 8-game Monday means small sample. Don’t force action on top of the four edges. The volume isn’t there today.

4. 📊 102 picks in. Model-pick accuracy is real (62.7%). Edge-play accuracy is the puzzle (~43%). Two different stories in one column.

Leave a comment

✅ What This Column Is, Two Days Into Week Two

A daily transparent read on where the MLB betting model and Vegas disagree. Wins, losses, blowouts, the occasional 3-6 edge-day disaster, and the kind of honest accounting that says “the audit data shows the edge filter isn’t working — here’s the plan.”

Tout services double down when they’re losing. Quant columns publish the audit.

If you’re new to Bad Bets:

• 📩 Subscribe for daily MLB picks at 10am ET

• 🐦 Follow @TradesAndGains for daily slate graphics

• 💬 Discord for JSON outputs, v0.5 internals, and v0.6 audit progress

See you tomorrow. ⚾

— T&G

Share

Today marks one week since Bad Bets relaunched (May 24). Seven days of daily publishing, seven slate graphics, ~10,000 words, one Skenes loss, three COL losses, one wind-boosted lock that didn’t survive a 5-7 final, and one 14-3 LAA blowout that arrived after two straight days of the same pick losing.

The math is honest. The honesty is the product. Let me show you week one. 📊

Subscribe now

📊 From The Ledger: Saturday Bounced Back HARD

After Friday’s 1-4 disaster, Saturday delivered: edge picks went 4-1 (80%).

Three blowouts in the four wins. This is what it looks like when the model gets the direction right and the variance cooperates simultaneously. LAA scoring 14 runs at the Trop was particularly satisfying — same team the model loved on Friday and lost on, back at the same park, against a different starter, with a bigger edge, and they delivered a season’s worth of run support in one game.

Model picks across all 15 games: 10-5 (66.7%). Solid all-around day.

Subscribe now

🔁 The LAA Story (And Why It Matters)

The Angels are now the column’s defining narrative through one week, and not for the reasons anyone expected.

Three straight days on LAA across two losing weeks of the season:

• 🟥 Friday: LAA +4.9% @ TB → TB 8, LAA 5 (loss)

• 🟩 Saturday: LAA +7.5% @ TB → LAA 14, TB 3 (massive win, edge increased after the loss)

• 🟦 Today: LAA +5.6% @ TB (round 4 — same series, edge trimmed)

This is the cleanest demonstration so far of what the model is actually doing. It’s not anchored to LAA. It’s recomputing the edge from scratch every day based on the day’s pitching matchup, lineup splits, park, and weather. Friday’s loss didn’t make the model gun-shy. Saturday’s blowout didn’t make the model overconfident. The number went 4.9% → 7.5% → 5.6% because the math went there. Three different pitching matchups, three different edges.

If you only watched the wins, you’d think the model was a genius. If you only watched Friday, you’d think it was broken. The actual story is just: it published the number every day. ⚾

📈 Week One Ledger

62.1% over a one-week sample of 87 picks. 9.7 points above the -110 breakeven of 52.4%. Still inside the noise band — the standard error on 87 picks is around 5%, so the true rate is somewhere between 57% and 67%. That range includes “real edge” and “lucky week.” We need a lot more data. 📊

Locks specifically: 2-5 across the week (28.6%). The high-conviction tier remains the column’s biggest puzzle — small sample, high variance, currently underperforming. Whether that’s structural or random, we don’t know yet.

Subscribe now

🎓 Five Things Learned In Week One

1. 🎲 Locks are a rollercoaster. Sunday’s CLE lock landed. Monday all five locks lost. ATL lock got shut out Wednesday. WSH wind-boosted lock lost Friday. NYY lock cashed Tuesday. The ≥10% tier is feast-or-famine in a way the 4-6% solid tier isn’t.

2. 📊 Boring is winning. Solid edges (4-6%) have outperformed expectations across the week. Quietly steady, not headline-making, but the actual signal.

3. 🌬️ Weather module shipped with mixed results. Friday’s wind-boosted lock lost. Saturday’s wind-suppressed lock (smaller edge) won. One week is nothing for evaluating a feature designed to improve over thousands of games.

4. ⛏️ COL on the road is a real model issue. Three straight road losses on COL +10%+ edges Sun-Tue. Then COL won at home Saturday with no edge from the model. The audit suggests the OPS-derived offense proxy was overrating Coors hitters in road games — v0.5 fixed this with true wRC+, and the model is no longer surfacing false COL locks. Probably solved. Will keep watching.

5. 🤨 Backing AND fading aces are both real plays. The model fades Cole (lost), backs Skenes (lost), today backs PHI against Yamamoto. The nuance is the math, not the brand of the pitcher.

Leave a comment

🎯 Today’s Slate

Headline numbers:

• ⚾ 15 games on the card

• 🔒 0 locks — third straight day with no ≥10% edge

• 🎯 9 actionable plays (≥4%) — biggest action slate since the relaunch Sunday

• 🔥 Slugfest: ATL @ CIN projected 11.18 r at Great American (Strider going)

• ❄️ Pitchers’ duel: BOS @ CLE projected 8.39 r (Ranger Suarez vs Bibee)

• 🌟 Big names going: Yamamoto (LAD), Strider (ATL), Robbie Ray (SF), Wacha (KC), Bibee (CLE)

Nine actionable edges is the most action this column has surfaced in a week. The model is finding small disagreements all over the slate — but interestingly, no big ones. The “0 locks for three straight days” pattern continues. This is the kind of slate where you bet small, bet many, and let the volume do its work.

Share

The Top Four 🎯

🎯 #1 — TOR +8.4% @ BAL (🔁 RERUN)

Model: TOR 52.9% · Market: BAL 55.5% · Total est: 9.16 r · PF: 102 · WX: 1.01

Top edge of the slate, and the second time this series TOR has been the play (TOR lost 5-6 Saturday). Today’s matchup is Spencer Miles vs Kyle Bradish, and the model’s pitcher composite has a meaningful gap in TOR’s favor. Edge went up from 5.2% to 8.4% after the loss, same as LAA last week — that’s the model recomputing, not anchoring. The lesson: when the same team appears again with a bigger edge, you trust the math more, not less.

🎯 #2 — PHI +7.1% @ LAD vs YAMAMOTO

Model: LAD 61.0% · Market: LAD 68.1% · Total est: 9.20 r · PF: 97 · WX: 1.04

The ace-pricing play. Yamamoto on the mound for LAD. Market has the Dodgers at 68% — pricing Yamamoto as an elite road favorite. Model has LAD at 61% — still favored but meaningfully less so. The 7.1% edge on PHI is the model saying the market overpays for Japanese ace name brand, especially at home, especially in a slugfest park. PHI countering with Painter (real FIP grade respectable but not elite).

This is the same kind of fade we ran on Cole on Wednesday — which lost. Two data points later, the thesis is now 0-1. Today is data point three. Variance permitting.

🎯 #3 — SEA +6.9% vs AZ

Model: SEA 63.6% · Market: SEA 56.7% · Total est: 8.65 r · PF: 92 · WX: dome

Boring home favorite at T-Mobile (dome, PF 92, low-scoring park). Bryce Miller’s composite grades meaningfully ahead of Merrill Kelly’s for Arizona. Market has Seattle as the slight favorite; model has them as a clearer one. This is exactly the “boring solid edge” the column kept flagging during the week as the most profitable bucket.

