Example Draft After-the-Fact Record Keeping

I used AI to develop a prompt to create these draft records for postseason analysis. I urge you to do the same for testing and personal improvement. 

The Science of Fantasy Football

How to Grade a Draft Without Trusting a Single Number

A fantasy draft is not a list of names. It is a portfolio assembled under scarcity, time pressure, and incomplete information. Most platforms grade that portfolio with one score: pick versus average draft position, or pick versus projected points. That is accounting. It is not analysis. A more scientific grade treats the roster the way a risk desk treats a book: multiple independent lenses, explicit disagreement, starter quality separated from bench noise, and a written record that can be scored after the season. What follows is a process used across several 12-team PPR, one-quarterback, tight-end-premium snake drafts. The goal is not a prettier letter grade. The goal is a verdict you can falsify in January.

Why single-metric grades fail

Commercial draft report cards usually lean on some mix of:

• pick versus ADP

• pick versus expert consensus rankings

• projected seasonal points

• positional balance

• sometimes a playoff-odds simulation

Those tools are useful and incomplete. ADP is a crowd price, not a true value. Expert rank is a board, not a market. Projections collapse a distribution into a mean. Balance counts bodies, not roles. A simulation that assumes independence between a quarterback and his running back is lying about football. Worse, most grades average the whole roster. A championship core plus six zero-floor darts looks average. An average core plus tidy depth looks fine. Those are different teams. A scientific grade must never let cuff zeros define the club.

A council, not a score

The working method is a judgment council. Each judge is allowed only one job. They are not averaged into mush until the end. Draft room

Construction and process. Did the seat take the format’s scarce piece? In one-QB, waiting on the passer can be correct even when the overall rank looks like a reach. In tight-end premium, a cheap tight end is a value pick and a format problem at the same time. Score both. 

Floor analyst

Only downside. Dead weeks, bye clusters, one-QB risk, committee backfields, availability flags on stars, correlated injuries (same offense, same bye, stacked skill players). Ends with a miss-playoffs script, not a vibe.

Ceiling analyst

Only the title path. Which three outcomes must hit. Which trade is allowed. Which player is not for sale. Ends with a championship script.

Consensus-rank judge

Pick versus expert board. Steal, fair, or reach. Separate the 1QB grade from the premium-tight-end grade. 

Market-price judge

Pick versus ADP band. Value is price paid minus market. Retail stars plus wholesale leftovers is a solvent draft. Two-round panic reaches are not.

Odds judge

Season-long props and futures only. Rushing yards and touchdowns, receiving lines, MVP and offensive-player boards. What the book will hang a poster on, and what it will not invent a number for. A workhorse rushing line is evidence. A committee “maybe” is not. 

Model judge

Internal projections and curve residuals only. The model is not the market and not the experts. It answers a different question: given where this player was priced, does the curve think the juice is above or below that price? Disagreement is the point. A player can be a market steal, an expert reach, a book fade, and a model leftover at the same time. That is information. A single letter grade erases it.

Obscure the index, keep the idea

Most private models bury the useful signal in branded columns. For public work, do not publish the codebook. Publish the jobs the indexes do. 

Any serious internal model is usually doing some version of these hidden jobs:

• Availability — will the player see the games you paid for?

• Competence/stability — is the weekly scoring distribution tight or wild?

• Explosion rate — how often does the player print a week that wins a matchup?

• Smash profile — is the usage shape (volume, red zone, receiving work) that of a starter, or of a dart?

• Award tail — is there a realistic path to a top-three finish at the position?

• High-end weeks versus floor weeks — the two tails, kept separate.

• Peak quality — best realistic season, not the mean.

• Curve residual — model juice minus what the ADP curve predicted for that price. Positive residual means the model likes the player more than the market. Negative residual means you paid a star sticker for replacement juice.

Name those jobs. Do not name your columns. A published article that dumps the proprietary header row is a leak. 

A published article that says “starter smash-rate versus bench explosion-rate” is a method. 

The operational rule is simple: grade the seven, not the seventeen. 

Compute the hidden indexes on the likely starting lineup and on the full skill roster, then compare both to the unweighted board of drafted skill players. 

If the full-roster mean looks poor and the starter mean looks rich, you drafted a core plus lottery tickets. 

That is a strategy.

 If both means are poor, you drafted hope.

Starters versus the board

A league baseline is not “average starter.” It is the unweighted mean of every skill player on the sheet. 

