Why Every Game Is a Coin Flip
NSBA plays like pure chance — and that is not proof the teams are equal. It's the fingerprint of a format engineered to hide real, unequal, persistent skill.
NSBA games are close — maddeningly, thrillingly close. A last-seed knocks off a top seed; a team is blown out one week and wins by two the next. The standings read like they were drawn from a hat. The obvious conclusion is that the draft made the teams equal. That conclusion is wrong, and this piece is the long walk through why.
The short version: NSBA's coin-flip look is real, but it is a measurement artifact stacked on a structural one. The skill is there — steep, persistent, and one level down from the standings. Four things conspire to bury it. We'll establish the pattern, rule out the easy explanation, then take apart the illusion piece by piece.
01The observation
how close is close, exactly
Because point scales differ across formats, the honest comparator is margin share — the winner's margin as a fraction of total points scored. A share of 0.20 means the winner edged the loser by 20% of the points; 0.50 is a rout where the winner doubles the loser. Ranked from tightest to most lopsided, the four datasets form a clean ladder:
The distribution tells the same story from both tails. Count the nailbiters (games decided within 15% margin share) and the blowouts (share ≥ 50%, i.e. the winner scored at least half the total):
| League | Nailbiters (<15%) | Blowouts (≥50%) |
|---|---|---|
| NSBA4 (draft fantasy) | 40.5% | 14.3% |
| SSB-2026 (open tournament) | 19.6% | 50.9% |
Two out of every five NSBA games come down to the wire; in open play barely one in five does, and instead half the field gets blown out. That is the pattern to explain. The tempting first move — "the draft equalized the teams" — is the one we are going to spend the rest of the piece dismantling.
02The mechanism
skill, luck, and how much of each the format allows
Model every game as outcome = skill_gap + luck. The skill gap is the true difference in team strength; luck is everything else — which questions come up, a couple of buzzer races, one mistimed neg. Luck is large, and crucially it is format-specific: a team's game-to-game margin swings by about ±38 points in NSBA and ±68 in open play. Wider luck alone would blur outcomes, but it is only half the mechanism. The other half is how often big gaps even occur. Bucket every gradeable game by the pre-game talent gap:
| Talent gap | Favorite wins |
|---|---|
| Small (evenly matched) | 56.3% |
| Medium | 66.3% |
| Large (mismatch) | 92.0% |
So the favorite-win rate is set by two things: how wide the luck band is, and how often the schedule serves up a genuine mismatch to overcome it. On both, the formats diverge. NSBA spends only ~7% of its games in the large-gap bucket; open play spends ~18% — more than double. Combine the tighter luck band with the scarcity of mismatches and you get the headline:
In open play the favorite wins roughly three in four (leave-one-out 76.2%); in NSBA barely more than half (55.6%). That much is settled and mechanical. The illusion begins at the next step — reading 55.6% as "the teams are equal." They are not.
03Is it just easier questions?
the difficulty objection, controlled twice
Before the interesting explanation, kill the boring one. Maybe NSBA is close simply because its questions are uniformly easy — everyone answers everything, scores compress, and margins shrink. The premise has some truth: open tournaments escalate difficulty as rounds progress, with the dead rate (tossups nobody answers) climbing from 11.7% early to 22.8% late, while NSBA stays essentially flat at 12.4% (and non-monotonic — it wobbles, it doesn't ramp). But the conclusion doesn't survive two controls.
- Control the format. Restrict the open tournaments to their homogeneous round-robin pool phase — the part most like NSBA — and they get more blowout-heavy, not less: SSB rises from 50.9% to 57.3% blowouts. Pool play pits top seeds against the weakest teams, so the bracket is actually the competitive part. Making open play structurally NSBA-like pushes it the wrong direction.
- Control the difficulty. Take SSB's single easiest round — the one whose dead rate matches NSBA's — so the questions are as gettable as NSBA's. SSB is still 49.6% blowouts versus NSBA's 14.3%. Difficulty and blowouts are decoupled: hand open-play teams NSBA-easy questions and half the games are still routs.
The objection runs backwards. Controlling for both format and difficulty leaves the gap intact — SSB stays roughly 3.5× more blowout-prone than NSBA at matched difficulty. Whatever makes NSBA a coin flip, it is not easy questions. Something about the teams is different.
04The parity mirage
why "the teams are equal" is the wrong conclusion
Measure balance the way sports economists do — the Noll-Scully ratio: the actual spread in team win% divided by the spread you'd expect if every team were identical and every game a coin flip. A value of 1.0 is the pure-luck floor; the more unequal the league, the higher it climbs. NSBA4 comes out at 1.02 — sitting right on the floor, below the standard values for every pro league:
| League | Noll-Scully | Reading |
|---|---|---|
| NSBA4 | 1.02 | on the pure-luck floor |
| NFL | ~1.5 | moderately unequal |
| MLB | ~1.8 | quite unequal |
| NBA | ~2.9 | steeply unequal |
It is tempting to crown NSBA the most balanced league on Earth. Don't — that comparison is a trap, and it is the door to the real story. Noll-Scully is not season-length invariant. Over few games the standings are mostly luck, so the ratio pins near 1.0 no matter how unequal the teams truly are. NSBA plays 6 games; those pro numbers come from 17-to-82-game seasons. You are comparing a snapshot to a time-lapse. And note: every one of those pro leagues drafts too, so drafting cannot by itself be why NSBA looks random. Three forces are — and only the third is really the draft.
