Daily Bet Research Report — Monday, September 21, 2026

Start: 11:10 ET | End: 11:17 ET

Bankroll: $60.23 | Stake (12.5%, one play): $7.53 Record: 3-1-0 (Main bets, main leg, since 9/14 reset) | Hit rate: 75.0% Slate: 12 scheduled / 4 bookable · 4 analysed · 26 sources · Playbooks: MLB, NFL · Verified at 11:10 ET

Section 1: Verification

Verified against statsapi schedule (MLB) and ESPN core events (NFL, NHL) at 11:10 ET — 12 games scheduled.

• MLB — 3 games. All three are series openers. Probable pitchers confirmed from statsapi probablePitcher and cross-checked against Baseball Savant's 2026 qualified-pitcher leaderboard (H3 second source) for Yesavage, Baz, Matthews and Tidwell; for River Ryan and DJ Herz the second source was statsapi career stats.
• NFL — 1 game. NYG @ LAR, Week 2 Monday nighter at SoFi.
• NHL — 8 games, all season.type: 1 (preseason). DraftKings returns count: 0 on the ESPN odds endpoint for these — tested on two of them (NYR @ NJ, COL @ WPG). They are not on the board, so they are not bookable and were not analysed. Details in Research Limitations.

Bookable slate is therefore 4 games, and all 4 were analysed.

Section 2: Game Analysis

Unanalysed: 8 of 12. All eight are NHL preseason. Two were confirmed unpriced by direct odds fetch; the other six were not individually fetched, and I am not claiming otherwise. Separately, the NHL playbook is a stub marked "NOT YET BUILT", and preseason goalie assignments are not published at 11:10 AM — so even a priced preseason game would have failed the confirmed-goalie requirement. Nothing was skipped for convenience.

Section 3: Today's Plays

Negative-edge flags: #1 (−1.9%), #3 (−0.1%) and #4 (−3.1%) all price above my calibrated confidence. Only #2 clears its own price, and by 0.4 points — which is noise, not an edge. This is a four-game bookable board with no soft spot on it; treat the whole card as thin and read Section 4 before staking anything.

All prices DraftKings, from the ESPN core odds endpoint. One pricing conflict to check before betting — see #2.

Section 4: Reasoning

#1 — Washington Nationals +1.5 (1.59)

This is the highest-probability ticket on the board, and the ranking rule is calibrated confidence, so it leads — but its rank comes mostly from run-line mechanics, not from an informational edge, and I want that stated plainly. The substantive case is the Detroit starter. River Ryan's entire MLB career is 4 starts and 20.1 innings (1.33 ERA, 9 walks, 295 pitches), all with the Dodgers, and statsapi returned no 2026 season split for him at all — he is effectively making a comeback start. MLB playbook H4 blocks a play on a starter with fewer than 3 career starts; Ryan has 4, so H4 does not bind, but H4's own carve-out is that a play built against that starter's inexperience is the legitimate use of the situation, and that is exactly this. Detroit is nonetheless priced -156, roughly a 61% favourite, over a Washington club with an identical 73-83 record. DJ Herz is no prize (career 19 GS, 4.16 ERA, 3.65 BB/9) but he is a known quantity with 88.2 career innings. Tier-1 supporting number: Comerica wind at first pitch is 6.1 m/s = 13.6 mph from 27°, below the 15 mph threshold, 62°F and overcast — no weather argument either way, and the playbook forbids guessing Comerica's centre-field bearing, so speed only. Line movement: the run line drifted from WSH +1.5 at -181 to -167, i.e. away from my side, which per Step 3 means sharp money is taking Detroit and I should discount accordingly — I have. Evidence tag [MODERATE] (§2 starting pitcher, §7 markets). The risk that beats me: River Ryan was a genuine prospect and a four-start 1.33 ERA is not nothing; if he throws five clean innings against a bad Washington lineup, Detroit wins by three and this loses. I also could not obtain xERA for either starter — neither reaches Savant's min=q cut — so the game's central matchup is the least-measured on the card. Source opened: statsapi.mlb.com/api/v1/people/689981/stats.

