Daily Bet Research Report — Saturday, September 19, 2026

Start: 11:11 ET | End: 11:42 ET

Bankroll: $50.40 | Stake (12.5%, one play): $6.30 Record: 2-1-0 (Main bets, main leg, since 9/14 reset) | Hit rate: 66.7% Slate: 15 games · 12 analysed · Playbooks: MLB v1 · Verified at 11:13 ET

Section 2: Game Analysis

Unanalysed: 3 of 15. SEA @ COL (Hard Rule H1 — no play at Coors — and Colorado's starter is unnamed anyway), SF @ LAD and BOS @ TB (both H3: starter unnamed; Tampa Bay additionally has no DraftKings market posted, so there is nothing to price).

Section 3: Today's Plays

⚠ Two plays carry negative edge: #3 (−2.5%) and #4 (−1.6%). Both are ranked on the handicapping read rather than the price, and the numbers are shown rather than buried. #1 and #2 are the only positive-edge plays, and both come from the same source — bullpen state, which §3 of the playbook calls the most under-priced factor available.

What came off the card. An earlier pass had Baltimore +1.5 at 1.64 as the top play, on Trevor Rogers' +0.54 ERA−FIP gap. Checking last night's box score killed it: Baltimore's starter went 3.2 innings and seven relievers covered 6.1 more in a 10-inning loss. A +1.5 play needs the game to stay close, and that is exactly what a spent bullpen cannot deliver. It is recorded here rather than quietly dropped.

Calibration caveat, restated: the 3-and-4-point overconfidence deductions were measured on a mostly-prop sample. There is very little history for them on game lines, so they remain a starting assumption.

Section 4: Reasoning

#1 — Chicago Cubs ML (Moneyline, 1.73)

This is a §3 bullpen play with a §2 starter read underneath it, and the two point the same way. Last night at Great American Ball Park the Reds got three innings from Chase Burns (67 pitches) and then used four relievers for six innings — Aguiar 3.0, Brock Burke 1.0, Antone 1.0, Pagán 1.0, 99 pitches between them. Chicago, in the same game, used two relievers for 2.2 innings and 46 pitches behind Clay Holmes' 5.1. §3's first check is "innings thrown by the pen last night," with 4+ flagged as a genuine next-day disadvantage: Cincinnati is at six, Chicago at 2.2. The starter matchup compounds it rather than offsetting it. Nick Lodolo carries a 5.12 ERA and a 5.81 FIP — 19 home runs and 14 hit batsmen in 96.2 innings, a 1.77 HR/9, a .868 OPS against and 17.76 pitches per inning, which is a pitcher who does not finish six. Matthew Boyd's FIP is 4.40 against a 4.16 ERA. That is a 1.41-run gap in the starters [MODERATE], handing off to a 1.41-run gap in the relief corps' freshness [MODERATE]. Line movement is the cleanest part of the case: DraftKings opened Chicago -136 and the price has not moved a cent, so the market has not priced last night's usage at all — which is precisely the slow-absorption edge §3 describes. The risk that beats me: Great American is a bandbox and Boyd's own 1.18 HR/9 plays badly there, so a two-homer inning off the Cubs' starter makes the bullpen comparison irrelevant. Sources opened: statsapi.mlb.com/api/v1/game/824463/boxscore; ESPN core API event 401816999.

#2 — Chicago White Sox ML (Moneyline, 1.72)

The same factor, larger in magnitude, in a game the market has if anything moved the wrong way on. Detroit lost 11-8 at Rate Field last night in a game where their starter, Andrew Sears, recorded two outs on 37 pitches. Five relievers then threw 8.1 innings: Brieske 2.1 (49 pitches), Sommers 2.0 (22), Hurter 1.2 (21), Kinley 1.1 (17), Jansen 1.0 (10). That is a full bullpen game, and §3 is explicit that 4+ relief innings is a real next-day disadvantage — Detroit is at more than double it. Chicago's pen worked too, 5.0 innings across six arms, but got four innings from Sandlin first. Today's starter makes it worse for Detroit rather than better: Jackson Jobe has thrown 34.2 innings across seven starts, under five per outing, so the Tigers will be back in that bullpen by the fifth. His surface line is also the softest on the slate — a 3.89 ERA against a 4.91 FIP, a 1.02-run gap past the playbook's 0.75 flag, on a 1.56 HR/9 and 3.38 BB/9. Sean Burke is the more trustworthy arm: 3.47 ERA, 3.68 FIP, 9.67 K/9, 1.20 WHIP over 161 innings. The market drifted from CHW -141 to -137 — four cents toward Detroit, inside the five-cent threshold and pointing away from the bullpen fact. The risk that beats me: Jobe's FIP rests on 34.2 innings, which is a noisy sample in its own right, and Detroit scored eleven runs last night against this same pitching staff. Sources opened: statsapi.mlb.com/api/v1/game/824547/boxscore; ESPN core API event 401817004.