🎯 #4 — LAA +5.6% @ TB (DAY 4)

Model: TB 59.5% · Market: TB 65.1% · Total est: 8.98 r · PF: 93 · WX: dome

The fourth straight day with LAA as an actionable play. Edge trimmed from yesterday’s +7.5% — McClanahan is a tougher matchup than Rasmussen, so the model dialed back. Discipline call: after yesterday’s 14-3 blowout, the temptation to chase is real. Resist it. Size to today’s edge, not yesterday’s result.

📋 Five More Honorable Mentions (4-5% edges)

The MIL flip is worth noticing. Friday HOU was a +5.3% edge (HOU won 9-2). Today the model has MIL +4.4% in the same series at the same park. Same series, opposite side. That’s because the pitching matchup completely changed — Misiorowski for MIL today is a much more respected arm by the composite than Sproa (Friday’s starter), and Imai for HOU is a much weaker arm than Lambert (Friday’s starter). The model isn’t taking sides on franchises. It’s taking sides on the day’s matchup.

Leave a comment

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Same caveat, week two.

2. 📊 87 picks in. Still inside the noise band. A 62% rate over one week could be skill or could be luck. By Day 30 we’ll know more. By Day 90 we’ll really know.

3. 🤨 Three straight days with 0 locks is worth noticing. The model has been progressively more conservative on conviction. Either the market got sharper (likely) or v0.5’s more honest math is correctly refusing to overpay (also likely). Probably both.

4. 🎲 Nine plays at 4-9% edges is a “let volume work” day. No single play is the headline; the basket is the bet.

Share

✅ What This Column Is, One Week In

A daily transparent read on where the model and market disagree. Wins, losses, weather modules that miss their first big game, blowouts that vindicate a streak, and the kind of honest accounting that says “the locks have been 2-5 this week, we don’t know if that’s real, here’s the data anyway.”

Week one is in the books. Onto week two. 🫡

If you’re new:

• 📩 Subscribe for daily Bad Bets at 10am ET

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for JSON outputs and v0.5 model internals

See you tomorrow. ⚾

— T&G

Subscribe now

Let’s get the bad news out of the way: edge picks went 1-4 (20%).

TEX +6.2% was the only winner, and it was a 9-1 blowout. The other four lost by a combined six runs across three one-run games and a two-run loss. Variance was unkind. The wind-boosted WSH lock — the headline play of v0.5’s launch day — lost 5-7.

The model PICK accuracy was actually fine: 8-6 (57.1%) across all 14 games. The split between model accuracy and edge-play accuracy is the painful one to sit with. The model identified the direction correctly more than half the time, but the games where we had a real edge were also the games where the variance hit hardest. That happens. It will happen again.

Subscribe now

🌬️ Weather Module: Day 1 Was A Miss

Here’s the awkward truth: v0.5’s flagship feature was on the wrong side of yesterday’s biggest loss.

The WSH lock at +10.9% was the model’s biggest edge of the day in large part because the WX 1.08 multiplier (wind blowing out at Nationals Park) added expected runs to the WSH side. The model thought weather would amplify Washington’s offense. It didn’t. SD scored 7. WSH scored 5. Wind blew, runs happened, the wrong team got them.

What this means for the weather module:

1. 🎯 It means nothing for the module’s validity. One data point against a feature designed to improve over thousands of games is not a signal. It’s an N=1 anecdote.

2. 🤨 It does mean we shouldn’t overweight WX 1.05+ games as “automatic upgrades.” The module is providing additional information, not certainty. A weather boost is a small factor that gets blended with everything else. Yesterday it was the visible factor on the headline play, which made the loss feel like a weather-module loss. It wasn’t. It was a baseball loss.

3. 📊 Real verdict comes in 90 days. When we have a sample of 80-120 WX-affected games, we’ll know whether the module is adding signal or just noise. Today we know nothing.

The honest version: the module shipped, the headline play lost, and that proves nothing either way. ⚖️

Subscribe now

📈 Running Ledger Through 6 Days

Still 8.7 points above breakeven on model picks. But the lock tier specifically is now 2-5 over the tracked window. Locks have a 28.6% hit rate. That’s worse than a coin flip. Either the high-edge tier is genuinely broken or the sample is too small to mean anything — and at 7 lock picks total, the latter is the boring-but-correct answer.

Don’t extrapolate. Don’t update priors yet. Keep publishing. 📊

Subscribe now

🎯 Today’s Slate

Headline numbers:

• ⚾ 15 games on the card

• 🔒 0 locks — second straight day with no ≥10% edge

• 🎯 5 actionable plays (≥4%) — LAA, HOU, TOR, ATL, WSH

• 🔥 Slugfest: ATL @ CIN projected 11.31 r at Great American (PF 109)

• ❄️ Pitchers’ duel: BOS @ CLE projected 7.36 r (Sonny Gray going)

• 🌬️ Weather watch: four games with WX 0.93-0.95 (wind blowing in), no boosts above 1.06

It’s another quiet conviction day. The model isn’t seeing big disagreements with the market anywhere — and after Friday’s blowup, “quiet” is fine.

Three repeat plays from yesterday’s slate appear again today: LAA, WSH, ATL. Each is a small lesson in how the model thinks about a team after losing on it.

Share

🔁 The Repeat Plays

This is the interesting structural feature of today’s slate. After yesterday’s painful edge results, three of yesterday’s actionable teams show up again with different numbers:

LAA → was +4.9% yesterday (lost 5-8), is +7.5% today

The model’s edge on LAA increased despite the loss. That’s because today is Detmers vs Rasmussen at Tropicana; yesterday was Ureña vs Martinez. Different pitchers, different number. The model isn’t doing emotional damage-control — it’s recomputing from scratch and landing on a bigger LAA edge today than yesterday. The “rerun” isn’t loyalty. It’s math.

WSH → was +10.9% yesterday (lost 5-7), is +4.5% today

Yesterday’s lock becomes today’s small edge. The model trimmed the conviction significantly: 10.9% → 4.5%. Why? Different pitcher matchup (King/Griffin today vs Giolito/Schultz yesterday) AND no weather boost (WX 0.95 today vs 1.08 yesterday). The model is not doubling down on yesterday’s miss. It’s reading a much smaller disagreement with the market. This is the model behaving exactly the way you’d want it to.

ATL → was won 8-3 yesterday, is +5.0% today again

Different opponent matchup setup but same series at Great American. The slugfest projection (11.31 r) is real — PF 109 + ATL’s offense + Singer’s run-allowing profile. If the model is going to keep flagging ATL at Cincinnati, watch the over.

Subscribe now

The Top Edge 🎯

🎯 #1 — LAA +7.5% @ TB

Model: TB 52.2% · Market: TB 59.7% · Total est: 8.03 r · PF: 93 · WX: dome

The market has Tampa Bay as a near-60% favorite at home. The model has them at 52.2% — basically a coin flip with a small TB lean. That 7.5% gap is the largest on the board today.

Why the gap: Reid Detmers grades much better in the v0.5 real-FIP composite than the market is pricing. The K-BB% number specifically is in the top quartile for active starters. Rasmussen for TB is a solid arm, but the model doesn’t see the gulf between the two pitchers that the market does.

Park caution: Tropicana, PF 93. Low-scoring environment. The 8.03 total puts this in the bottom third of the slate. If the LAA edge is real, the path is win a low-scoring game on the road — which is harder than winning a slugfest because one bad inning is fatal.

