Synthefy's 2026 Fantasy Football Rankings: Predictions from 1,900 Player-Seasons of Data

1from synthefy_nori import NoriRegressor
2
3nori = NoriRegressor(model="nori-6m")
4nori.fit(X_2015_2025, ppr_through_week_16_capped_at_15_appearances)
5
6p10, p50, p90 = nori.predict(
7    X_2026,
8    output_type="quantiles",
9    quantiles=[0.1, 0.5, 0.9],
10)
11ranked_players = X_2026.assign(median_points=p50).sort_values(
12    "median_points", ascending=False
13)

1Help me build a reproducible fantasy football forecasting notebook with Nori-6M. Use Python and synthefy_nori.NoriRegressor for every model forecast.
2
3Use this published notebook as a reference implementation:
4https://github.com/Synthefy/synthefy-nori/blob/main/examples/notebooks/nori-fantasy-football-2026.ipynb
5
6Start by installing Nori:
7pip install synthefy-nori
8
9Use these free data sources:
10- nflverse (https://github.com/nflverse/nflverse-data/releases) for player statistics, rosters, player metadata, and NFL draft information.
11- Fantasy Football Calculator API (https://help.fantasyfootballcalculator.com/article/42-adp-rest-api) for preseason PPR average draft position and fantasy positions.
12
13For every backtest, use the preseason data recorded before that specific season began. Do not replace historical draft-position snapshots with today's data.
14
15Goal
16- Predict each player's low, median, and high season-long PPR point total.
17- Rank players from highest to lowest median forecast.
18- Export the final rankings as a CSV.
19
20Data
21- Build one row per player and season using only information available before Week 1.
22- Include preseason average draft position, age, professional experience, draft information, position, career games entering the season, and up to five prior seasons of fantasy production, usage, and games played.
23- Set a historical season to zero only when the player had not entered the NFL. Leave other unknown values missing.
24- Do not give the model player names, team names, season labels, future results, or any games-played value that was unknown before Week 1.
25- Use PPR points from each player's first 15 regular-season appearances through Week 16 as the result to predict.
26
27Evaluation
28- Replay at least two completed seasons. For each test season, use only earlier seasons as Nori's examples and use preseason data captured before that test season began.
29- Report mean absolute error for the median forecast. Lower is better.
30- Add checks that prevent future-season information from entering the inputs.
31
32Forecast
33- Create NoriRegressor(model="nori-6m") and fit it on every completed historical row.
34- Call predict with output_type="quantiles" and quantiles=[0.1, 0.5, 0.9]. Treat those values as the low, median, and high forecasts.
35- Sort the upcoming-season players by the 0.5 quantile, highest first.
36
37Before writing code, ask me to confirm the forecast season, scoring rules, player pool, and dated preseason data sources. Ask what additional football knowledge or signals I want to test beyond the notebook's baseline features, such as injuries, depth-chart changes, coaching changes, offensive line quality, or quarterback changes. Use an additional signal only if it was available before the forecast season began. Then list the columns you plan to use. Flag anything unavailable instead of inventing it. Explain each step for a reader who knows fantasy football but is new to machine learning.