#3 — New York Yankees ML (Moneyline, 1.68)

The largest talent gap on the slate, held back by the loudest warning sign. Cam Schlittler has a 2.62 FIP over 184.2 innings — 228 strikeouts, 44 walks, 16 home runs, 11.11 K/9, 0.90 WHIP. Brandon Pfaadt's FIP is 4.11 on a 6.10 K/9. A 1.49-run gap in the metric §2 treats as the baseline skill measure is why New York is favoured on the road, and it is a real edge. What stops me ranking it higher is the movement: DraftKings opened New York -162 and the price is now -146, sixteen cents against my side, far past the five-cent threshold at which the methodology says to lower confidence rather than ignore it. Schlittler is at 184.2 innings in his first full season, and an innings cap or a short leash is the obvious thing money might know that I do not. I did not check Arizona's bullpen state, which given that leash is the gap I would close first with more clock. The risk that beats me: exactly that — a Yankees starter pulled at 75 pitches hands the game to relievers I have not examined, in a park that rewards contact. Sources opened: statsapi.mlb.com/api/v1/people season splits; ESPN core API event 401817010.

#4 — Houston Astros ML (Moneyline, 1.83)

The headline divergence on the board, and a fade — which §2 notes is the harder direction to trust. Grant Holmes shows a 3.49 ERA over 141.2 innings sitting on a 4.75 FIP: 20 home runs, 59 walks, just 110 strikeouts, a 6.99 K/9 and a 1.86 K/BB. The 1.26-run gap clears the playbook's threshold for making it the headline of the writeup. Hayden Wesneski is unremarkable but honest — 3.70 ERA, 3.77 FIP, 1.13 WHIP — though across only 48.2 innings in nine starts, which is a genuine caveat on my own side of the matchup. The market has already made this move and made it expensive: DraftKings opened Atlanta as a -103 favourite and Houston is now -120, a seventeen-cent swing toward my side. Per the methodology that means the market agrees and I am late, which is why this ranks fourth despite carrying the widest metric gap on the slate. The risk that beats me: Atlanta is 89-64, beat Houston 6-2 last night, and Holmes has outrun his peripherals for 28 starts — sequencing and defence that have held all season can hold one more night. Sources opened: statsapi.mlb.com/api/v1/people; ESPN core API event 401817003.

Note on MIN @ LAA — read but not played

Reid Detmers has the best underlying line on the slate that nobody is paying for: a 3.09 FIP against a 3.36 ERA over 174 innings, 10.29 K/9, 1.05 WHIP, with the ERA−FIP sign pointing up — mildly unlucky, which §2 makes a buy signal. Joe Ryan is legitimate but a step back at 3.74 FIP. The Angels at -112 at home are a live price on pitching alone. What kept it off the card is that DraftKings has already dropped the total from 7.5 to 7.0 and flipped the juice hard to the over (-126 / +104) — the market priced this low-scoring read overnight — and a 59-94 club has every September incentive §6 warns about to rest regulars, a lineup I could not confirm.

Sources

• https://statsapi.mlb.com/api/v1/schedule?sportId=1&date=2026-09-19&hydrate=probablePitcher
• https://statsapi.mlb.com/api/v1/schedule?sportId=1&date=2026-09-18 (prior-day finals, for bullpen work)
• https://statsapi.mlb.com/api/v1/people?personIds=695549,680732,688107,669432,691725,681035,682052,676440&hydrate=stats(type=season,group=pitching)
• https://statsapi.mlb.com/api/v1/people?personIds=656550,669713,667755,615698,657746,672282,693645,694297&hydrate=stats(type=season,group=pitching)
• https://statsapi.mlb.com/api/v1/people?personIds=571510,666157,702070,696149,700241,674841,663554,691587&hydrate=stats(type=season,group=pitching)
• https://statsapi.mlb.com/api/v1/game/824547/boxscore (DET @ CWS, 9/18 — bullpen usage)
• https://statsapi.mlb.com/api/v1/game/824790/boxscore (MIL @ BAL, 9/18 — bullpen usage)
• https://statsapi.mlb.com/api/v1/game/824463/boxscore (CHC @ CIN, 9/18 — bullpen usage)
• https://sports.core.api.espn.com/v2/sports/baseball/leagues/mlb/events?dates=20260919&limit=100
• http://sports.core.api.espn.com/v2/sports/baseball/leagues/mlb/events/401816997 (event detail + Cleveland weather)
• ESPN core API odds, one per game: events 401817004 (DET@CWS), 401817001 (MIL@BAL), 401816998 (BOS@TB — empty), 401817002 (PHI@NYM), 401816997 (ATH@CLE), 401816999 (CHC@CIN), 401817000 (KC@PIT), 401817005 (TOR@TEX), 401817003 (ATL@HOU), 401817006 (WSH@STL), 401817010 (NYY@ARI), 401817008 (MIA@SD), 401817009 (MIN@LAA) — each at https://sports.core.api.espn.com/v2/sports/baseball/leagues/mlb/events/{id}/competitions/{id}/odds
• https://www.fangraphs.com/api/leaders/major-league/data (three attempts — all failed, see limitations)
• Local: C:\Users\mitch\Documents\BettingTracker\playbooks\mlb.md, bet_tracker.xlsx (Log sheet)