Verdict: Today’s top edge. Smaller edge than Friday’s blown lock, but the cleanest read on the slate. ⚾

Subscribe now

📋 Honorable Mentions

The TOR pick is the most underrated of the group. Wind blowing in at Camden Yards turns a hitter park into something close to neutral. The model captures that via WX 0.94 — a 6% run suppression — and the lower projected total favors the team with the better starter. The market doesn’t seem to be pricing the wind yet.

Leave a comment

🔁 COL Watch Update

SF @ COL is on the slate today (SF +1.1%, pass tier).

COL won at home yesterday 8-6, so the road-loss streak from last week is officially over — and the model is showing essentially zero edge today (1.1% SF). That’s the right behavior: when COL is in their natural Coors environment, the model treats it as a near-coin-flip and doesn’t generate false locks.

The road-game audit is still on the v0.6 list, but the data this week suggests the issue might have been concentrated in a 3-game variance window rather than a structural model bug. We’ll keep watching.

Share

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Period.

2. 🌬️ v0.5 had a rough first 24 hours. The flagship feature was on the wrong side of the biggest loss. That’s variance, not a verdict on the module.

3. 🎲 72 picks. Still nothing. A 61% model-pick rate is meaningfully above breakeven but inside the noise band for a sample this small. By 200 picks we’ll know more.

4. 🤨 Two straight days without a lock means the model is being conservative. That’s good. The right move on quiet days is to respect the quiet, not force action.

Leave a comment

✅ What This Column Is

A daily transparent read on where the model and market disagree. Wins, losses, blowups, recoveries, and the honest accounting of the days when the flagship feature loses its first big game.

If you’re new:

• 📩 Subscribe for daily Bad Bets at 10am ET

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for JSON outputs and v0.5 model internals

See you tomorrow. 🫡

— T&G

Subscribe now

This is the biggest model upgrade since the column relaunched. Four real changes, all shipped in the v0.5 push:

1. ✅ True wRC+ from FanGraphs — no more OPS-derived proxy. Real offense numbers, with the proper context adjustments.

2. ✅ Real FIP — actual fielding-independent pitching, not the ERA-based shortcut. The pitcher composite is now what it should have been all along.

3. ✅ Weather module (WX) — wind speed + direction + temperature, applied as a runs multiplier on outdoor games. Domes get a passthrough.

4. ✅ Shin de-vig — replaces the proportional method on market prices. More accurate for favorite-longshot bias; small but real ROI lift.

The footer of today’s slate now reads “offense: true wRC+ (FanGraphs) · real FIP × weather × pythagenpat · market: ESPN consensus moneyline, shin-devig.” That methodology line is finally honest. Took five iterations to get there.

What’s still on the roadmap: COL audit (next), doubleheaders, lineup news ingestion. But the bones are now in good shape. 🛠️

Subscribe now

📊 From The Ledger: Thursday

Edge picks: 2-1 (66.7%). Model picks: 3-3 (50%). Mixed day.

The wins:

• 🎯 TOR +8.0% cashed. TOR beat BAL 2-1 in a tight one. The top edge held.

• 🎯 CWS +7.2% cashed AGAIN. CWS beat MIN 6-2. That’s now two straight days on CWS — Wed they won 15-2, Thu they won 6-2. The model’s pet team is delivering.

The loss:

• ❌ PIT +4.5% busted. This was the “Skenes back” play — model said ride the ace. CHC won 7-2. Imanaga out-pitched Skenes. The same week the “Cole fade” lost on Wednesday and the “Skenes back” lost on Thursday. The model’s nuanced ace-pricing read isn’t finding traction yet. Worth a flag.

The ironic footnote: Wednesday’s only lock was ATL +13.5% @ BOS, which got shut out 8-0. Thursday, with no lock and only +1.0% on ATL, they won 10-2. Same teams, same park. Sample of one. Means nothing. Funny anyway.

Subscribe now

📈 Running Ledger Through 5 Days

Above -110 breakeven (52.4%) by 9.7 points. Five days, 58 picks. Still nothing. The model needs 200-300 graded plays before any number means anything. We’ll get there. 📊

Leave a comment

🎯 Today’s Slate

Headline numbers:

• ⚾ 14 games on the card

• 🔒 1 lock — WSH +10.9% (with weather boost)

• 🎯 5 actionable plays total — WSH, TEX, MIA, MIN, LAA

• 🔥 Slugfest: ATL @ CIN projected 10.84 r at Great American (PF 109)

• ❄️ Pitchers’ duel: PHI @ LAD projected 8.18 r (Wheeler going)

• 🌬️ Weather impact today: SF @ COL has WX 0.96 (wind blowing in, slight run suppression). WSH home has WX 1.08 (wind out — see lock).

Share

The Lock 🎯

🌬️ WSH +10.9% vs SD (with the wind blowing out)

Model: WSH 60.9% · Market: WSH 50.0% · Total est: 10.08 r · PF: 99 · WX: 1.08

Today’s only lock, and the weather module is doing real work on this number. Without weather, WSH’s expected runs at home would be around 5.30. With the WX 1.08 multiplier (wind blowing out at Nationals Park), the model bumps it to 5.74. That’s 0.44 extra expected runs for WSH from weather alone.

The market hasn’t priced this — moneylines move slower than weather forecasts, and the line was probably set when wind direction was unclear. The model now sees what the line doesn’t. That’s the entire reason the weather module exists.

The pitching matchup: Lucas Giolito (SD) vs Paxton Schultz (WSH). The model’s real-FIP composite (the v0.5 upgrade) has Schultz substantially ahead of Giolito on underlying skill metrics, which the market is also discounting.

Park factor is neutral (99), but the 10.08 total projection puts this in over-watch territory if you’re scouting team totals.

Verdict: Today’s only lock, and the cleanest read the model has produced this week. The weather is the unlock. ⚾

Subscribe now

📋 Honorable Mentions (4-10% edges)

Two contrarian plays in the top 5 (MIA, MIN). The model is taking the away dog against the market’s preferred home favorite in both. Three days ago the contrarian COL plays lost three in a row. Today the model has spread its contrarian conviction across MIA and MIN with much smaller edges. Sizing accordingly.

Leave a comment

🌬️ The Weather Module: First Live Day

Today is the first slate where weather is actually doing visible work. Here’s what to watch:

• 🟢 WSH (WX 1.08) — wind out → run boost → contributes to the +10.9% lock

• 🟢 BOS @ CLE (WX 1.06) — wind out at Progressive → BOS +2.4% edge gets a small lift

• 🔵 SF @ COL (WX 0.96) — wind in at Coors → run suppression → contributes to COL +3.5% edge being smaller than typical Coors numbers

• ⚪ Most other parks at WX 0.99-1.03 — weather neutral or near-neutral

• ⚫ Domes (KC@TEX, LAA@TB, MIL@HOU, AZ@SEA) get no WX adjustment

The interesting test: do the weather-boosted edges (WSH, BOS) outperform a model run without weather? We won’t know from one slate. But after 60-90 days of tracked WX-affected vs WX-neutral games, we’ll have a real answer about whether the module is adding signal or noise.

Subscribe now

🔁 COL Watch

SF @ COL is on the slate today (COL +3.5%, lean tier).

After three straight COL road losses last week, the model is showing a much smaller edge today on COL at home — and that’s actually a good sign. It means the road-bias issue I flagged last week isn’t generating false locks now that COL is in their natural environment.

Still on the v0.3 audit list to make sure the road-game pricing is being handled correctly. Not closing the investigation yet.

