Every number AutoTiers shows you is the output of four steps: blend the projections, apply situational adjustments, convert points into value over replacement, then cut the resulting curve into tiers. This page documents all four, including the constants we chose and why.
A ranked list implies precision that season-long projections do not have. The gap between the RB4 and the RB5 on a cheat sheet is usually smaller than the error bar on either projection — treating that ordering as meaningful leads to bad picks, because you reach for a name instead of taking the best available player at a position of need.
Tiers encode the thing that actually matters on the clock: where the real drop-offs are. If five running backs sit in the same tier, they are interchangeable enough that you should take the one who fits your roster, trade down, or address another position. If only one player remains in a tier, that is a genuine cliff and a reason to reach. The whole point of the pipeline below is to find those cliffs honestly rather than imposing a fixed number of tiers on the data.
AutoTiers starts from multiple projection sources plus last season's actual finish, and computes a weighted average across them. You control the weights in Settings.
The important detail is what happens when a source is missing. A naive weighted sum silently scores a player lower simply because one source had no line for them. Instead, AutoTiers computes the average over the active sources only — those that are both present and carry a non-zero weight — and redistributes the missing weight proportionally across the rest. A player covered by two of three sources is scored on the same scale as a fully covered player, preserving the relative importance you assigned.
Data confidence is expressed separately and explicitly, through the Projection Unavailable rule (a −50% penalty at default weight), rather than being smuggled in as an accidental side effect of arithmetic. Note also that a value of zero is not treated as missing: a player who genuinely scored zero points is a different case from a player with no stats on record.
Last season's point total understates a player who missed time, because the missing games are baked into the total as zeros. A back who averaged elite production across four games before getting hurt looks, by total, like a waiver-wire body.
So the prior-year term is scaled by games played before it enters the blend:
scale = clamp(games / 14, 0, 1). Fourteen is deliberately set below a full
NFL schedule so that near-complete seasons still receive full weight; rookies (no prior
season) and healthy 15–17 game seasons are unaffected. The constant is calibrated so that
an injury-shortened star's season lands within roughly one tier of their projection-only
score.
A steeper option squares that fraction, discounting partial seasons more aggressively — useful if you consider a half-season closer to noise than to signal.
The blended score is then adjusted by rules you switch on and weight yourself. Each rule is a transparent, bounded percentage adjustment, and every rule that fired on a player is recorded so you can see exactly why they moved. A representative sample of the defaults:
| Rule | Default effect | Trigger |
|---|---|---|
| RB Committee Penalty | −15% | Carry share under 50% |
| Target Share Premium | +7% | ≥25% of team targets |
| Declining Snap% | −10% | Offensive snap share dropped under 55% |
| New Team Penalty | −10% | Adjusting to a new team or scheme |
| New Head Coach | −7% | Team hired a new head coach |
| Sophomore Leap | +8% | Second-year WR/QB |
| Injury History | −12% | Played under 12 games last season |
| TD Regression | −10% | Scored 3+ more TDs than red-zone opportunity implied |
| Red Zone Usage Premium | +7% | 25+ looks inside the 20 |
| Over the Hill | −15% | Past the position's decline age |
| Projection Unavailable | −50% | No current-season projection from any source |
Some rules are informational only and change no score — Contract Year Flag, Handcuff RB, and Availability Risk exist to surface context on the clock without quietly moving a player for you.
The age thresholds in Over the Hill are position-specific rather than a single cutoff: RB at 28, WR and TE at 31, K at 40. Quarterbacks are split by playing style, because rushing production ages differently from pocket passing — a dual-threat QB (60+ rushing attempts the prior season) is treated as declining from 31, a pocket passer only from 38.
Raw projected points cannot be compared across positions. A quarterback projected for 320 points is not more valuable than a running back projected for 240, because the quarterback you could stream instead also scores a lot. What matters is the gap between a player and the player you would otherwise have gotten for free at that position — the replacement level.
AutoTiers sets that baseline per position, as a multiple of your league size:
| Position | Multiplier | Replacement rank in a 12-team league |
|---|---|---|
| QB (start 1) | 0.67 | ~QB8 |
| QB (superflex / 2QB) | 1.5 | ~QB18 |
| RB | 2.5 | RB30 |
| WR | 2.5 | WR30 |
| TE | 1.25 | ~TE15 |
| K / DST | 1.0 | 12th |
The QB multiplier is the one worth explaining. A 12-team league starting one quarterback does not have its replacement level at QB12 — it sits at the last reliably startable quarterback, around QB8. Anchoring at QB12 inflates elite quarterback value enough to crowd premium overall tiers ahead of elite receivers and backs, which is exactly the failure mode that makes people take a quarterback two rounds too early.
Superflex and 2QB leagues invert that pressure. With two quarterbacks starting, replacement drops roughly a full league deeper — to about QB18 rather than the literal QB24, with the difference reflecting the streamers who remain available on waivers.
With every player on a common VBD scale, the last step is finding the breakpoints. AutoTiers uses Jenks natural breaks, a one-dimensional clustering method that chooses cut points minimising variance within each group and maximising it between groups. In plain terms: it puts the boundaries where the data is already separating, which is precisely the cliff-finding job described at the top of this page.
Jenks is not used unconditionally. Above 11 overall tiers, AutoTiers switches to quantile splits (equal-size groups) instead. The reason is that the score distribution gets smooth at high tier counts, and past that point Jenks starts fitting noise rather than real gaps in player value — the goodness-of-variance-fit gain from each additional break falls below 0.002 around a dozen tiers, which makes 11 the last defensible Jenks count.
Positional tiers are capped separately, since positions differ in how many meaningful strata they actually contain: 5 each for RB and WR, 3 each for QB, TE, and DST, and 2 for kickers. Claiming eight distinct tiers of kicker would be false precision.
If clustering cannot run — every score identical, for example — AutoTiers falls back to quantile tiers rather than collapsing everyone into a single group.
Being clear about the limits matters more than the feature list:
Every tier list AutoTiers generates records which rules fired on which players. If a player sits somewhere surprising, that trail tells you which adjustment moved them and by how much — and if you disagree, you can down-weight or disable that rule and regenerate. The defaults are a starting position, not a verdict.