PUCKSON.NET

1THE HIGH-LEVEL APPROACH

THE QUESTION WE ARE TRYING TO ANSWER: WHICH ANALYSTS’ FINAL PUBLIC RANKINGS AND MOCK DRAFTS MOST CLOSELY ALIGNED WITH THE ACTUAL FIRST ROUND OF THE NHL DRAFT EACH YEAR?

WE SCORE TWO TYPES OF FINAL PUBLIC LISTS:

DRAFT RANKINGS

AN ANALYST’S EVALUATION OF THE PLAYERS IN THE CLASS BASED ON THEIR VIEWINGS, ANALYTICS, DEVELOPMENT CONTEXT, AND CONVERSATIONS WITH VARIOUS SOURCES.

MOCK DRAFTS

AN ANALYST’S PREDICTION OF WHEN PLAYERS WILL GET PICKED BASED ON THEIR EVALUATIONS, TEAM PREFERENCES & NEEDS, AND INFORMATION THEY’VE GATHERED.

WE SCORE THE BOARD, NOT THE LOGO. IF A PLAYER GOES 12TH, WE CARE WHETHER THE ANALYST HAD HIM AROUND 12 — NOT WHETHER HE GUESSED THE EXACT TEAM THAT ENDED UP MAKING THE PICK. PICK TRADES CAN SCRAMBLE THE TEAM SIDE OF A MOCK, SO THE CLEANEST APPLES-TO-APPLES TEST IS WHICH PLAYER CAME OFF THE BOARD AT EACH SLOT.

WHAT THIS ISN’T: A SCOUTING GRADE, OR A MEASURE OF LONG-TERM PLAYER QUALITY. THE DRAFT RESULTS ARE WHAT WE’RE COMPARING AGAINST — BUT THEY ARE NOT A GUARANTEE OF WHO WILL BECOME THE BEST NHL PLAYER. THIS MEASURES HOW CLOSELY A PUBLIC LIST MATCHED WHAT TEAMS ACTUALLY DID ON DRAFT DAY.

BUILT FROM 113 PUBLISHED LISTS — 67 PROSPECT RANKINGS AND 46 MOCK DRAFTS — ACROSS THE 2022–2026 DRAFT CYCLES. NOT EVERY ANALYST PUBLISHES EVERY YEAR OR IN BOTH FORMATS — ALL-TIME SCORES AVERAGE ONLY THE YEARS AN ANALYST WAS ACTUALLY SCORED.

2026 LISTS ARE TRACKED BUT AREN’T SCORED YET — SCORING WILL HAPPEN AFTER FINAL LISTS ARE PUBLISHED AND THE DRAFT IS COMPLETED.

2THE SCORING COMPONENTS

SLOT
ALIGNMENT + RELATIVE
ORDER + TIER
ALIGNMENT = DRAFT ALIGNMENT
SCORE × 1/3× 1/3× 1/3× 1/3

EACH COMPONENT EVALUATES SOMETHING DIFFERENT. SLOT ALIGNMENT CHECKS HOW CLOSE THE ANALYST WAS TO THE ACTUAL PICK. RELATIVE ORDER CHECKS WHETHER THEY HAD PLAYERS IN THE RIGHT ORDER. TIER ALIGNMENT CHECKS WHETHER THEY HAD THE RIGHT NAMES IN THE RIGHT PARTS OF THE FIRST ROUND. EACH ONE COUNTS FOR EXACTLY ONE-THIRD OF THE FINAL SCORE.

WHY THREE? AND WHY EQUAL WEIGHTS?

THE THREE SCORES USUALLY MOVE TOGETHER — AN ANALYST WHO DOES WELL ON ONE USUALLY DOES WELL ON ALL THREE. BUT THEY AREN’T REDUNDANT: SLOT AND TIER ALIGNMENT TRACK EACH OTHER CLOSELY, WHILE RELATIVE ORDER OVERLAPS THE LEAST — IT’S THE COMPONENT MOST LIKELY TO SEPARATE TWO ANALYSTS WHO LOOK SIMILAR ON THE OTHER TWO. EQUAL WEIGHTS ARE SIMPLE, TRANSPARENT, AND HARD TO OVERFIT.

1SLOT ALIGNMENT

SLOT ALIGNMENT ASKS: HOW FAR WAS EACH PLAYER FROM WHERE THE ANALYST HAD HIM? A SMALL MISS COSTS A LITTLE. A BIG MISS COSTS MORE. AND MISSES AT THE TOP OF THE DRAFT COUNT EXTRA, BECAUSE THE TOP PICKS CARRY THE MOST VALUE.