Share

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Same caveat every day.

2. 🚧 v0.5 is new. First slate with weather + real FanGraphs data running in production. Bugs are possible. The model is more accurate but the signal-to-noise ratio on the upgrades will take 30-60 days to verify.

3. 🤨 58 picks is still nothing. A 62% hit rate is encouraging but inside the noise band. Regression is coming.

4. 🎲 One lock + four edges = a slate with limited action. That’s a feature, not a bug. The model isn’t seeing forced opportunities today.

Leave a comment

✅ What This Column Is

A daily transparent read on where the model and market disagree. Wins, losses, model upgrades, weather modules, and the occasional honest accounting of where my code is still wrong.

The product is the math + the honesty. v0.5 makes the math better. The honesty is the same.

If you’re new:

• 📩 Subscribe for daily Bad Bets at 10am ET

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for JSON outputs and v0.5 model internals

See you tomorrow. 🫡

— T&G

Subscribe now

Model picks: 9-3 (75.0%). Edge picks: 7-5 (58.3%). A good day on both counts.

By edge bucket:

• 🤏 Lean (2-4%): 5-2 (71.4%)

• 🎯 Solid (4-6%): 2-2 (50.0%)

• 💪 Strong (6-10%): 0-1 (0.0%)

The lock lost. Again. ATL +13.5% — yesterday’s only lock — got shut out 8-0 at Fenway. The model had Atlanta as a 65.6% favorite; they scored zero. That’s the kind of result that has nothing to do with the model and everything to do with baseball being baseball, but it counts the same in the ledger.

The wins: CWS demolished MIN 15-2 (CWS +7.8% cashed huge), SEA rolled OAK 9-1, PHI shut out SD 3-0, NYM beat CIN 4-2. The boring solid plays did their job.

The misses worth noting:

• 🎌 The Cole fade didn’t work. I flagged KC +5.8% as the “fade the name-brand ace” play. Cole threw a shutout — NYY won 7-0. The fade lost. (The model itself picked NYY to win, so model-pick was correct, but the edge play on KC was wrong.) One data point against the fade-the-ace thesis.

• ⚾ deGrom delivered for the wrong side. TEX +8.0% lost — HOU won 4-3.

Subscribe now

📈 Running Ledger Through 4 Days

🎲 The Pattern Worth Noticing: Locks Are A Rollercoaster, Edges Are Steady

Four days in, here’s what the data is whispering:

The locks (≥10% edge) have been wild. Sunday’s CLE lock landed. Monday all five locks lost. Tuesday’s lone lock cashed. Wednesday’s lone lock got shut out. Across four days the highest-conviction tier has been the highest-variance tier — feast or famine, no in-between.

The boring solid plays (4-6% edge) have been quietly steady. They’re the ones that don’t make the headline of the article but keep showing up in the win column — CWS, SEA, PHI, NYM yesterday.

This is worth sitting with, because it’s counterintuitive. The instinct is that the biggest edges are the best bets. But a +15% edge against an efficient market usually means the model is missing something — and when it’s missing something, it’s missing it big. The 4-6% edges are smaller disagreements, which means the model and market mostly agree, which means the model is more likely right.

If this holds over 90 days, the column’s real product might not be the flashy locks at all. It might be the unglamorous 4-6% plays nobody screenshots. We’ll let the sample decide. 📊

Share

🎯 Today’s Slate

Headline numbers:

• ⚾ 6 games — the thinnest slate yet (Thursday getaway day)

• 🔒 0 locks — no ≥10% edges anywhere on the board

• 🎯 3 actionable plays (≥4%) — TOR, CWS, PIT

• 🔥 Slugfest: LAA @ DET projected 9.80 r

• ❄️ Pitchers’ duel: ATL @ BOS projected 6.53 r (Sale vs Tolle)

This is a discipline day. Six games, zero locks, three plays worth a mention. The model isn’t seeing big disagreements with the market anywhere — which, after four days where the big disagreements went 2-3, might be a feature, not a bug. When the slate’s thin, the right move is to bet the slate you have, not force the slate you want.

The Three 🎯

🎯 #1 — TOR +8.0% @ BAL

Model: TOR 53.9% · Market: BAL 54.1% · Total est: 9.43 r · PF: 100

The top edge, and a dueling-direction setup — model has Toronto as a slight road favorite, market has Baltimore as a slight home favorite. Patrick Corbin vs Chris Bassitt, two crafty veterans the model’s composite reads closer than the market does. The disagreement is the whole edge.

🎯 #2 — CWS +7.2% vs MIN

Model: CWS 64.5% · Market: CWS 57.3% · Total est: 7.59 r · PF: 96

The streak play. The model loved CWS yesterday (+7.8%) and they delivered a 15-2 beatdown of these same Twins. Now they’re back, Davis Martin on the mound again, and the model still sees value. Back-to-back days on the same team is a small caution flag (am I just chasing yesterday’s result?), but the model recomputed from scratch and landed here independently. The number is the number.

🎯 #3 — PIT +4.5% vs CHC

Model: PIT 65.4% · Market: PIT 60.8% · Total est: 7.99 r · PF: 98

Skenes is going — and this time the model backs the ace. Yesterday I flagged the model fading Gerrit Cole (it lost). Today the model is backing Paul Skenes. That’s the important nuance: the model doesn’t blanket-fade big-name pitchers. It fades when the composite says the price is inflated (Cole) and backs when the composite agrees the guy is elite (Skenes). Skenes grades as a true ace on the model’s tape, and PIT at home behind him reads as undervalued.

Subscribe now

🔁 COL Watch

No COL game on today’s six-game slate, so the streak gets a day of rest at 0-3. The v0.3 audit (most likely culprit: OPS-derived offense proxy overrating Coors hitters on the road) is still on the bench. Back to it when the schedule fills out.

⚠️ Caveats Still Apply

1. ❌ Not betting advice. Model output. Period.

2. 🚧 v0.3 priorities: weather, doubleheaders, the COL audit

3. 🎲 Thin slate = small sample. Three plays today. Don’t manufacture action where the model doesn’t see it.

4. 📊 52 picks in. Still nothing. A 63.5% model-pick rate is an early hot streak. Regression is coming. It always is.

✅ What This Column Is

A daily transparent read on where the model and market disagree. Wins, losses, shutout locks, thin slates, and the occasional uncomfortable insight that the boring plays might be better than the flashy ones. The honest accounting is the entire product.

Monday was 4-9 with all five locks losing. Tuesday cleaned up the math.

By edge bucket:

• 🔒 Locks (≥10%): 1-0 (100%)

• 💪 Strong (6-10%): 2-1 (66.7%)

• 🎯 Solid (4-6%): 2-0 (100%)

• 🤏 Leans (2-4%): 2-3 (40%)

Total on graded plays: 7-4 — that’s 63.6% on plays that cleared the 2% edge threshold. The conviction tiers held: solid and lock both perfect, strong above 60%. The leak was leans, which we expect to behave like coin flips anyway. 🪙

The big result of the day: the NYY +17.2% lock landed. NYY 15, KC 1. A blowout in the column’s favor. Monday’s +18.9% MIN lock losing had me suspicious of all big-edge calls; this one cashing means the column is now 2-2 on its biggest single-game edges (CLE +15.0% hit, MIN +18.9% missed, NYY +17.2% hit, WSH +12.8% hit). Sample of 4. Still means nothing yet.

The pain point of the day: COL got blown out again, 15-6.

