I Ran Five Mock Drafts and Found the Same Mistake

By the third mock I had discovered that my strategy was a strategy only when nobody else interfered #

I ran five mock drafts in three days, which felt productive in the way that opening five tabs to read the same article feels productive. The first mock went beautifully because everybody else drafted the way my script assumed they would. The second mock went well enough that I assumed the first was reproducible. The third mock went sideways when my target running back was taken two picks before my turn, and I spent the next thirty seconds discovering that my plan had no Plan B and the screen was politely waiting for me to do something humiliating. I drafted a wide receiver I had not been thinking about, then a tight end I did not want, and arrived at Round Six with a roster that was, by any honest assessment, slightly worse than if I had let the autodraft button handle it.

The fourth mock did the same thing in a different round. The fifth mock did the same thing again, in yet another round. By the time I sat down to review the five drafts together, the pattern was visible enough that I felt embarrassed not to have noticed it earlier. I was not practicing drafting. I was practicing a script. The script was, on average, a good script. The problem was that I had never written the branches, which meant that the moment the script broke, I was a worse drafter than someone who had not prepared at all.

The market mechanic of an ADP-driven draft #

A snake draft is a market with twelve participants, all of whom are working from approximately the same set of public ADP rankings. The plan that works in Round 1 is the plan that respects ADP; the plan that breaks in Round 4 is the plan that assumed two specific players would still be available at a pick number where the public market disagreed. The disagreement is not always rational, but it is reliable. The bettor who builds a plan with no branches is implicitly assuming the room will behave optimally, which the room does not.

The interesting feature of ADP is that it is most accurate at the top of the draft and degrades sharply between rounds four and eight, which is precisely the range where the pivot pressure is highest. The reason is structural. The top of the draft is dominated by consensus; the middle of the draft is where individual managers begin to express preferences; the late rounds are dominated by lottery tickets where the variance is high but the cost of being wrong is low. The middle rounds are the rounds the prepared drafter has to plan branches for, because they are the rounds where the script most reliably breaks.

What the mock log recorded that my memory would not have #

I built the log as a single sheet with one row per pick across the five mocks. The columns were the round, the player I had planned to take, the player I actually took, the position of the pick I had planned to take, the position I actually took, and a one-sentence note on why the deviation occurred. The first observation, after I had filled the sheet, was that the deviations were not random. They concentrated on the same positions — running back in Rounds 3 to 5, tight end in Round 3, wide receiver in Rounds 6 to 8 — and on the same kinds of misreads. The plan was assuming a tier of players would remain available at a round where the consensus had moved them up.

The second observation was about my response to the deviations. In about two-thirds of the cases I reached for a memory of the player I had wanted, rather than evaluating the available pool with the same criteria I had used to select the original target. The reach was almost always a downgrade. The exception was the small number of cases where I had pre-written an alternative — a tier-2 RB if the tier-1 RB was gone, a different TE if the elite TE was gone, a specific WR with a similar profile if my preferred WR was off the board. The pre-written alternatives, when I had them, produced picks that the log judged equivalent to or better than the original plan. The unwritten alternatives produced picks the log judged worse than the plan in roughly two cases out of three.

The branch-writing exercise that fixed the third mock #

After the third mock I sat down and tried to write the branches I had been missing. The format was simple: for each of the first six rounds, my Plan A pick, my Plan B if Plan A was taken, and my Plan C if both A and B were taken. The exercise took about forty minutes for the first three rounds and accelerated quickly thereafter, because most of the work was thinking, not writing. The hardest part was forcing myself to commit to a Plan B that I would not be allowed to reach past on draft day. A Plan B that exists in principle but gets ignored under pressure is not a plan; it is a regret with stationery.

The fourth and fifth mocks were the test runs for the branched plan. The fourth mock pivoted twice in the rounds the plan covered, and both pivots executed cleanly — the Plan B was waiting, I took it without spending twenty seconds explaining to myself why Plan A was no longer realistic, and the rest of the draft proceeded from there. The fifth mock pivoted three times, including once in a round I had not branched, and the unbranched pivot reproduced the original failure mode exactly. The lesson, which I would have happily not received but which the log made unavoidable, is that the rounds without branches are the rounds where the plan will break the most expensively.

Five mocks did not reveal the perfect draft. They revealed the spot where I needed a second plan. That is less glamorous and far more useful on the clock.

The protocol I will use on real draft day #

The protocol I have committed to for the real draft is the same protocol I used in the fourth and fifth mocks, with one addition. Each of the first eight rounds has a Plan A, Plan B, and Plan C with named players. Each pick I make is recorded in the log within thirty seconds of the click, along with whether it executed against Plan A, B, or C. After the draft I will compare the team I built to the team the plan would have built if every pick had been Plan A available, and the gap between the two will be the cost of the deviations. The addition is that I will commit to not amending the branches in the last hour before the draft, because the last-hour amendment is almost always the manager s anxiety speaking in the voice of preparation.

This is, like the rest of these experiments, a small discipline that does not photograph well. It does not produce a draft grade. It does not produce a screenshot of a perfect lineup. It produces a slightly better team in a slightly more defensible way, and it produces a log I can take into next season s planning instead of relying on memory of which mock did what. The bankroll equivalent is the same as the CLV diary equivalent, which is the same as the disagreement-log equivalent: the workflow is the product, the result is downstream of the workflow, and the workflow improves only as fast as the bettor is willing to record it.

The closing argument #

The five mocks did not produce the perfect draft. They produced something more useful: a log of the moments at which my preparation actually failed, with enough specificity to fix the failure before draft day. The pattern was so consistent that I am slightly embarrassed not to have noticed it sooner. I was prepared for the draft I had imagined. I was unprepared for the draft I actually got. The branches close the gap between the two, and the discipline of writing them in advance is what converts a preparation regimen from a script-rehearsal exercise into something that can survive the eleven other managers in the room.

I will run two more mocks before the real draft, with the branched plan in front of me, and I will log them with the same protocol. I expect the pivot count to be similar and the pivot cost to be lower. I do not expect the mocks to be more fun. They are not, in any direct sense, more fun. They are, in the direct sense that matters, more useful. The mock that produces a screenshot of a beautiful roster has taught me nothing. The mock that produces a log of two pivots executed cleanly and one pivot mishandled has taught me exactly what I need to know on the clock.