A six-game season is a telescope pointed the wrong way
You cannot identify the better team in six games. To see how badly, simulate leagues with a known talent spread and play just six games each — then measure the Noll-Scully you'd observe:
| True talent spread | Observed Noll-Scully | Favorite wins |
|---|---|---|
| Truly equal teams | 1.00 | 49% |
| NFL-sized gaps | 1.15 | 59% |
| NBA-sized gaps | 1.21 | 62% |
| NSBA4 (observed) | 1.02 | 55.6% |
NSBA4's true talent spread is therefore statistically unidentified. An observed 1.02 is consistent with anything from perfectly equal up to roughly NFL-sized gaps (wider-than-NBA isn't strictly ruled out, but the point estimate argues against it). The reliability check agrees: a team's scoring in the first half of the season barely tracks its second half — split-half r ≈ 0.14, with a 95% interval spanning −0.4 to +0.6. That interval is uninformative, not a demonstration of zero: with 6 games we simply can't tell. Individual players, by contrast, each face ~150 tossups a season against a team's 6 games, and they replicate far more cleanly at r ≈ 0.5–0.8. With this much data we can already estimate the players; we cannot yet estimate the teams.
The core trap, stated plainly. A Noll-Scully of 1.02 is not evidence of equal teams. It is what any talent spread — equal, NFL, or NBA — produces when you only give it six games to show itself. "Equal" and "hidden but real" are not separable at this sample size, so the coin-flip look proves nothing about the teams.
05Two more reasons
availability, and the draft itself
The short season hides team talent. But two structural forces make NSBA look random on top of that — and neither has anything to do with the teams being equal.
You don't know which version of the team showed up
A pro roster is fixed game to game. NSBA re-rolls its effective strength weekly. Teams draft 6–7 players but only about 3.8 take the floor on a given night — roughly 40% of the roster sits. That turns the paper standings into a poor guide to what actually played:
When the best player is out, the team scores about 13 fewer points and wins roughly a third of the time versus over half when present — a large swing. In 14% of games the paper favorite actually fields the weaker lineup that night, so the "upset" was never an upset at all; the stronger roster simply didn't show. A real slice of "NSBA is random" isn't parity — it's not knowing which version of each team turned up. A fixed-roster pro league has none of this. (n ≈ 21 for the star-out split, and absence tangles with other weak-week factors, so treat the magnitude as directional.)
Stars exist; the draft just deals them evenly
Here the draft genuinely acts — and note what it does not do. Individual skill is steeply unequal and persistent: career win-shares have a Gini of 0.54 (the top 20 of 181 players hold 46% of all value), and the same players stay elite year to year (r = 0.49 season to season). The talent gaps between people are enormous and durable. But the snake draft scatters those stars so evenly that only 7.5% of that talent variance ends up between teams — the rest cancels out inside each balanced roster.
And this compression is the draft, not the short season. Undrafted Science Bowl over the same six games runs Noll-Scully 1.24–1.44 versus NSBA's 1.02 — visibly less balanced with identical season length. So the draft is real and it does compress talent hard. But it is not why the standings look random: even after the draft scatters the stars, undrafted play over six games would still show more spread than NSBA does. The extra flattening down to 1.02 is the short season and the shifting lineups finishing the job.
06Resolution
what the coin flip actually is
Put the three forces together. Steep, persistent individual skill exists — a Gini of 0.54, replicating at r = 0.49. The draft scatters it so evenly that only 7.5% lands between teams. A six-game season is too short to resolve even the talent that remains. And a different lineup shows up every week, with the best player missing a quarter of the time. Each force alone would blur the standings; stacked, they erase the signal almost perfectly — which is exactly what a 1.02 looks like.
NSBA's parity isn't proof the teams are equal. It's the fingerprint of a format built to hide skill that is real, steeply unequal, and persistent: the draft scatters the stars, a six-game season can't resolve what's left, and a different lineup takes the floor every week. The talent is all there — one level down, in the players.
So the right analogy isn't "a salary cap like the NFL," which merely narrows how much talent a team can hoard while keeping rosters fixed and seasons long. It's a fantasy re-draft league — every player redistributed every season by serpentine order, zero roster continuity — stacked on top of a season too short to measure teams and a lineup that changes weekly. The re-draft compresses talent harder than any cap could; the short season and the shifting lineups do the rest. The standings look random. The skill underneath them is anything but.
07Caveats
This is an exploratory read of one small league against three outside datasets, not a controlled study. Read the NSBA figures as directional. Specifically:
- NSBA4 is tiny — 14 teams, 6 games each — so every NSBA number here is directional, and its Noll-Scully (1.02) carries a wide interval. The whole point of the measurement section is that at this size "equal" and "hidden talent" are not separable.
- Pro-league Noll-Scully values are cited, not recomputed, and are not directly comparable to a 6-game season — which is exactly why the piece leads with the simulation rather than the raw comparison.
- Availability and star-out effects are observational, on small, confounded cells (n ≈ 21; a team missing its star is often just having a weak week for other reasons). Read them as associations and directional magnitudes, not clean causal swings.
- Correlation ≠ causation throughout; raw talent confounds nearly everything, and no multiplicity correction was applied across the dozens of comparisons.
NSBA Lab · built from scibowl.live tournament exports + the NSBA season database · reviewed against an adversarial statistician / quant panel.