Buzz Science — Timing, Stealing & Discipline

Deep-dive companion · Weeks 1–5 · 958 buzzes / 691 tossups (94 dead) / 30 games · built 2026-07-21

Read this before you quote a leader

Every buzz is stamped with a position in the question (buzz_pos: 0 = first word, 1 = final word). That timestamp is reliable in aggregate — it tells you how hard a question was — but it is noisy as a per-player skill grade: player-level correlation of buzz position with itself is only r ≈ .08. So the "Ice," "Itchy," and "Robbery" leaderboards below are descriptive color, not talent rankings. Sample sizes are printed on every row; say them out loud.

SOLID structural / safe to state as factCOLOR descriptive, buzz-position based — frame as funTINY n read the sample out loud

Rate leaderboards require ≥15 correct buzzes (mechanics) or ≥10 buzzes (discipline) unless a smaller n is printed. Contested-tossup / streak lines require ≥6–8 reps. One neg-cost model constant (11.18) is used throughout — its derivation is shown in the Neg Economy section.

THE HEADLINE · STRUCTURAL FACT

The Rebound Rule

On every tossup this season where both teams buzzed (a "contested" tossup), the person who buzzed first was wrong all 179 times. The game was won on the rebound: the second buzzer converts 76% of the time. SOLID

Why this isn't a fluke — it's the format. A second buzz can only exist because the first buzzer negged and opened the door. If the first buzzer had been right, the tossup is over and there is no "second." So "contested" is, by construction, a set of tossups where buzzer #1 already failed. That's why buzzer #1's conversion in this pool is 0 / 179 — it's a definition, not a talent gap. The live lesson: jumping first isn't a skill, it's a warning sign; the points are on the rebound.

Say "179 contested tossups this season. The person who got there first was wrong every single time — all 179. Speed to the buzzer isn't winning these games; cleaning up the other guy's neg is. The game is won on the rebound."

The Closers — who punishes the opponent's neg SOLID TINY n

Rebound conversion rate when the opponent negs first (min 6 contested reps). The counts are small — this is the reliable part (the 0/179 first-buzz fact), told player by player.

Player	Contested	Rebounds	Rebounds won	Closer rate
Edward C	12	7	7	100%
Sohil R	9	6	6	100%
Lucas W	8	6	6	100%
snapninja	7	5	5	100%
Kian D	11	9	8	89%
akhil	12	8	6	75%
Eric L	8	7	6	86%

Trigger-happy — most often the one who jumps (and negs) first TINY n

Share of a player's contested tossups on which they were the first buzz. In this format that first buzz is the neg that hands it over.

Player	Contested	Buzzed first	First-buzz rate
edwin h	9	8	89%
kensuke o	8	7	88%
Monish S	6	5	83%
pancreas uvula	15	12	80%
Rohan G	8	6	75%

Method: "contested" = both teams buzzed on a non-dead tossup; "first" = strictly lower buzz-word index (exact ties dropped). Min 6 contested reps to list. The 0/179 first-buzz record is the season aggregate across all listed and unlisted contested tossups.

TIMING · DESCRIPTIVE COLOR

Ice vs. Itchy

How buzz position works. Each buzz gets a fraction buzz_pos ∈ [0,1] for how far into the question it landed — 0.00 = buzzed on the first word, 1.00 = buzzed on the last word. "Ice" = high average position on correct buzzes (waits for near-certainty and still gets it). "Itchy" = low average position (early fingers). Caveat, prominent: as a per-player skill metric buzz position is unreliable — the player-level self-correlation is only r ≈ .08. It is trustworthy in aggregate (question difficulty), so treat these names as color, not a ranking of who is "clutch."

The Gem — Lucas W COLOR TINY n

Lucas W lets the question read to ~95% of the way through on his correct buzzes and still converts 76% (19/25). Late and lethal — pure knowledge, no panic. (25 buzzes; buzz position is descriptive.)

Say "Lucas W waits until basically the last word of the question and still beats you three times out of four. That's not speed — that's knowing it cold. Small sample though: 25 buzzes."

Full timing board (qualified buzzers, ≥15 buzzes)

Sorted by average position on correct buzzes — latest (iciest) at top, earliest (itchiest) at bottom.