That’s three straight days of COL losing as a road pick:

• 🟥 Sunday: COL @ AZ → AZ 9, COL 1 (model picked COL +12.7%, lost)

• 🟥 Monday: COL @ LAD → LAD 5, COL 3 (model picked COL +11.8%, lost)

• 🟥 Tuesday: COL @ LAD → LAD 15, COL 6 (model picked COL +11.8%, lost)

Three straight published locks. Three straight losses. COL outscored 29-10 across the run.

If the model is going to keep generating 10%+ edges on Colorado road games, I need to look under the hood. That’s now a v0.3 priority. More on it below.

Subscribe now

📈 Running Ledger Through 3 Days

Above -110 breakeven (52.4%) by 7.6 points. Encouraging but 40 picks is nothing. Mean reversion is coming. By August we’ll have enough sample to actually talk. 📊

Leave a comment

🎯 Today’s Slate

Headline numbers:

• ⚾ 13 games on the card

• 🔒 1 lock (≥10% edge) — just ATL +13.5%

• 🎯 7 actionable plays (≥4%) — ATL plus six 4-8% edges

• 🔥 Slugfest: CHC @ PIT projected 10.61 r (unusual — PNC is PF 98, not a hitter park)

• ❄️ Pitchers’ duel: PHI @ SD projected 5.98 r (Buehler at Petco, lowest total on the board)

• 🌟 Big names going: deGrom (TEX), Cole (NYY), Ohtani (LAD), Buehler (SD), Gausman (TOR)

It’s a thin slate. The model isn’t seeing the kind of disagreements with the market that Sunday and Tuesday produced. Most lines are within a few points of the model’s number — which usually means the market is sharp today, and the disciplined call is to bet less, not push harder.

But there’s one play worth talking about, and a few honorable mentions that share a theme.

Subscribe now

The One Lock 🎯

🎯 ATL +13.5% @ BOS

Model: ATL 65.6% · Market: ATL 52.1% · Total est: 9.06 r · PF: 107 (Fenway)

Fenway Park, PF 107, Bryce Elder vs Connelly Earl. The model has Atlanta as a real road favorite (-180-ish if you priced from the model); the market basically has it as a coin flip with a small lean ATL.

The gap is mostly the pitcher math. The model’s composite weighs Elder substantially ahead of Earl on both the FIP- input and the K-BB% input. The Atlanta lineup vs RHP grades well in the splits the model is pulling. Combine the two, and you get an offense outscoring BOS by ~1.5 runs even after Fenway’s hitter-park lift.

The Fenway watch: PF 107 means this game is run-environment-friendly to both sides. The total estimate of 9.06 puts the OVER on the team total or game total in scout territory. If the moneyline edge is real, the over often is too.

Verdict: Today’s only lock. Smaller edge than the multi-lock days, but the cleanest single read on the slate. ⚾

Leave a comment

📋 Honorable Mentions (4-10% edges)

The six other actionable plays today. Notice the theme — the model is fading big-name starting pitchers in three of them.

The standout from this group is NYY @ KC. Gerrit Cole is one of the three most overpriced starting pitchers in baseball on any given night — the market loves a name-brand ace and prices accordingly. The model’s composite has Cole as above-average but not the ace the market is paying for. KC at +5.8% on the moneyline is the kind of public-money fade that works long-term when you’re disciplined about identifying it.

The TEX edge with deGrom going is the inverse setup — a name-brand pitcher the market is under-pricing relative to the model, possibly because his health/innings history has trained the books to discount him.

Subscribe now

🔁 The COL Question (I Owe You An Answer)

Three straight days of COL +12% locks losing demands accountability. Let me lay out what I think is happening.

The pattern:

• 📊 Model thinks COL hitters are systematically underpriced when they go on the road

• 📉 Market disagrees — has had this opinion for years

• 🏟️ Reality (last 72 hours): COL outscored 29-10, market correct in all three games

The three most likely causes:

1. 📊 OPS-derived wRC+ proxy overrates COL hitters. The current v0.2 model uses OPS as a wRC+ proxy. Colorado hitters post inflated OPS at home (Coors). If I’m not adjusting for that before scoring their road games, COL’s offense number is too high. This is the most likely culprit and the easiest to fix.

2. 🌬️ Park factor logic is misapplied. The model uses the road park’s factor. But COL specifically is the team where home-vs-road environment is the biggest. A team that hits .280 at Coors might be a .240 team on the road. The fix here is a separate “home advantage” multiplier that’s specifically punitive for COL road games.

3. 🎲 It’s variance, and three games means nothing. Three losses in a row is a 12.5% probability event if each game is a 50/50. So purely on coin flips, you’d expect this once every 8 such sequences. Plausible — but not the bet I’d take when the alternative is “your model has a known bug.”

What I’m doing about it:

✅ Today’s slate has COL @ LAD with Ohtani going. Model says +0.4% edge on LAD — basically nothing. This is one to watch as a calibration check: when the model isn’t generating a juicy COL edge, does it correctly say “this is a coin flip” or worse for COL? Today says yes. That’s actually evidence in favor of “the model’s COL number isn’t structurally wrong — it just produced three bad random walks.”

But I’m still digging. v0.3 will include a Colorado-specific audit before it ships.

Leave a comment

⚠️ The Caveats Still Apply

1. ❌ Not betting advice. Model output. Period.

2. 🚧 v0.3 priorities now: weather (especially relevant for Fenway today), doubleheaders, AND a COL audit

3. 🤨 One lock today is just one lock. Slate quality varies. The discipline call when the model isn’t seeing big disagreements is “respect the slate’s verdict and size smaller.”

4. 🎲 40 picks is nothing. Don’t extrapolate from a 60% hit rate. Don’t extrapolate from a 0-3 COL streak either.

Leave a comment

✅ What This Column Is

A daily transparent read on where the model and market disagree. We publish the wins (Sunday: 10-3). We publish the losses (Monday: 4-9). We publish the methodology audits (COL, today). We publish the slow days too (today: one lock).

The honest accounting is the whole product. If you ever feel like you’re getting a different version of the column on bad days than on good days, the column is failing.

If you’re new:

• 📩 Subscribe for daily Bad Bets at 10am ET

• 🐦 Follow @TradesAndGains for slate graphics

• 💬 Discord for the JSON outputs and the model internals

See you tomorrow. 🫡

— T&G

Subscribe now

Let’s start with the receipts.

Monday’s slate: 4-9 on completed picks. All 5 locks lost.

The model picked five games as ≥10% edges. Five. All five lost.

• 🔴 The +18.9% MIN lock @ CWS? CWS 3, MIN 1. Lost.

• 🔴 The +13.6% SF “home arm at a pitcher’s park” lock? AZ 6, SF 2. Lost.

• 🔴 The +12.6% COL contrarian dog @ LAD? LAD 5, COL 3. Lost.

• 🔴 The +12.5% TEX slugfest call vs HOU? HOU 9, TEX 0. Lost. Badly.

• 🔴 The +11.0% NYM “quiet lock — trust it most”? CIN 7, NYM 2. Lost.

If you backed every lock at flat -110, Monday cost you 5 units. If you fired all eight actionable plays (≥4% edge), you went 2-6 on edges of ≥5%, and 4-9 overall. 🫠

This is what a -2σ day looks like. The honest take: the model went 0-5 on the highest-conviction picks on the same day it had its widest spread of edges. That’s not a coincidence — that’s the model finding what looks like value in spots where the market actually knew more than we did.

