RESEARCH NOTEBOOK — 2026 Preseason Week 2 (seat 4, claude-opus-5)
GOAL PRIORITY: exact scorelines > exact team scores > correct outcomes > SSE. Because per-game exact-scoreline probability in the NFL is only ~1%, the realistic differentiators are (a) exact individual team scores and (b) outcomes. So the method is:
- Derive market-implied scores from spread + total: fav = (T+S)/2, dog = (T-S)/2.
- Round each side to the highest-frequency NFL team scores (17 > 20 > 24 > 13 > 21 > 10 > 16 > 23),
preferring the frequent value that stays closest to the implied number and preserves the market's winner. This trades a tiny amount of SSE for meaningfully higher hit rates on the higher-priority "exact team score" tiebreaker.
- Never predict a tie (preseason tie rate ~2-3%; a wrong tie costs 25 in the outcome dimension).
EVIDENCE (two dated market snapshots + playing-time reporting)
- bookmakersreview.com, Aug 17 board (sr_d265267d6fc3884840685de8): all 16 games; every total
36.5-40.5, board average 38.2; 14 of 16 spreads <= 2.5; DEN -4 (-205) heaviest chalk; four road favorites (LV, NO, NYG, KC); unders shaded in LAR/TB/IND/DET/TEN, over shaded in CIN.
- Yahoo/The Big Lead betting guide, Aug 20-21 (sr_3eea693f8afbd75dbc9aa355): later numbers —
DET -4.5 (37.5), CLE -1.5 (36.5), IND -3.5 (40.5), MIN -2.5 (36.5), NO -3 (37.5), NYG (36.5), CIN -1.5 (40.5), PHI -3.5 (36.5), TB -5.5 (36.5), ARI -1.5 (37.5), TEN -4.5 (38.5).
- FoxSports/Sporting News board (sr_6d1a4c9564b6b744a8457434): DEN -6 with O/U 41.5; NYG -2.5;
TB -5.5; TEN -4.5. VSiN (sr_68f34b7bfb144df64df0ce1c) has DEN -6.5 and notes GB lost 28-9 in Wk1 with Jordan Love expected to play a little; Cardinals have scored 30 and 27 in two outings.
- Covers model picks (sr_8aafd4e6d596804989d32172) broadly agree with favorites: DEN 67%, DET 64%,
CLE 56%, IND 64%, MIN 62%, NYG 57%, TB 67%, TEN 64%.
KEY PLAYING-TIME NOTES
- GB@DEN: biggest spread on the board; DEN moneyline -205/-245. Love brief, Tyrod Taylor after.
- WAS@DET: Goff sits, Dobbs gets heavier load; Campbell said Lions will play starters. Washington
without Mariota, no Jayden Daniels; Hartman/Kaliakmanis, "good amount" of starters per Quinn.
- BUF@CLE: Josh Allen not expected to play; Shedeur Sanders starts for Cleveland at home.
- ATL@IND: BOTH teams sitting starters (joint practices took the reps). Jack Strand vs Riley Leonard.
Total nonetheless drifted up to 40.5.
- BAL@MIN: Murray not expected to play; J.J. McCarthy extended look; Ravens ran Joe Fagnano heavy.
- NO@LAR: McVay sits Stafford again; Shough gets real work after sitting Wk1. Road favorite NO.
- NYG@MIA: Giants QB room thin (Haener just signed, Brandon Allen struggling); Miami down to
Willis/Miller/Gronowski with Ewers missing practice. Market still likes NYG by 1.5-3.5.
- CHI@CIN: week's highest total (40.5, over -115). Burrow limited (13 snaps Wk1, banged up);
Caleb Williams may get a small taste; Bagent hamstring questionable.
- PHI@NE: line flipped from NE -1.5 (Aug 17) to PHI -3.5 (Aug 20-21). Neither Hurts nor Maye likely.
NE tied the Colts in Week 1 — reminder that preseason ties happen, but still not worth predicting.
- KC@TB: biggest line move of the slate — KC -1.5 to TB -5.5 after Bowles confirmed Mayfield plays
(he sat all of last preseason). Mahomes out all preseason; Fields/Nussmeier for KC. Lowest total (36.5, u -120).
- DAL@ARI: Prescott out; Milton likely starts over Howell. Cardinals cautious in their 3rd outing
(HOF game included) and Jeremiyah Love out, which caps the Arizona blowout scenario.
- SEA@TEN: TEN -4.5 (up from -1.5). Cam Ward home rebound spot, Levis heavy work, Trubisky banged up;
Seattle leaning on Lock/Milroe with Darnold minimal.
DERIVATIONS AND FINAL PICKS (implied -> chosen) GB@DEN T~40.5 S6.0 -> 17.3/23.3 -> GB 17, DEN 24 (margin 7, total 41) WAS@DET T 37.5 S4.5 -> 16.5/21.0 -> WAS 17, DET 21 (exact fit on DET) BUF@CLE T~38.0 S1.5 -> 18.3/19.8 -> BUF 17, CLE 20 ATL@IND T 40.5 S3.5 -> 18.5/22.0 -> ATL 17, IND 21 (both starter-less; shaded to under) BAL@MIN T 36.5 S2.5 -> 17.0/19.5 -> BAL 17, MIN 20 (near-perfect fit) NO@LAR T 37.5 S2.5 -> LAR 17.5 / NO 20.0 -> NO 20, LAR 17 NYG@MIA T 37.0 S2.5 -> MIA 17.3 / NYG 19.8 -> NYG 20, MIA 17 CHI@CIN T 40.5 S1.5 -> 19.5/21.0 -> CHI 20, CIN 23 (highest total on board, over-juiced) PHI@NE T 37.0 S2.5 -> NE 17.3 / PHI 19.8 -> PHI 20, NE 17 KC@TB T 36.5 S5.5 -> 15.5/21.0 -> KC 17, TB 21 (kept KC at the modal 17 rather than 15/16) DAL@ARI T 37.5 S1.5 -> 18.0/19.5 -> DAL 17, ARI 20 SEA@TEN T 38.5 S4.5 -> 17.0/21.5 -> SEA 17, TEN 21 (near-perfect fit)
Predicted totals average 38.0, matching the board average of 38.2. All 12 winners follow the current market favorite; expected outcome accuracy ~6-8 of 12, which is about the ceiling for August exhibitions. Expected exact-team-score hits ~1.9 (eleven 17s, seven 20s, four 21s).
MAIN RISKS: (1) preseason variance — blowouts (28-9 type results) and 9-6 slogs are common and un-forecastable; (2) late scratches after my research window; (3) concentration on 17/20/21 means a slate that skews high (30s) or low (single digits) hurts my SSE broadly rather than in one game.
