## Homepage per-pair percentage summary

espn: n=250 pairs (500 drafts); lowest -3.5458%; highest +14.2070%; arithmetic mean of paired percentage changes +4.0740%.
sleeper: n=250 pairs (500 drafts); lowest +0.1322%; highest +25.5274%; arithmetic mean of paired percentage changes +10.6895%.

The averages above are means of individual percentage changes. The aggregate relative lift below divides the two group means and is a different statistic.

# 1,000 draft simulations — internal planning benchmark

500 ESPN-format drafts and 500 Sleeper-format drafts: 250 paired comparisons per format. Algorithms were frozen before running; no new recommendation weights were fitted.

## Results

| Measure | ESPN weekly-all-play format | Sleeper elimination/Superflex format |
|---|---:|---:|
| Evaluation target | Season projected starter points | Week 1 projected starter points |
| DraftAid average | 2130.70 | 122.67 |
| Baseline average | 2048.81 | 110.97 |
| Relative projected lift | 4.00% | 10.54% |
| Higher paired score | 223/250 | 250/250 |
| Average projected rank, DraftAid / baseline | 1.65 / 5.50 | 1.14 / 9.49 |

## Scope and limitations

These are synthetic drafts against seeded rank/ADP-based bots, **not actual ESPN or Sleeper Auto-Pick**. The ESPN approximation matched 48/140 exact next picks in one observed room; Sleeper baseline fidelity has not been established. Neither number measures actual wins, elimination survival probability or championship probability. Both algorithms use the same projections being used to evaluate them.

ESPN: 10 teams, 14 rounds, 1QB/2RB/2WR/TE/FLEX/K/DEF. 25 seeds at each slot. The retained 250-player pool includes all 140 observed picks but not the full ESPN universe. Weekly data was missing, so only season totals were used. Rank baseline: +/-8 seeded noise, max 2QB/2TE, K/DEF in final two rounds and mandatory slot filling.

Sleeper: 18 teams, 12 rounds, QB/2RB/2WR/TE/FLEX/Superflex, four bench spots, linked PPR scoring. 14 seeds in slots 1–16, 13 in slots 17–18. The saved September 9 dataset supplies 383 QB/RB/WR/TE players with positive season and Week 1 projections. Players with zero/missing weekly points are excluded from both arms; this restricts the pool and omits some future-value/injured assets. The missing observed Travis Hunter DB entry is excluded. This is not an exact replay of that room. Current public weekly coverage was checked for all retained IDs; saved values were retained.

Sleeper baseline: ADP +/-8 noise; starter-need bonus 8; max 3QB/2TE; force mandatory holes late. To avoid an incomplete baseline roster, a team with no QB prioritizes one when available QBs fall to the count of QB-less teams plus four. A preliminary attempt without this protection failed roster validation and was discarded before the 500-draft run. Both arms use the corrected opponent policy.

Pairs share player data, slot, rules and counter-based preference randomness. Changing DraftAid's picks can change subsequent opponent choices. Seeds reused across slots are correlated scenarios; this is not 500 independent real-world trials. Earlier 200-run ESPN cases overlap this expanded experiment and must not be added as independent evidence.

No FAAB bidding, waiver acquisitions, eliminated-roster redistribution, calibrated score variance, NFL actual results or multiweek championship simulation is included. These results can support a clearly labeled internal projected-lineup simulation card, not an assertion of superior league-winning probability. Live screenshot comparison results remain separate.

## Verification

All 1,000 drafts completed. All 178,000 selections were unique within each draft. All 14,000 rosters had complete eligible starters. An independent position-based scorer matched every team total. All 6,500 recommendation calls returned picks. Source/data hashes and full pick histories are retained alongside this report.

## ESPN slots

| Slot | Pairs | DraftAid | Baseline | Lift |
|---|---:|---:|---:|---:|
| 1 | 25 | 2134.46 | 2068.54 | 3.19% |
| 2 | 25 | 2124.88 | 2067.72 | 2.76% |
| 3 | 25 | 2130.86 | 2057.67 | 3.56% |
| 4 | 25 | 2135.84 | 2048.28 | 4.28% |
| 5 | 25 | 2131.68 | 2057.04 | 3.63% |
| 6 | 25 | 2142.15 | 2064.32 | 3.77% |
| 7 | 25 | 2143.60 | 2043.33 | 4.91% |
| 8 | 25 | 2133.82 | 2010.08 | 6.16% |
| 9 | 25 | 2116.77 | 2027.84 | 4.39% |
| 10 | 25 | 2112.96 | 2043.30 | 3.41% |

## Sleeper slots

| Slot | Pairs | DraftAid | Baseline | Lift |
|---|---:|---:|---:|---:|
| 1 | 14 | 126.21 | 113.02 | 11.67% |
| 2 | 14 | 125.33 | 108.76 | 15.24% |
| 3 | 14 | 125.40 | 110.21 | 13.79% |
| 4 | 14 | 125.41 | 109.97 | 14.04% |
| 5 | 14 | 125.38 | 112.24 | 11.71% |
| 6 | 14 | 124.97 | 112.30 | 11.28% |
| 7 | 14 | 124.15 | 112.39 | 10.46% |
| 8 | 14 | 123.62 | 113.03 | 9.36% |
| 9 | 14 | 123.38 | 110.59 | 11.56% |
| 10 | 14 | 122.01 | 112.45 | 8.50% |
| 11 | 14 | 122.07 | 111.42 | 9.56% |
| 12 | 14 | 121.34 | 111.66 | 8.68% |
| 13 | 14 | 120.56 | 112.35 | 7.30% |
| 14 | 14 | 120.47 | 109.52 | 9.99% |
| 15 | 14 | 119.88 | 109.72 | 9.27% |
| 16 | 14 | 119.40 | 109.27 | 9.27% |
| 17 | 13 | 118.72 | 108.62 | 9.29% |
| 18 | 13 | 119.20 | 109.70 | 8.66% |

## Reproduce

Run node espn.cjs and node sleeper.mjs, then node report.cjs. They use frozen engines in this directory. Sleeper room rules come from ../../actual-draft.json. Do not treat this season-only/Week-1 split as a full-season survival model.