Methodology

FiveThirtyEight-style, in spirit: show the distribution, not just the point estimate.

How the numbers are built — sample model

1. Median (P50). The central projection for the week, in full PPR. On the weekly pages the median is the median of the player's actual weekly lines through Week 9 (Pro Football Reference).
2. Volatility (SD). How wide the range is. High-volatility players have a higher ceiling and a lower floor — the model shows both.
3. Distribution. A normal-distribution approximation turns median + SD into P10 / P50 / P90, and into probabilities: boom, bust, and Top-12 at the position.
4. Head-to-head. Start/sit and trade comparisons use the difference of two normals to get a win probability.
5. Rest-of-season. A separate per-game ROS median, used only for trade context — not for this week's start/sit call.

Where the edge is

The top of the board is not where this wins. The money is in out-researching key waiver pickups, PAR trades, and the middle of the draft. The top tiers are held to a different standard: they are built to be safe, and if a top call is wrong, it should be because of injury — not a misread.

What this model does not do

• It is not trained on real game logs — the inputs are each player's actual weekly lines through Week 9, and the distribution is a normal approximation around the real median and real weekly standard deviation.
• It does not account for injuries, weather, or late-week news unless the median / SD are updated by hand. No player on injured reserve is ranked as playing.
• It does not use Sleeper, FantasyPros, or any other ranking source. Your preference for a consensus mode across about five sources in full PPR is noted for a live version — it is not implemented here, because there is no live source connected.
• It is not betting advice, and it is not a real forecast.

What's still placeholder

• Weekly player data is real now — src/lib/real-week-10.ts (through Week 9, Pro Football Reference). The old sample set in src/lib/data.ts is no longer shown on any weekly page; that file is kept only for the Substack URL.
• Substack — live at https://routetreereport.substack.com (created Nov 10, 2026). The signup form sends readers to the real Substack subscribe page; a local browser copy is kept only as a backup
• No live data source, no auth, no payments, no email backend
• No live data source, no auth, no payments, no email backend

What a live version would need

• A weekly projections source (with a licence / API key, stored securely — never in the repo)
• Actual game logs to estimate volatility per player, not a hand-set SD
• Matchup adjustments that are computed, not hand-written
• Real Gumroad and Substack URLs, and a real signup backend