#2 — New York Giants +8.5 (1.90)

The only play on the board with a positive edge, and the case is the §2 fallback working as designed. Week 1: Giants 28 points on 395 yards; Rams 7 points on 290 yards. Turnover differential — the luck flag §2 exists to check — is NYG 0 (1 takeaway, 1 giveaway) and LAR −1 (1 takeaway, 2 giveaways), so the Rams' bad week was not turnover variance being unkind to them; it was a genuinely poor offensive performance, with Stafford 15/25 for 155 and an interception. The Rams are still laying between 7 and 8.5 at home, and playbook H4 caps home-field at 2.0 points, so the market is rating Los Angeles roughly 5.5–6.5 points better on neutral ground off that evidence. Key numbers matter here [STRONG]: +8.5 sits on the far side of 7, and ~7–9% of NFL margins land exactly on 7. H1 check: presumptive starters only — Jaxson Dart (23/29, 230, 3 TD in Week 1) and Matthew Stafford — no injury feed available today; the ESPN summary endpoint returned ten Questionable designations all dated 19–23 August, stale preseason tags with zero Out or Doubtful entries and no quarterback among them. The H1(b) line-move block does not trigger: the spread moved 1.5 points (under the 2-point trigger), landed on rather than through 7, and the total moved only 1.0 (under the 1.5-each opposite-direction trigger). Line movement and a pricing conflict you must resolve: ESPN's two feeds disagree. The core odds endpoint reports current LAR -8.5 (-110), opened -8.5; the site scoreboard reports LAR -7 (-118) / NYG +7 (-102), opened -8.5, with the total up from 47.5 to 48.5. If DraftKings is actually at +7, this play loses its key-number value and I would not make it. Check the app before staking. Evidence tag [MODERATE] at best — n=1 game, which is a genuinely weak sample and I am not dressing it up as more. The risk that beats me: one week is one week, the market's preseason prior on the Rams' roster is far stronger than my two data points, and a 24-10 Rams win covers comfortably. Source opened: sports.core.api.espn.com/.../seasons/2026/types/2/teams/14/statistics.

#3 — Toronto Blue Jays ML (1.92)

The single largest measured starter gap on the card, priced as a coin flip. From Baseball Savant's 2026 expected-statistics leaderboard: Trey Yesavage — 3.65 ERA, 3.38 xERA, era_minus_xera_diff +0.271; Shane Baz — 4.11 ERA, 4.18 xERA, diff −0.07. That is an 0.80-run xERA separation in Toronto's favour, and both gaps sit well inside the playbook's 1.00-run regression flag, so neither pitcher is being flattered or punished by luck — the xERA gap here is not a regression signal, it is a straight statement that Yesavage has been the better pitcher on contact quality. Yet DraftKings has Baltimore -111 and Toronto -108. The market is paying for home field, which is the only thing Baltimore has: both clubs are eliminated (77-79 and 75-81), so §6's September lineup caveat cuts both ways and I am not leaning on records, per §6's explicit instruction. H4 satisfied: Yesavage faced 386 batters in 2026 and Baz 742, both far past the 3-start floor. Line movement is flat — the moneyline opened -108/-112 and sits -108/-111, so nothing has happened that I am late to. Evidence tag [MODERATE], per §2's honest limitation that xERA is the best obtainable estimator rather than the best one. Tier-1 weather number: Camden at 6:35 PM is 3.6 m/s = 8.1 mph from 33°, 71°F — wind is a non-factor, but MET Norway has rain arriving from roughly 8 PM ET and turning heavy overnight (0.9 mm at 00:00 UTC, 1.2 mm at 01:00 UTC). The risk that beats me: rain. A delay or a shortened game is a live possibility, and a postponement voids the bet rather than losing it — annoying, not costly — but a five-inning rain-shortened game hands more variance to the worse team. Source opened: baseballsavant.mlb.com/leaderboard/expected_statistics.

#4 — Minnesota Twins ML (1.85)

The under-priced factor is bullpen state, which §3 calls the slowest-moving input the market carries. Minnesota's leverage arms did not pitch last night. The Twins won 8-0 at the Angels and the save went to Bailey Ober — a starter absorbing bulk relief innings in a blowout, which is what a club does precisely to avoid using its late-inning group. San Francisco, by contrast, lost 1-3 at the Dodgers with reliever Sam Hentges taking the loss, so the Giants worked their late innings in a close game. That is a one-night edge, not the three-consecutive-day breakpoint §3 identifies as the documented decline threshold, so I am weighting it as support, not as the argument — [MODERATE], leaning weak. The starters are a genuine wash and I will not pretend otherwise: Zebby Matthews 4.74 ERA / 4.45 xERA (+0.289) against Blade Tidwell 4.56 / 4.48 (+0.076) — three hundredths of a run apart on expected ERA. Tier-1 weather number: Oracle Park at 9:45 PM is 4.8 m/s = 10.7 mph from 250°, 61°F and overcast. The playbook permits an out/in call at Wrigley only, so this is speed-only: no wind argument. Line movement is essentially nil (MIN -118 to -117). Given a 73-83 road team against a 64-92 home team priced near pick'em, the market is charging full freight for Oracle's home edge, and I think it is charging slightly too much — but only slightly, which is why this ranks last and why its edge is the worst on the card at −3.1%. The risk that beats me: I never obtained a three-day bullpen usage picture for either club — one night's decisions is a thin proxy — and Oracle genuinely is a hard place for a visiting offence. Source opened: statsapi.mlb.com/api/v1/schedule?date=2026-09-20&hydrate=decisions.

