Every dynasty site claims to be right. We publish the scoreboard: all 560 historical disagreements between our hybrid model and the market, wins and losses, computed with no lookahead. If a rankings site won't show you its misses, its hits aren't worth much.
Eleven times the model stood apart from the market and the seasons that followed settled it. These are the biggest, cleanest examples; the full 311–239–19 record above is the honest denominator.
The market beat us here, and each one taught us something. The five biggest misses, with the diagnosis.
The other side of the ledger: the five disagreements where the model was most right. Same data, same honesty.
Every historical valuation above used only data available at the time. For each preseason window we used NFL production from earlier seasons only, player ages as of September 1 of that year, and that year's rookie draft capital, with rookie priors drawn from earlier draft classes alone. No season outcomes, injury news, or results from that year or later entered any valuation. The "realized" ranks were computed from the following seasons' actuals (discounted career points), which were strictly unavailable when the model made its calls. That's what makes a backtest a backtest.
Why the model fades old producers even when they're still scoring: dynasty value is discounted career points, not next-season points. A 24-year-old projected at 15 ppg with 8 expected seasons left is worth more than a 32-year-old at 17 ppg with 3 left - and every future season is discounted 12%. The formula:
Projected seasons come from historical survival data (what fraction of players each age/position are still playing years later), with longer careers for high producers. Year-by-year scoring declines by position. One discount rate for everyone, tuned, not chosen, by walk-forward backtest.