How Roster Choices Affect the 20-0 Simulation

"Take the best player available" is useful advice until the best player available creates a fourth luxury and leaves a starting slot empty. In 20-0, the model makes that cost visible. Six offensive slots and six defensive slots feed two averages, which are combined equally to calculate each game's win probability.

Equal-looking sides receive no special bonus

The model does not reward matching offense and defense ratings. If offense is 94 and defense is 76, the combined rating is 85. A roster rated 85 on both sides also has a combined rating of 85 and the same win probability against every opponent. Adding three points to either side raises the combined rating by 1.5; equal-sized gains have equal effects wherever they land.

The useful comparison is not "which side is weaker?" but "which legal pick produces the larger gain over the slot it replaces?" A candidate's value depends on its rating, the open slot it can fill, and the future flexibility spent to place it.

Empty slots are expensive

An empty or skipped slot contributes 50. If five defenders average 88 and the sixth defensive slot is empty, that side falls to about 81.7. Filling the slot with an 80 raises the side to 86.7. That five-point side gain adds 2.5 points to the combined rating before the opponent is generated.

This explains why forcing another superstar into a crowded position can be worse than taking a lower-rated player who legally replaces a 50-point slot: compare the resulting contribution change, then divide by six for the side average and by two for the combined rating. The flex slots give you room, but they do not eliminate the twelve-slot constraint.

From one slot to a win chance

The roster model lets you trace that decision all the way to one matchup. Assume offense averages 90. Five defenders average 88 and the sixth slot is empty, so defense is (5 × 88 + 50) ÷ 6 = 81.67. The combined rating is (90 + 81.67) ÷ 2 = 85.83. Against the regular-season center opponent rated 82, the logistic formula gives about a 65.4% chance to win that game.

If an eligible 80-rated defender replaces the empty slot, defense becomes (5 × 88 + 80) ÷ 6 = 86.67 and the combined rating becomes 88.33. Against that same 82-rated opponent, the single-game estimate rises to about 74.2%. This isolates the value of one roster fit; the real model draws regular opponents from 74 to 90, so the example does not predict a full season.

The right decision changes through the draft

Early rounds favor portable value. A high-rated player at a scarce position or a flexible offensive/defensive option keeps future paths open. Late rounds favor need because there are fewer spins left to repair a hole.

A useful decision sequence is:

1. List the open slots the current player can legally fill.
2. Check whether taking that player would consume a flex slot needed by another position.
3. Compare the player's rating with the contribution of the slot it would replace, then account for the future value of any flex slot used.
4. Use a reroll only when the current team-era offers no acceptable path, not merely because the top name is unfamiliar.

This is more precise than "always draft defense" or "always take the highest number." The board state determines the answer.

Flex slots are option value

Two offensive flex slots and one defensive flex slot let you bank talent without committing to a single position. The mistake is spending them automatically.

If your first strong offensive spin is a wide receiver, placing that player in WR preserves both flex slots. If WR is already occupied and you use OFLEX1, the pick can still be correct, but you have spent future flexibility. The cost is invisible until a later spin offers a second elite running back and nowhere useful to place him.

Treat a flex slot like a reroll: valuable because it changes future choices. Use it when the player's advantage is worth that lost option.

Rerolls solve dead boards, not imperfect boards

You receive two rerolls. A weak-looking team-era can still contain a player who improves an open slot, while a famous team-era can be useless if all of its relevant positions are closed.

Before rerolling, ask one concrete question: "Can any eligible player improve an open slot over its 50-point replacement value?" If yes, the spin may be good enough. If no eligible player fits, a reroll gives you another chance to find a useful fit; it does not guarantee one or guarantee that every slot will be filled.

Saving both rerolls until the final two rounds is not automatically disciplined. If a mid-draft spin has no legal or credible fit and several slots remain, rerolling now protects more future paths. The principle is to spend rerolls on constraint failure, not disappointment.

The postseason raises the opponent rating for every roster

Regular-season opponents are centered at 82 and range from 74 to 90. The three modeled postseason games add 6, 12, and 18 points to opponent ratings. Week 18 combines the Wild Card and Divisional labels into one simulated game, followed by a Conference Championship and Super Bowl; this is a 20-game model, not a complete NFL bracket.

The matchup formula uses one combined rating: (offense + defense) / 2. At fixed center values, a combined rating of 85 against an 82-rated regular opponent gives 1 / (1 + exp(-(85 - 82) / 6)), or about a 62.2% chance to win. Against the center opponent plus the final +18 boost (rating 100), the same roster rating gives about 7.6%. These are single-game estimates; actual regular opponents vary, and each result also depends on the seeded random draw.

The model does not detect or target a roster's weak side during postseason games. A low offense or defense average matters because it pulls down the combined rating every week; the later opponent boosts raise the challenge for every roster. A low side rating is useful to inspect when it reflects missed slot value, but the formula does not add a separate penalty for imbalance.

The final record can differ from the projection for the same reason. The preview estimates wins from 200 deterministic scenario runs, then approximates perfect-season odds by raising the estimated average win probability to the twentieth power. At 17.2 expected wins, that approximation is (17.2 / 20)^20, about 4.9%; it is a planning heuristic, not a directly counted share of perfect runs. The final saved run uses one seeded path, so a preview is not a promise about the final record.

A practical late-round rule

With three or fewer slots remaining, stop comparing candidates only to the best player already on your roster. Compare each candidate to the contribution that slot will otherwise receive.

An 86 safety filling the last defensive slot replaces 50, a 36-point gain in that slot and a 3-point gain to the combined rating after both averages. A 94 receiver replacing an 89 in a flex slot gains five at that slot, or about 0.42 combined points. The scores do not make the safety inherently better; the comparison shows why an open replacement-level slot can be more valuable than a small upgrade elsewhere.

Compare the gains across all twelve slots

An upgrade to any one slot flows through the same arithmetic: its contribution change is divided by six for that side average, then by two for the combined rating. A 20-point improvement in an empty slot and a 20-point improvement in an occupied slot have the same model effect, if both placements are legal. Empty-slot repairs often look compelling because 50 is a low starting value, not because the simulation tracks position-specific vulnerabilities.

The preview's perfect-season figure is multiplicative: it raises estimated average win probability to the twentieth power. That is a rough planning shortcut; it does not separately calculate twenty different matchup probabilities or directly count perfect runs.

The shortest version is: compare each pick with the slot it replaces, preserve options when they have real future value, and read the projection as an estimate rather than a promise. Review the published scoring model, then play a draft and compare the preview with the final seeded season.