Or: variance. Could also be just variance. After one day, you can’t tell. 🎲

Now let’s look at Tuesday. 🫡

Leave a comment

🎯 Today’s Slate

Headline numbers:

• ⚾ 14 games on the card

• 🔒 4 locks (≥10% edge) — NYY, WSH, COL, SEA

• 🎯 8 actionable plays (≥4%) — the 4 locks plus MIN, TEX, MIL, MIA

• 🔥 HOU @ TEX projected for 9.75 — slugfest watch (the same matchup that went 9-0 yesterday on the wrong side of the model)

• ❄️ PHI @ SD projected for 7.14 — Nola going, low total

• 📈 Biggest single-game edge: NYY +17.2% in Kansas City

After Monday’s reality check, today’s slate has the biggest single-game edge the model has produced in this column’s short life: +17.2% on the Yankees in Kansas City.

We’re going to walk through all four locks, including the COL one — same team the model picked yesterday and got crushed on. Same shape, same matchup type, similar edge. We’re not changing the read just because yesterday hurt. That’s the entire point.

Subscribe now

🧠 Methodology, In Case You’re New

The column runs a quant model called T&G MLB Edge every morning, compares it to the de-vigged DraftKings moneyline, and surfaces the games where the two most disagree. Anything ≥4% is actionable. ≥10% is a lock.

The math is fully transparent — pitcher composite from FIP- and K-BB%, bullpen blend at 58/42 weight (matches actual SP/RP innings split), team offense by handedness, park-adjusted run environment, Pythagenpat win probability. Full breakdown in the comeback issue if you want it.

No machine learning. No proprietary data. No vibes. Just the math, the disagreement, and a published number.

Subscribe now

The Four 🎯

🎯 #1 — NYY +17.2% @ KC

Model: NYY 81.5% · Market: NYY 64.3% · Total est: 8.82 r · PF: 98

The biggest edge this column has surfaced.

The model has the Yankees scoring 6.08 expected runs in Kauffman, against a Royals lineup it thinks puts up just 2.74 runs against Schlittle. That’s a +3.34 run gap. The market has NYY as a real road favorite (~-180), but the model prices NYY closer to -440. That spread is unusual against a liquid major-market line. 🚨

Past +17%-tier calls in this column:

• CLE +15.0% (comeback issue) → ✅ HIT

• MIN +18.9% (Monday) → ❌ MISS

Sample size: 2. The model is currently 1-1 on its biggest single-game edges. Means nothing.

What I’d verify before fading or backing: Royals lineup card (any IL news on their top three?), Falter’s last three starts (the model uses season FIP-, which can mask recent regression), and the closing line direction once lineups post.

Verdict: Model loves NYY. Past calls of this magnitude have been a coin flip in the lab. Size accordingly. 🪙

🃏 #2 — WSH +12.8% @ CLE

Model: WSH 57.7% · Market: CLE 55.1% · Total est: 8.25 r · PF: 98

The rarest setup the column will surface: model and market disagree on direction.

Market has Cleveland as a small home favorite. Model has Washington as a slight road favorite. Somebody is meaningfully wrong, not just off by a few points.

Cavalli (WSH SP) is the bet. The model’s composite has him substantially ahead of Cantillo, who’s been hit around recently. Combine with a CLE offense the model rates as middling and an away dog priced cheaply (WSH on the road is structurally fade-able money), and you get the read.

Verdict: The model is contrarian, and the contrarian side has structural value. Worth the price. 📈

🔁 #3 — COL +11.8% @ LAD

Model: LAD 55.0% · Market: LAD 66.8% · Total est: 11.12 r · PF: 99

This is the rerun. 🔁

Sunday: model had COL +12.7% @ AZ. Lost 9-1. Monday: model had COL +12.6% @ LAD. Lost 5-3. Tuesday: model has COL +11.8% @ LAD. Same series, same shape, same edge.

I’m publishing the same read I’d publish if the last two days hadn’t happened. The model doesn’t remember yesterday’s loss, and neither should I. Either:

• (a) The model has a structural bias toward overrating COL on the road, and I need to dig into the run estimator

• (b) Variance — two losses on small samples is well within normal range for +EV plays

• (c) Both

The total here is 11.12 expected runs — a slugfest at Dodger Stadium. If the model is right that COL hangs around, the over/under might be the more interesting play than the moneyline.

Verdict: Backing the third straight COL lock requires conviction. If you’ve faded the prior two, you’re 2-0. If you backed them, you’re 0-2. Process over outcome. 🎯

🎯 #4 — SEA +11.0% @ OAK

Model: SEA 62.0% · Market: SEA 51.0% · Total est: 7.86 r · PF: 91

The cleanest lock of the four. 💎

Severino has a recognizable name but the model’s composite weighs recent form and underlying skill metrics heavier than reputation, and both are unkind to him right now. Hancock (SEA) profiles as the better starter on tape.

The Oakland park (PF 91) suppresses the run environment for both sides equally, which means small advantages compound — and the model thinks SEA’s small advantages are real. The total of 7.86 also reads under, if that’s your kind of bet.

Verdict: Boring lock. Same profile as Sunday’s MIA quiet-lock that landed. These are the locks I trust most. 😴

Subscribe now

📋 Honorable Mentions (4-10% edges)

⚠️ The Caveats (Still True, Maybe More So Today)

After yesterday, this section is worth posting in full again:

1. ❌ Not betting advice. Model output. Period.

2. 🚧 The model doesn’t yet know: lineup news, last-72-hour bullpen state, weather, umpire factor, catcher framing, doubleheaders, late scratches. v0.3 is targeting weather and doubleheaders next.

3. 🤨 A +17% edge is almost certainly partly wrong. I gave this same warning before the +15% CLE landed and the +18.9% MIN lost. Sample size 2. Means nothing yet.

4. 🎲 MLB variance is the entire genre. Going 4-9 yesterday doesn’t mean the model is broken. Going 10-3 Sunday didn’t mean it was solved. Sample size is the only judge.

Through two iterated days, the column sits at 14-12 on completed plays — 53.8%. -110 breakeven is 52.4%. We’re up roughly 1.4 percentage points after two days of noise. Do not extrapolate from this. Do not extrapolate from anything yet. By August we’ll have enough sample to talk seriously.

Today we just publish. 📰

Subscribe now

✅ What This Column Is

A daily honest accounting. The model says what it says. The math is the math. Sometimes the locks land. Sometimes all five lose on the same day. Yesterday I had to publish a 4-9 day. I’ll do it again whenever it happens. That’s the deal.

If you’ve ever been frustrated by sports-betting Twitter promising 70% win rates and locks of the century, this is the antidote. We post the wins. We post the losses. The losses with extra emphasis. 🧯

Share

Quick reminder on the engine, now running v0.4: team offense is real wOBA (weighted by what each hit is actually worth), pitchers are scored on real FIP (strikeouts, walks, homers — the stuff a pitcher controls), and everything gets run through park factors, weather, and a Pythagenpat win-probability model. Then I de-vig the DraftKings line and subtract. 📊

edge = my win probability − the market’s win probability. Anything ≥ 4% clears the vig. Today, five games clear 10%.