Research limitations

Counts: 12 of 15 games analysed. 27 sources fetched (2 schedules, 3 player-stat batches, 3 box scores, 1 event index, 1 event detail, 13 odds endpoints, 3 failed FanGraphs attempts, 1 local workbook).

• xFIP and SIERA could not be obtained — the playbook's named source is broken for this toolchain. FanGraphs' leaderboard API is the only source §9 lists for them, and three separate requests all returned payloads that hit the fetch tool's ~59,000-character ceiling at 58,842, 58,762 and 58,869 characters. Narrowing with team= and with a custom type=c,13,120 column list changed the roster but not the response size, and the saved-file fallback could not be sliced: the tool-results directory sits outside the Linux sandbox's mounts, and the JSON is a single line too long for chunked reads. Recommend adding to §9's failure list: the FanGraphs leaderboard endpoint is unusable through this fetch tool at any page size or column selection tested.
• FIP was substituted for xFIP throughout, computed by hand from statsapi counting stats as (13·HR + 3·(BB+HBP) − 2·K) / IP + 3.15. The playbook is explicit that FIP "provides no predictive value over simply using prior ERA" [STRONG], so every ERA−FIP gap quoted above is a weaker signal than the ERA−xFIP gap the method actually calls for. This is the single biggest weakness on today's card and it touches all four plays. It also cost a specific read: Washington's Andrew Alvarez shows a 3.32 FIP that is almost entirely a 0.44 HR/9 artefact — the precise illusion xFIP is designed to strip out — and without xFIP I could only flag it, not size it.
• Bullpen state checked for three games only — DET @ CWS, MIL @ BAL, CHC @ CIN — chosen because they were the live candidates. Not checked for the other nine, and that gap is load-bearing on play #3: Arizona's pen matters given Schlittler's likely innings limit, and I did not look.
• No lineups confirmed for any game. §6 warns this matters most in September, and two analysed games involve eliminated clubs (Athletics, Angels) likely to rest regulars.
• No umpire assignments (§5, worth 0.3–0.5 runs). Not attempted. No total was played, which limits the cost.
• Weather obtained for one park only — Cleveland (9 mph E, 70°F, 34% precipitation), and only incidentally via an event-detail fetch. No wind data for any other outdoor park. No Wrigley game on the slate, so H2 did not bind; but §4's "check direction and speed for every outdoor total" was not met, which is part of why no total was played.
• Three games could not be analysed at all: SEA @ COL (H1, and Colorado's starter unnamed), SF @ LAD and BOS @ TB (H3 — starter unnamed; Tampa Bay also has no DraftKings market).
• H3 partially met, not fully. The rule requires both starters confirmed on two independent sources. I confirmed via statsapi probablePitcher and cross-checked participants against the ESPN events feed, but ESPN's probables are referenced by athlete ID rather than name in the core API, so the second confirmation is of the matchup, not independently of the pitcher, for most games.
• Direct HTTP from the Linux sandbox is blocked (curl exit 56); all fetching went through the web-fetch tool, whose ~200-character URL cap also forced the player-stat queries into three batches without field filtering.
• ⚠ The site build needed a workaround — flagging for Mitch. python3 build.py fails at clean_dist() with PermissionError: [Errno 1] Operation not permitted trying to unlink dist/reports/2026-09-14.html. File writes to that mount succeed; only deletes are blocked from the sandbox. I rebuilt by executing build.py with that single stale.unlink() line skipped — every report page is regenerated and overwritten either way, and nothing has been archived out of reports/ today, so no stale page is left live. build.py itself was not modified. The 12:00 deploy should publish correctly, but the underlying delete permission is worth fixing, and running rebuild_and_deploy.bat from Windows would sidestep it entirely.
• No Hard Rule was violated. All prices are DraftKings, read from the ESPN core API. VegasInsider was not used.