Player	Buzzes	Correct	Conv.	Neg rate	Avg pos (correct)
Lucas W	25	19	76%	8%	0.95	Ice · Gem
Daniel Y	20	8	40%	30%	0.93	Ice
bluewater16	16	10	62%	6%	0.92	Ice
Rohan G	29	18	62%	21%	0.92	Ice
Harry G	25	17	68%	8%	0.92	Ice
vishnu m	44	25	57%	11%	0.89	Ice
Edward C	41	24	59%	15%	0.89
akhil	32	23	72%	16%	0.81
Kian D	30	25	83%	7%	0.80
suzuko	26	18	69%	19%	0.76	Itchy
ishaan k	23	9	39%	39%	0.76	Itchy
roshan	27	19	70%	19%	0.75	Itchy
varnite	22	12	55%	23%	0.73	Itchy
pancreas uvula	29	13	45%	41%	0.71	Itchy · Gunslinger
edwin h	24	12	50%	38%	0.70	Itchy · Gunslinger

The two ends tell the story: the itchy names (edwin h, pancreas uvula, ishaan k) carry neg rates of 38–41%; the iciest names mostly sit under 12%. But remember r ≈ .08 — a single player's position bounces around, so read the pattern, not the ordinal rank.

STEALING · MOSTLY SOLID

The Vulture — points stolen off other people's categories

Method. A player's "own subjects" = their top-2 subjects by combine IRT ability (theta). "Vulture %" = the share of their correct tossups that land outside those two subjects. Denominator is n_correct (not total buzzes), so the thin-tail rows are shaky — the safe ones are the high-volume leaders.

Leaders SOLID (top two) · TINY n (tails)

Player	Own subjects (top-2 θ)	Correct	Outside	Inside	Vulture %
roshan	Earth/Space + Physics	19	10	9	53%
Kian D	Bio + Chem	25	10	15	40%
Jianyu	Chem + Earth/Space	13	5	8	38%
Justin K	Chem + CS	8	3	5	38%
suzuko	Bio + Physics	18	6	12	33%
Edward C	Math + Physics	24	8	16	33%
vishnu m	Earth/Space + Math	25	8	17	32%
Sean	Bio + Earth/Space	16	5	11	31%
akhil	Chem + Earth/Space	23	6	17	26%
Harry G	CS + Physics	17	4	13	24%

Say "Forty percent of Kian D's points come on Bio and Chem — subjects that aren't even his best two. Roshan's over half. These guys aren't waiting for their category; they're robbing everyone else's." SOLID "on Roshan's 19 and Kian's 25 gets."

Only Roshan (10/19) and Kian D (10/25) clear a comfortable sample bar. Justin K's 38% is 3/8 and Jianyu's is 5/13 — real, but thin. The floor is instructive too: Lucas W (Chem+Physics) and Eric L (Bio+Chem) hoard their own subjects (11% / 8% outside) — specialists, not vultures.

STEALING · CRUDE PROXY · COLOR

Robbery Index — beating the field to the punch

Crude proxy — say so on air. Every question has a "hardness spine" = the average correct buzz position across the field (how late everyone else needed to hear it). A player's Robbery Index credits how much earlier than that field average they lock the same questions in, with extra credit on near-dead questions. A "heist" = converted ≥20 points-of-question earlier than the field. This leans entirely on buzz_pos, so the difficulty proxy is crude and descriptive — not a talent measure. Small heist counts (3–4). Treat as fun.

Player	Correct	Robbery idx	Earlier than field	Own avg pos	Heists
edwin h	12	0.161	+13 pts	0.70	3
boraginaceae	9	0.149	+14 pts	0.73	4
pancreas uvula	13	0.116	+11 pts	0.71	3
ishaan k	9	0.113	+10 pts	0.76	2
varnite	12	0.105	+7 pts	0.73	3
roshan	19	0.104	+7 pts	0.75	2

Notice the overlap with the "Itchy / Gunslinger" list — edwin h, pancreas uvula, ishaan k all buzz early. Robbery Index rewards the same early-trigger behavior the Neg Economy punishes; it's the upside of the risk, not a separate talent. Present as color only.

DISCIPLINE · SOLID (definitional)

The Neg Economy — what a neg really costs

The ~11.2-point neg. A neg looks like a −4. It isn't. When you neg, the tossup bounces to the opponent, who then gets a free look. Observed in this season's data: the opponent converts that free tossup 74.7% of the time (+4 each), and a converted tossup earns a bonus worth on average 56% × 10 = 5.6. So the expected swing is:
4 (your −4) + 0.747 × (4 + 5.6) = 4 + 7.17 ≈ 11.18 margin points.
That constant (11.18) scales every player's neg count below. It is a definitional model, not an independent measurement — the leaderboard is really "who negs most, cost-weighted."

Most costly neggers SOLID

Player	Buzzes	Negs	Cost/neg	Est. points bled
pancreas uvula	29	12	11.18	134.1
ishaan k	23	9	11.18	100.6
edwin h	24	9	11.18	100.6
kensuke o	14	7	11.18	78.3
Edward C	41	6	11.18	67.1
Rohan G	29	6	11.18	67.1
Daniel Y	20	6	11.18	67.1

Say "A neg isn't minus four — it's about minus eleven, because you hand the other team a free tossup they cash three times out of four, plus the bonus. pancreas uvula has bled an estimated 134 points to opponents on negs alone this season."