Let’s go. 👇

Subscribe now

🚨 The five locks

🎯 1. MIN @ CWS — model says MIN, +18.9%

The biggest gap I’ve seen the model spit out. 😳 The market has this as a coin flip (MIN ~51%); my model has Minnesota at nearly 70%. An 18-point disagreement is a screaming flag — and not necessarily a good one. My honest read: the White Sox are a weak offense (the wOBA model docks them hard) facing a Twins arm the model likes, but a gap this size usually means something I can’t see — a bullpen note, a lineup the market knows is resting. Verdict: I’m intrigued, but I trust this one the least precisely because it’s the loudest. 🔍

🔒 2. ARI @ SF — model says SF, +13.6%

San Francisco at home, in Oracle Park (park factor 94 — one of the league’s best pitcher’s yards), behind an arm the model rates well. 🌉 The model has the Giants at ~70% vs the market’s ~56%. This is the kind of edge I like more than #1: it’s built on a real, legible reason (a good pitcher in a run-suppressing park against a road offense), not a mystery gap. Verdict: defensible. The park is doing real work here.

🃏 3. COL @ LAD — model says COL, +12.6% (yes, the dog)

Read this one carefully. ⚠️ The model still thinks the Dodgers win — it just thinks LA should be ~61%, while the market has them at a massive 73%. So the value is on Colorado as a live underdog, not because the Rockies are good (they’re not), but because the market may be overcharging for the Dodgers’ name. 🎲 This is a classic public-money spot: everyone wants LA, the price gets bid up past fair. Verdict: a value stab on the dog, not a “COL wins” call. Tiny stake, eyes open.

🔥 4. HOU @ TEX — model says TEX, +12.5% 🌭 SLUGFEST

Texas at home, and the model is also flagging this as the highest-scoring game on the board (projected total 10.5). 💥 Rocker projects for a heavy line, but the model sees the Rangers’ bats outscoring it. The interesting wrinkle: when the projected total is this high, the favorite’s edge gets a little noisier (more runs = more variance), so I’d weight the side slightly less and note the over as the louder signal here. Verdict: edge on TEX, but the total is the headline.

💎 5. CIN @ NYM — model says NYM, +11.0%

The quiet one — and usually the one I trust most. 😌 Mets at home, a clean ~70% model number vs the market’s ~59%, no park weirdness, no rookie-pitcher noise, no slugfest variance. Just a steady model preference the market has only half-priced. 💎 Verdict: the calmest lock on the slate. If I’m putting weight anywhere, it’s here and on SF.

Subscribe now

🤏 Below the locks

Four more clear the 4% bar: 🟢 MIL +7.8% (Milwaukee a big home favorite the model likes even more), TB +5.9%, and a pair of leans in CHC +4.2% and NYY +4.2%. The full 13-game board — model vs market, color-coded by edge — is in the graphic up top. 📈

And the boring-but-important truth: the model can’t see everything yet. Confirmed lineups don’t post until the afternoon, and my umpire/framing inputs are still being calibrated. So treat the double-digit edges as “go look at this” flags, not blank checks. 🚧

Share

🧾 The standing rule

Bad Bets isn’t a picks service. 🙅 It’s a transparent model, published daily, with a scoreboard that doesn’t lie. 📅 One day of five locks means nothing — I’ll grade every one of these tonight and the record goes in the ledger, wins and losses. If the model’s a fraud, you’ll watch it happen in real time. If it’s real, same. 🫡

Today’s gaps are unusually wide. That makes it a great day to watch — not necessarily a great day to bet the house.

See you at the grade. ⚾

🆘 Model output only — not betting advice, and nothing here is a guarantee. Must be 21+ (or legal age in your jurisdiction). If you or someone you know has a gambling problem, call 1-800-GAMBLER or text 800GAM to 800-522-4700. Bet what you can afford to lose, or don’t bet at all.

Subscribe now

Bad Bets has been quiet for a minute. Not because the market got efficient — it didn’t — but because I was off building something that I think actually justifies bringing the column back.

It’s called T&G MLB Edge. It’s a quant model. It runs every morning on the day’s slate, scores each starting pitcher on a 0-100 composite, estimates park-adjusted expected runs for each side, converts that to a fair win probability, and compares the answer to the de-vigged DraftKings consensus moneyline. 🎯

When the two disagree by 10+ percentage points, the column writes itself.

Today they disagree on four games. That’s a lot. We’re going to walk through all four, including the ones that are probably partially wrong, and you’re going to leave knowing how to think about that kind of disagreement — not because I’m handing you locks, but because I’m not.

No fire emojis on parlays. No “lock of the century.” No PDFs in your DMs. Just the math, the disagreement, and an honest read.

Let’s go. 📊

Subscribe now

Today’s Slate 🗓️

The headline numbers:

• ⚾ 15 games on the card

• 🎯 8 actionable edges (model vs market gap ≥5%)

• 🔒 4 “locks” (gap ≥10%)

• 📈 Top edge: CLE +15.0% @ PHI

• 🔥 Slugfest alert: STL @ CIN at Great American Ball Park (PF 109, model total 11.20)

• ❄️ Pitchers’ duel: ATH @ SD (model total 6.56)

The four ≥10% games are where this issue lives. 👇

Share

How The Model Reads A Game 🧠

Before we walk through anything, the methodology — because if you don’t know what a number means, you can’t tell when it’s lying to you.

For each game, the model builds:

1. Pitcher composite score (0-100) ⚾ — blended from FIP-, K-BB%, and recent form. Skubal-tier guys score in the high 80s. Replacement-level call-ups land in the low 20s.

2. Park-adjusted run estimate 🏟️ — opponent offense quality, adjusted by the starter facing them, scaled by a 3-year rolling park factor for the venue. Coors gets a 1.14 multiplier. Petco gets 0.91.

3. Pythagenpat win probability 📊 — converts the two run estimates into a fair win probability using an adaptive exponent (handles high-scoring environments better than the textbook Pythagorean).

4. De-vigged market price 💰 — pulls live DraftKings moneylines via the ESPN pickcenter endpoint, converts to implied probability, normalizes to strip the vig.

Edge = model probability minus market probability. Positive = model thinks this side is undervalued. ≥4% gets a look. ≥10% gets a column.

That’s the whole job. No vibes. No “I have a guy.”

The Four 🎯

🎯 #1 — CLE +15.0% @ PHI

Model: CLE 66.1% · Market: CLE 51.1% · Total est: 9.08 r · PF: 105

The biggest gap on the board. The model has Cleveland’s offense outproducing Philly’s by nearly two runs at Citizens Bank.

This is the play to be most suspicious of, not most confident in. 🚨 When a model spits out a 15% disagreement against a liquid major-market line, the realistic question is what does the market know that the model doesn’t? Possibilities, in rough order of likelihood:

• 📋 A late lineup scratch the morning fetch didn’t catch

• 🪫 A bullpen state issue (PHI’s pen burned three days running?)

• 📝 The Philly starter has scouting reports we’re not modeling

• 📉 The Cleveland starter’s recent line is hiding underlying regression

If you can clear all four of those after checking lineup news, weather, and the last week of pen usage, you’ve got an unusually big number on a real market. If you can’t clear them, that 15% is probably the market pricing in something we don’t yet see.

Verdict: The model loves CLE. Do your homework before you do too. 🔍

🃏 #2 — COL +12.7% @ AZ

Model: AZ 51.6% · Market: AZ 64.3% · Total est: 10.08 r · PF: 101

The most counterintuitive number on the slate, which is exactly why it’s interesting. 🤔

Colorado on the road is historically one of the worst bets in baseball. The market knows this. The market has been right about it for twenty years. Books have a structural edge baked in by recreational money that automatically fades the Rockies anywhere outside Coors.

