We publish our own scorecard, including the gates we haven't cleared.

Every model, every gate: exactly as it scored.

Each model carries one of three honest states, set by the live backtest and never by hand: validated (cleared its pre-registered gate), market test pending (can't be tested yet), or in development (a bar it hasn't met). Here is why each sits where it does:

• Dynasty DNA Value earns a real signal: it sorts players top-to-bottom on future production, in order. But we can't yet test it against the market: our own value history only begins in 2026, so there is no market baseline to out-rank. That's untestable for now, not a failure.
• Projection (the published equation) is a calibrated range, graded on whether its error beats a 2-year weighted average by the 3% our gate demands. Its verdict is in the ledger, set by the live backtest, and so is that of an experimental age-adjusted variant we publish beside it.
• Breakout / bust are graded on their walk-forward (out-of-sample) AUC against a 0.58 floor, never the flattering in-sample figure. Each signal's verdict is in the ledger; neither is sold as predictive unless its held-out AUC earns it.

Dynasty DNA Value

The Dynasty DNA Value is our own dynasty value on a fixed 0–10,000 scale, built from an additive model of a player's production and then calibrated onto that range. It is not the market value: the public market value is what the market pays today; the Dynasty DNA Value is what our model says a player is worth. We show the two side by side so you can see where they disagree. It is glass-box: the waterfall on every player's Strand sums to the model's raw score, line by line. No black box. The inputs:

• Base: recency-weighted recent PPG, the production floor a player starts from.
• × Age: the player's spot on his position's aging curve, peak-aware.
• + Production trend: the season-over-season percentile slope.
• + Trajectory: a phase adjustment for ascending or declining players.
• + Scarcity: value above positional replacement level.
• + Draft capital: a centered draft-pedigree prior that fades to zero once a player has a few seasons of real production behind him. No draft record means no adjustment, never a guessed discount.

Those are all of them. This page also used to list an Opportunity term and a Risk term; the published equation computes neither, so they are gone rather than quietly restated. The lines sum to the raw score on a points-per-game scale, not to the 0–10,000 value. The value is the next step: the raw score places the player on the market's own sorted value ladder, so the scale is the market's and the placement is ours.

The earned claim above is the proof it works as a ranking: sort by Dynasty DNA Value and real future production falls in order. That ordering is measured out of sample. The gate on this model itself has not been met: the 0–10,000 scale is calibrated to market, and whether it out-ranks the market is a separate, harder test we can't run yet. See the ledger.

Trajectory: phase, breakout / bust

Trajectory reads where a player is on his arc, and never hides how:

• Phase blends the aging-curve prior (the sign of the curve's slope at his age) with the player's own 2-season percentile slope. When they disagree, the player's own data gets the vote at lower, disclosed confidence.
• Breakout / bust are logistic models fit on labeled history, applied with shown coefficients. Each 0–100 score is a sum of named contributions, not an opaque grade. We grade them on their walk-forward AUC, shown honestly in the ledger.
• Regression-to-mean shrinks thin samples toward the positional mean by reliability weight (w = n / (n + K)). Few games trust the prior, a full sample trusts the player.

Projection: the honest band

A projection should be a range: a floor, a ceiling, and the zone most outcomes land in. Never a single fabricated decimal. Ours is a published equation with five named stages, and its range is calibrated: the share of real outcomes that land inside it sits within our pre-registered 80%±7 target (the exact figure is in the ledger).

On error, the published equation is graded against that same 2-year weighted average. It must beat it by the 3% our gate requires to claim an accuracy edge. Whether it does today is the ledger's call, and that verdict drives the accuracy badge on every Strand. An experimental age-adjusted variant is graded the same way, separately. It is published unproven, not hidden.

A failure you can see beats a number you can't.

This page reports gates we haven't cleared, on purpose. A scorecard is only worth something if it can come back negative, and ours does, in public, with the exact evidence beside it. Nothing here is hand-set: the statuses, the metrics, and the criterion-by-criterion verdicts all read from the live backtest. The day a gate clears, the same endpoint that shows you today's shortfall will show you the win, and you'll be able to check that too.

• Every number on screen is live model output, or clearly labeled as developing, in-validation, or untestable.
• Glass-box always: every verdict is followed by its receipts.
• A projection is a range, never a fabricated decimal.
• No “out-ranks the market” claim until the public scorecard earns it.