That baseline is ugly on purpose. 

It includes handcuffs and seventh wide receivers. What you want is the gap:

• full roster versus board

• starters versus board

A healthy title core is usually above the board on stability, smash-rate, floor weeks, and model juice. It can sit near the board on explosion if the explosion lives in one or two swing players rather than across the whole lineup. 

A common failure mode in these drafts was the opposite shape: gorgeous pillars and a running-back room that lived entirely in committee lines. 

Another failure mode was two locked backs and a premium-format tight end that the book priced like a 575-yard player. 

Both teams can have similar starter juice. They lose in different weeks.

What every lens should have to agree on

Before you write a composite grade, force a short agreement list. If the council cannot agree on three facts, you do not have a grade. You have a fight. 

Examples from this process:

• The first running back at the turn is either a workhorse or a role bet. Do not let ADP call a committee a star.

• In tight-end premium, taking the scarce tight end on time is not a reach. Taking a cheap tight end is a bargain and a leak.

• In one-QB, paying a third-round sticker for the best passer on the board is process. Waiting and still landing a dual-threat number is also process. Streaming after drafting one mid passer is hope.

• Bench leftovers with huge positive residuals are options, not starters.

• Don't trade the engine unless the return includes what you actually lack.

Then write the fight in a conflict matrix: 

One row per disputed player, one column per judge. 

1. Floor, 

2. ceiling, 

3. board, 

4. price, 

5. book, 

6. model. 

If all six columns say the same thing, the player is not interesting. If they split, that player is the season.

Interdisciplinary upgrades the public tools skip

Borrow from fields that already price uncertainty. From portfolio theory

Correlation is a feature and a bug. A quarterback and his running back spike together and die together. Stacking is a ceiling tool. It is not free insurance. 

From gambling theory

Closing-line value has a draft cousin. If you take a player at pick 39 and the market still treats him as a first-round talent at the position, you beat the closer. If you take a player at pick 34 and the curve residual is sharply negative, you paid tomorrow’s price for yesterday’s role.

From behavioral economics

Recency and herding show up as ADP runs. The council’s job is to notice when the table is drafting a narrative (the shiny rookie, the revenge wide receiver) and you are drafting a role. From operations research

Opportunity cost is the real VORP. The question is not “is this tight end good.” The question is “what starter did I not take, and can the rest of this roster survive that hole?”

From forecasting

Write the miss script and the title script in advance. A grade that cannot be wrong cannot be right.

The written record

The last product is not a tweet. It is a document you can paste into a file and reopen in January. 

It should freeze:

• team name, slot, format, date

• position counts and the expected seven

• every judge’s letter and one-line lens

• the agreement list

• floor script and ceiling script

• board ledger, price ledger, odds board

• model card with residuals described, indexes unnamed

• starter averages versus board

• a conflict matrix

• a to-do list with trade rules

• a blank postseason card: record, playoffs, title, hits, misses, would-draft-again

The postseason card is the science. If Cook’s workhorse line cashed and the cheap tight end still killed weeks, the TEP warning was the correct judge. If the model leftovers never hit, the residual board was a lottery ticket, not a strategy. Update the priors. 

Do not update the memory with a story.

A composite that stays honest

Letter grades are a courtesy. Use them, then pin them to a range.

• Floor — what happens if the committees stay committees and the leftover explosion rates stay at zero.

• Base — what happens if the pillars hold.

• Ceiling — what happens if one swing player hits and one hole is traded shut.

Across the Who Dis tickets graded in this process, the pattern was stable. Cores built on a workhorse, a true wide receiver one, and an elite passer sat in the same starter-juice band as cores built on a premium tight end plus that same passer. They were not the same team. One lost on empty running-back weeks. One lost on empty premium-tight-end weeks. One had to trade. One already had the scarce piece. That is the scientific claim, stated carefully: starter juice is comparable across good drafts; the failure mode is not. Grade the failure mode, or you are grading the logo.

What not to publish

Do not publish the header row of a private sheet.

Do not publish the fitted curve that turns ADP into expected juice.

Do not publish the exact residual formula.

Do not pretend Vegas lines are projections; they are prices.

Do not pretend expert rank is truth; it is a committee. Publish the jobs, the splits, the scripts, and the January scorecard. 

The science of fantasy football is not a better average. It is a roster treated as a book: several prices, two tails, one core, and a record you are willing to lose to.