Most disciplined high-volume buzzers SOLID

Accuracy on the heaviest workloads — volume and a clean sheet.

Player	Buzzes	Correct	Negs	Accuracy	Neg share
Kian D	30	25	2	83%	7%
akhil	32	23	5	72%	16%
roshan	27	19	5	70%	19%
suzuko	26	18	5	69%	19%
Rohan G	29	18	6	62%	21%
Edward C	41	24	6	59%	15%
vishnu m	44	25	5	57%	11%

Say "Kian D is the model of discipline — 30 buzzes, 83% right, and only two negs all season. And Vishnu M went 30 straight buzzes without a single neg — the highest-volume clean streak in the league." SOLID

The gunslingers TINY n

Recklessness = a blend of high neg share, early buzzing, and early-neg rate (qualified: ≥10 buzzes).

Player	Buzzes	Negs	Neg share	Mean buzz pos	Recklessness
kensuke o	14	7	50%	0.70	0.149
pancreas uvula	29	12	41%	0.73	0.112
edwin h	24	9	38%	0.72	0.106
ishaan k	23	9	39%	0.75	0.097
varnite	22	5	23%	0.77	0.053

Say "Kensuke O is the biggest gunslinger in the league — he buzzes early and negs on half of everything he touches. Fun to watch, brutal on the scoreboard. Small sample, 14 buzzes." TINY n

MOMENTUM · MIXED

Hot hand & who's trending

Streaks SOLID

Longest run of consecutive correct buzzes, and longest run without a neg (qualified: ≥8 buzzes).

Player	Buzzes	Correct	Negs	Longest correct streak	Longest neg-free streak
Kian D	30	25	2	13	14
Eric L	14	12	1	11	12
akhil	32	23	5	10	10
roshan	27	19	5	9	9
Sean	19	16	2	9	9
vishnu m	44	25	5	5	30
Lucas W	25	19	2	5	23

Say "Kian D buzzed correctly 13 times in a row — the hot hand of the season on 30 buzzes. But the discipline record belongs to Vishnu M: 30 buzzes deep, not one neg." SOLID

Risers & fallers SOLID (top) · COLOR

Change in overall player rating from Week 1 to Week 5 (players with ≥10 buzzes; the ratings model, not raw buzz counts).

Player	Team	Games	Wk1 → Wk5	Momentum
Edward C	cryo	5	0.26 → 0.87	+0.60	Breakout of the season
Akhil B	James W	4	0.77 → 1.31	+0.54	Rising
Roshan A	mingle/Yunyi	3	0.13 → 0.66	+0.53	Rising
Lucas W	sumin	4	0.29 → 0.70	+0.41	Rising
Edwin H	jonathan	4	0.71 → 0.44	−0.27	Cooling
Ryan K	(Stitch)²	4	0.44 → −0.03	−0.47	Biggest faller

Say "Edward C is the breakout — biggest rating climb in the league, and he posted the single biggest one-week jump of the year, +0.39. On the other end, Ryan K is the coldest arrow on the board, down nearly half a point since Week 1." SOLID "Short season, so trends will move."

Caveats & method (honor these on air):
• buzz_pos is descriptive, not a skill grade. Player-level self-correlation is only r ≈ .08. The Ice/Itchy board and the Robbery Index are color, tagged COLOR; the Robbery difficulty proxy (field-average buzz position) is crude — do not call it talent.
• The Rebound Rule is the one structural fact here SOLID: first buzzer wrong 0/179 on contested tossups is true by construction (a 2nd buzz only exists because the 1st negged); rebound conversion 76%. The per-player Closer/Trigger tables are on small n (6–15 contested reps) — read the counts.
• Vulture uses top-2 combine-θ as "own subjects" and n_correct as denominator; only Roshan (19) and Kian D (25) clear a safe sample — tails are TINY n.
• Neg Economy is a definitional model: every neg is scored at a fixed 11.18-pt swing (derived from the observed 74.7% opponent free-tossup conversion + 56%×10 bonus). It's neg-count, cost-weighted — not an independent measurement.
• A few display names span two raw user-ids or map with low confidence (e.g. snapninja ↔ Rahib H, conf 0.50; Abhinav A split across two ids). 72 of 958 buzzes (7.5%) are unmapped and kept separate in these handle-grouped tables. Numbers are Weeks 1–5; a short season means ranks will move. Full rigor log: outputs/podcast/AUDIT.md.