So when the model says the line is 13 points too wide on the AZ side, the model is essentially claiming a known market bias is over-extended today. That’s a strong claim. It’s the kind of claim a model needs to back up over a 100-game sample, not a one-shot. 📈

Verdict: Model is shouting at the market. Take the COL number seriously, but size accordingly. 🎰

🔥 #3 — STL +10.4% @ CIN

Model: STL 57.9% · Market: CIN 52.5% · Total est: 11.20 r · PF: 109

The slugfest game of the day. 💥 Great American Ball Park is a 109 park factor — the league’s third-most hitter-friendly venue. The model total of 11.20 runs is screaming OVER.

The moneyline edge here is real, but the more interesting market read might actually be the total. ⬆️ When the model and market disagree on a side at a park where any hanging slider exits the building, the play often isn’t the side — it’s the over. We don’t compute totals lines yet (it’s on the v0.2 roadmap), but eyeball whatever the book is offering and ask yourself if you’ve seen a higher number than 11 today.

Verdict: Take the STL number with one eye on the total. 👀

Subscribe now

💎 #4 — MIA vs NYM

Model: MIA 64.1% · Market: MIA 54.1% · Total est: 7.28 r · PF: 96

The quiet one. 🤫 No park drama, no narrative twist — just a model that sees Miami’s home edge being underweighted by the market. loanDepot is a slight pitcher’s park, both starters profile as control types rather than strikeout artists, and the model expects a low-scoring game where the home offense plays a bigger role than the line implies.

This is the kind of edge that wins over a full season and doesn’t make highlights. Not flashy. Probably correct.

Verdict: The most boring +10% on the board. Often the best kind. 😴

Honorable Mentions 📋

The other four actionable edges (5-8%) don’t get full writeups but deserve a row in the ledger:

What This Column Is Not ⚠️

A few honest disclaimers before anyone runs to the cashier window:

1. ❌ This is not betting advice. It’s model output. The model is v0.1.1 and openly imperfect.

2. 🚧 The model doesn’t yet know about: lineup news, bullpen state from the last 72 hours, weather, umpire factor, catcher framing, or full L/R splits. All of these move lines, and the model will miss them. v0.2 is targeting bullpens and lineup splits next.

3. 🤨 A +15% edge is almost certainly partly wrong. The market is efficient enough that 15% gaps usually mean we’re missing something. The middle of the edge distribution (4-8%) is generally where the model is most defensible.

4. 🎲 MLB is a high-variance sport. Even a 60% pick loses 40% of the time. A bad day doesn’t mean a bad model. A good day doesn’t either. Sample size is the only thing that knows.

What This Column Is ✅

A daily honest accounting of where my model disagrees with the market — published openly, with the math shown, so you can decide whether to trust it.

If you’ve ever been frustrated by sports-betting Twitter promising 70% win rates and locks of the century, this is the antidote. 🧯 We show the numbers. We tell you when we’re probably wrong. We bench-test the model in public.

Tomorrow we run it again. And the day after. By August we’ll have enough sample to start scoring the column against itself. 📅

Leave a comment

The Yankees did not have a great showing yesterday only racking up 2 hits in their 27 plate apperances while the cubs were able to shell out 7 hits on 33 attempts. Yankees also had to start dipping into their bullpen just after 5.1 innings with Rodon on the mound starting. Things would go downhill after putting in Hamilton as the backup.

Share

A PITCHERS DUEL YOU SAY? 🤺

The Yankees will be starting Gerrit Cole on the mound tonight, with a 5-1 record at home he is sporting a 2.82 ERA with 77 strikeouts. In his last appearance against a streaky Cardinals team Cole pitched 6 innings where he gave up 6 hit and 2 runs while striking out 5 batters and walking 2 in a 5-1 loss at St. Louis. In his last 5 games he has a 2-3 record so far, is this the game where he can turn things around?

The Cubs will be starting Drew Smyly who comes into this game with a 4-3 record on the road where he has dealt 37 strikeouts and earning himself a 3.08 ERA. In his last start against the Brewers he only pitched 3.2 innings where he gave up 4 hits and 3 runs, while striking out 5 batters and walking 3. It seems Smyly is not a guy they take past 5 innings and in his last 2 starts he has only pitched 3.2 innings in each game.

Subscribe now

Both of these pitchers had ugly June records and have not had a great start to their July debuts, with the halfway point of the season hitting look for both of these pitchers to start tightening up their spots on the mound to gear up for a pennant run.

THE CUBS ARE HITTING WELL? ⚾️

Chicago Cubs | 7 Games | .239 AVG | .308 OBP | .355 SLG Rating

The Cubs have been playing decent baseball as of late, the biggest question mark heading into the season was the Cubs ability to hit the ball and get on base. The beginning of the season was rough, last night was a testament to how far the Cubs batting line up has come. In recent months they have seemed more poised and composed. I would look for Swanson, Hoerner, Mancini, Madrigal & Happ to all be big contributors and factors if they want to go up 2-0 in the series.

N.Y. Yankees | 8 Games | .221 AVG | .301 OBP | .369 SLG Rating

The Yankees have had a rough past couple months, from incosistent play, keeping teams in games that have no business being there, pitchers giving up 3-4 run leads in one inning. The Yankees have not looked like themselves consistently this season and their record has suffered for it., Volpe, Rizzo, Donaldson, McKinney, & Co. need to step up their play if they want to win this series. The 2 hits the Yankees got yesterday will not help them win any game.

Read more

A PITCHERS DUEL DOWN SOUTH ⚾️

The Braves are starting Charlie Morton on the hill tonight, in his last start against the Marlins he pitched 5.2 innings where he gave up 4 hits, 0 runs, while striking out 5 batters and walking one. In his last 5 starts he 4-1 but he never goes over 5 innings it seems, so lets hope the Braves bullpen is rested and ready to deal out some heat.

The Rays will be starting Tyler Glasnow on the mound tonight, in his last start against the Mariners he pitched 5.2 innings where he gave up 7 hits, 3 runs, and 1 home run while striking out 11 batters and only walking 1 in the 8-3 loss. So far in the month of June he has a 3-2 record.

Atlanta will have the pitching advantage headed into this game and I would look for them to take full advantage of it here tonight and make sure their bullpen is on point to close out.

Subscribe now

OFFENSE OFFENSE OFFENSE 🏟️

These teams have different levels of production for offense so we are going to take a look at their last 7 days of productivity and give you some of the key players to lookout for.

BRAVES | 6 GAMES | .319 AVG | .368 OBP | .657 SLG RATING

Murphy is currently leading the team in AVG & OBP in the last 7 days sporting a .533 AVG and .556 OBP respectively. Quite frankly the whole Braves batting line up is just disgusting no matter who they have in it. I would look for The Braves to hop out early tonight and try to maintain a cushion between them and the Rays. It does not help that the Rays will have a lesser experienced pitcher kicking off the mound, but luckily he will not be going past 5 innings if he can make it that far…

Share

RAYS | 6 GAMES | .215 AVG | .269 OBP | .402 SLG

Since their hot start to the season it seems they Rays hit a mid season wall or slump of some sorts, now this is not to say they are playing poorly, but the numbers do speak for themselves. The Braves have almost a whole percentage point advantage in the same amount of games. I would look out for Lowe, Bethancourt, Franco, & the OTHER LOWE to output some heavy production if they want the win at home tonight. If they come out flat foot this could be another 5-6 run blowout for the Rays

Read more