2. Format Mapping & Methodology#
2.1 What the historical data actually shows#
The brief described the old format as a 12-week regular season with playoffs from Week 13. The real data (confirmed by every Sleeper league's settings: playoff_week_start = 14) shows the old format actually was:
| Phase | NFL weeks | What happened |
|---|---|---|
| Regular season | 1–13 | 10 teams, 2 divisions of 5; division foes ×2 + cross-division ×1 = 13 games, no byes |
| Playoff Rd 1 | 14 | 4 teams (championship bracket) — the top seeds had a BYE here — plus Sleeper-run consolation games for seeds 7–10 |
| Playoff Rd 2 (semis) | 15 | 4 teams (byes + Rd 1 winners) + consolation games for everyone else |
| Championship (BCC) | 16–17 | 2 teams, two-week combined score + consolation games / idle lineups for the other 8 |
Crucially, Sleeper recorded a real, scored lineup for every team in every week 1–17 — even weeks where a team had no official matchup (matchup_id = None). Those recorded scores power the backtest.
2.2 Old → New mapping used in this backtest (10-team adaptation)#
Historical divisions fielded 10 teams, not 12 (the expansion to 12×3 has not happened yet — 2026 is still 10-10-10). Per the agreed approach, the new-format rules are applied to each historical division's real 10 teams: seeds 1–10, five Week-17 Showdown matchups, AFP component ranks 1–10, bottom 3 relegated (30% of the tier, same proportion as 3-of-12).
| New-format week | Source of each team's game |
|---|---|
| Weeks 1–13 (regular season) | Real league matchups, real scores |
| Week 14 (regular season) | Real leg-14 games for the 8 teams that played (Rd 1 + consolation). The 2 bye teams: constructed optimal lineups (brief §5.3) from their reconstructed Week-14 rosters, paired against each other as "simulated opponents"; W/L by those scores |
| Week 15 (regular season) | Real leg-15 games — all 10 teams had a real matchup in every league |
| Week 16 (regular season) | Real leg-16 games (incl. BCC game 1); the 2 teams with no official matchup use their real recorded lineups and are paired vs each other |
| — seeds lock — | Seeds 1–10 by W-L (tiebreak: total PF, then head-to-head) |
| Week 17 Seeding Showdown | Pairings #1v#2, #3v#4, #5v#6, #7v#8, #9v#10; each team's real Week-17 lineup score; winner of each pairing takes the better placement rank |
2.3 AFP calculation (new rules)#
AFP = Showdown placement rank + H2H record rank (seed) + Points-For rank (lower = better)
Relegation: bottom 3 AFP scores in the tier.
AFP ties: 1) higher raw regular-season PF, 2) better raw W-L, 3) regular-season head-to-head, 4) alphabetical.
H2H and PF ranks are computed exclusively from Weeks 1–16; the Showdown rank exclusively from Week 17 — exactly per the brief. Statuses shown: 🏆 CHAMPION / 🥈 Runner-up / 🥉 Podium by placement; ❌ RELEGATED for the AFP bottom 3; ⚠️ Bubble for the two AFP ranks just above the drop zone; ✅ Safe otherwise.
2.4 Week-14 optimal-lineup construction (brief §5.3)#
For each of the 24 bye-team slots:
- Roster reconstruction — the team's end-of-season Sleeper roster with weeks 15–17 waiver/free-agent transactions reversed (adds removed, drops restored). Verified: every recorded Week-14 starter appears in the reconstructed pool (3 ghost-drop edge cases unioned back in from the recorded starters themselves).
- Player scoring — raw NFL Week-14 stats from the Sleeper stats API, scored under that league-year's exact scoring settings (the pipeline that reproduced all 22,554 historical player-week scores with zero error).
- Optimal legal lineup — exhaustive assignment over the season's real starting slots (which varied by year/division: e.g. 2022 Third = 9 slots, no IDP; 2025 Elite = 14 slots incl. DL/LB/DB + IDP_FLEX), maximizing total points with every slot filled by an eligible player (empty only if every candidate scores ≤ 0).