Sources

URLs actually fetched this run (26 successful, 1 failed):

MLB — statsapi 1. https://statsapi.mlb.com/api/v1/schedule?sportId=1&date=2026-09-21&hydrate=probablePitcher 2. https://statsapi.mlb.com/api/v1/schedule?sportId=1&date=2026-09-20&fields=... (final scores) 3. https://statsapi.mlb.com/api/v1/schedule?sportId=1&date=2026-09-20&hydrate=decisions 4. https://statsapi.mlb.com/api/v1/people/689981/stats?stats=season,career&group=pitching&season=2026 (River Ryan) 5. https://statsapi.mlb.com/api/v1/people/687792/stats?stats=season,career&group=pitching&season=2026 (DJ Herz) 6. https://statsapi.mlb.com/api/v1/people/689981?hydrate=stats(type=season,season=2026)&fields=... 7. https://statsapi.mlb.com/api/v1/game/824546/boxscore?fields=... (DET @ CWS) 8. https://statsapi.mlb.com/api/v1/game/823896/boxscore?fields=... (SF @ LAD)

MLB — pricing and metrics 9. https://sports.core.api.espn.com/v2/sports/baseball/leagues/mlb/events?dates=20260921&limit=100 10. https://sports.core.api.espn.com/v2/sports/baseball/leagues/mlb/events/401817027/competitions/401817027/odds (TOR @ BAL) 11. https://sports.core.api.espn.com/v2/sports/baseball/leagues/mlb/events/401817028/competitions/401817028/odds (WSH @ DET) 12. https://sports.core.api.espn.com/v2/sports/baseball/leagues/mlb/events/401817029/competitions/401817029/odds (MIN @ SF) 13. https://baseballsavant.mlb.com/leaderboard/expected_statistics?type=pitcher&year=2026&min=q&csv=true

NFL 14. https://sports.core.api.espn.com/v2/sports/football/leagues/nfl/events?dates=20260921&limit=100 15. https://sports.core.api.espn.com/v2/sports/football/leagues/nfl/events/401872947/competitions/401872947/odds 16. https://site.api.espn.com/apis/site/v2/sports/football/nfl/scoreboard?dates=20260921 17. https://site.api.espn.com/apis/site/v2/sports/football/nfl/summary?event=401872947 18. https://sports.core.api.espn.com/v2/sports/football/leagues/nfl/seasons/2026/types/2/teams/19/statistics (NYG) 19. https://sports.core.api.espn.com/v2/sports/football/leagues/nfl/seasons/2026/types/2/teams/14/statistics (LAR)

NHL 20. https://sports.core.api.espn.com/v2/sports/hockey/leagues/nhl/events?dates=20260921&limit=100 21. https://site.api.espn.com/apis/site/v2/sports/hockey/nhl/scoreboard?dates=20260921 22. https://sports.core.api.espn.com/v2/sports/hockey/leagues/nhl/events/401879645/competitions/401879645/odds (returned count: 0) 23. https://sports.core.api.espn.com/v2/sports/hockey/leagues/nhl/events/401878107/competitions/401878107/odds (returned count: 0)

Weather — MET Norway 24. https://api.met.no/weatherapi/locationforecast/2.0/compact?lat=37.7786&lon=-122.3893 (Oracle Park) 25. https://api.met.no/weatherapi/locationforecast/2.0/compact?lat=42.339&lon=-83.048 (Comerica Park) 26. https://api.met.no/weatherapi/locationforecast/2.0/compact?lat=39.284&lon=-76.6217&altitude=10 (Camden Yards)

Failed: https://sports.core.api.espn.com/v2/sports/baseball/leagues/mlb/events/401817296/competitions/401817296/odds — empty body. My error: I guessed the event ID before fetching the events list. Refetched correctly at #9–12.

Research limitations

The three counts, leading:

• Games analysed: 4 of 4 bookable (12 scheduled across all four sports). This is below the 8-game floor, and the floor was not met. The reason is the slate, not the effort: MLB has only 3 games on a Monday this late in September, the NFL has a single Monday-night game, the NBA is out of season, and all 8 NHL games are preseason with no DraftKings price. The floor's own escape clause — "or the whole bookable slate if fewer" — applies, and I analysed the whole bookable slate.
• Sources fetched: 26 successful, 1 failed. Above the 15 floor.
• Games where Tier-1 factors were unavailable: 0. Every analysed game carries at least one Tier-1 factor with a real number. Partial gaps are itemised below.

What I could not obtain:

• xERA for DJ Herz and River Ryan. Neither reaches Savant's min=q qualified cut, so the WSH @ DET starting-pitcher matchup — the game carrying my top-ranked play — rests on career counting stats rather than the playbook's §2 metric. I substituted statsapi career lines rather than guessing, and did not fall back on FIP, which §2 says adds nothing over ERA.
• Three-day bullpen usage for any team. I attempted per-pitcher extraction from two statsapi boxscores (DET @ CWS, SF @ LAD). Both saved to files as expected, but the pitching stats interleave game and season totals and the player names are not adjacent to the pitching objects, so no regex I tried paired a name to an innings count. This is not a truncation problem — the files were reachable and greppable. It is a JSON-shape problem, and the cheap workaround I used instead was hydrate=decisions on yesterday's schedule, which yields winner/loser/save and therefore tells you which closers worked. That is a real but much coarser signal, and §3's documented three-consecutive-day breakpoint is not testable from it. Worth adding to the MLB playbook as a known shape problem with the boxscore endpoint.
• Statcast park factors — not fetched. §4 says to use Statcast's multi-year figures and explicitly forbids inferring park effect from this season's scores. I did neither: I made no park-factor claim for Oracle, Comerica or Camden beyond the measured wind and temperature. The unverified source baseballsavant.mlb.com/leaderboard/statcast-park-factors remains untested.
• Umpire assignments — not fetched, still unverified. §5 values them at 0.3–0.5 runs. No total was played today, so nothing turned on it.
• The NFL injury feed is confirmed thin, exactly as the playbook predicted. site.api.espn.com/.../summary?event=401872947 returned ten designations, all Questionable, all dated 19–23 August — preseason tags never cleared — with zero Out or Doubtful and no quarterback among them. Puka Nacua (LAR) and Malik Nabers (NYG) both carry stale August Questionable flags; those are not current status and I did not treat them as such. H1 was satisfied via the presumptive-starter path. Official inactives drop ~6:45 PM ET for an 8:15 kickoff — after this run. Check late.

Freshness check: passed. All four analysed games were confirmed against the league schedule at 11:10 ET, and every pitcher, quarterback and price cited belongs to a game on tonight's board.

Playbooks: MLB (v1, 2026-09-16) and NFL (v1, 2026-09-20), both real and both read in full. No stub playbook was used — the NHL stub was not invoked, because no NHL game was analysed. No Hard Rule was violated: MLB H1 (no Coors) n/a, H2 (no Wrigley) n/a, H3 both starters double-sourced, H4 checked explicitly on Yesavage, Baz, Matthews, Tidwell and Ryan, H5 all four plays inside 1.50–2.10. NFL H1 satisfied via presumptive starters with the line-move block checked and not triggered, H2/H3 n/a (no total played; SoFi is roofed in any case), H4 home-field capped at 2.0 and cited in play #2, H5 in band.

Two things to treat with suspicion:

1. ESPN's two feeds disagree on the NYG @ LAR spread — core says LAR -8.5 (-110), site scoreboard says LAR -7 (-118). I priced from the core endpoint per the playbook, but play #2's key-number argument depends on the 8.5 being real. Verify in the DraftKings app before staking.
2. The calibration subtraction is still running on a mostly-prop sample with very little game-line history behind it. Four plays today, all at or below a 61% calibrated figure, all but one with a negative edge. That is a thin, efficiently-priced board, and "the ranking says #1" is not the same as "this is a good bet."

Cutoff: not hit. End stamp read at 11:17 ET, well inside the 11:50 limit.