MISSING PLAYERS AND PLAYERS RANKED OUTSIDE THE FIRST ROUND ARE CAPPED AT SLOT 33. THAT KEEPS THE PENALTY BOUNDED — MISSING A FIRST-ROUNDER HURTS, BUT ANALYSTS WHO PUBLISH DEEPER LISTS ARE NOT PUNISHED MORE HARSHLY THAN ANALYSTS WHO STOP AT 32.

THE REVERSE COSTS TOO: PLAYERS AN ANALYST HAD IN THE FIRST ROUND WHO WEREN’T ACTUALLY PICKED THERE PUSH REAL FIRST-ROUNDERS OUT OF POSITION — SO THAT PENALTY SHOWS UP INDIRECTLY.

FOR THE MATH NERDS

FOR EVERY PLAYER SELECTED IN THE FIRST ROUND, WE COMPARE HIS ACTUAL DRAFT SLOT TO WHERE THE ANALYST RANKED OR MOCKED HIM. THE MISS IS THEN WEIGHTED BY SLOT IMPORTANCE AND NORMALIZED SO THAT HIGHER SCORES ARE BETTER (1.0 = PERFECT / 0.0 = WORST POSSIBLE).

THE SLOT WEIGHT w(i,y) IS AN 80/20 BLEND. 80% COMES FROM PICK VALUE, USING A DRAFT-VALUE CURVE ADAPTED FROM DOM LUSZCZYSZYN’S GSVA WORK — PICK 1 = 100, PICK 2 = 64.2, PICK 3 = 50.3, PICK 5 = 37.1. THE OTHER 20% IS A DRAFT-YEAR ADJUSTMENT BASED ON WHERE PUBLIC LISTS WERE MOST SPLIT. MOST OF THE MODEL FOLLOWS PICK VALUE: A MISS AT THE TOP OF THE DRAFT SHOULD HURT MORE THAN THE SAME MISS LATE IN THE ROUND. BUT EACH DRAFT CLASS IS DIFFERENT, SO THE MODEL ALSO GIVES A SMALL EXTRA WEIGHT TO RANGES WHERE ANALYSTS WERE MORE DIVIDED. THE SAME RECIPE IS USED FOR EVERY ANALYST EACH YEAR.

w(i,y) = 0.8 × PICK VALUE(i) + 0.2 × DRAFT-YEAR SPLIT(i,y)

THE RESULT: THE TOP OF THE DRAFT CARRIES MOST OF THE SIGNAL. THE TOP 5 PICKS ACCOUNT FOR ROUGHLY A THIRD OF THE WEIGHT, THE TOP 10 ABOUT HALF, AND THE TOP 16 ABOUT 70%.

rᵢ* = ANALYST RANK FOR PLAYER i, CAPPED AT 33
SlotAlignment(A,y) = 1 − [ Σ w(i,y) × | rᵢ* − i | ] / MaxMiss(y)

i = ACTUAL DRAFT SLOT. w(i,y) = THE 80/20 SLOT WEIGHT. MaxMiss(y) = THE YEAR’S WORST-POSSIBLE WEIGHTED MISS TOTAL. 1.0 = PERFECT, 0 = WORST.

2RELATIVE ORDER

RELATIVE ORDER ASKS: DID THE ANALYST HAVE PLAYERS IN THE SAME ORDER THE DRAFT DID? A FIRST ROUND HAS 496 HEAD-TO-HEAD PLAYER PAIRS. WE CHECK EVERY ONE. IF PLAYER A WENT BEFORE PLAYER B, THE ANALYST GETS CREDIT IF HE ALSO HAD PLAYER A AHEAD OF PLAYER B. THIS REWARDS ANALYSTS WHO UNDERSTOOD THE ORDER OF THE CLASS, EVEN IF THE EXACT PICK NUMBERS WERE OFF. IF ONE PLAYER IS LISTED AND THE OTHER IS NOT, THE LISTED PLAYER IS TREATED AS BEING AHEAD. IF BOTH PLAYERS ARE UNLISTED, THE ANALYST GETS NO CREDIT FOR THAT PAIR.

RELATIVE ORDER USES THE ANALYST’S ORDERING WHEREVER AVAILABLE, INCLUDING PLAYERS RANKED BEYOND THE FIRST ROUND — THIS IS WHERE DEEPER LISTS STILL ADD SIGNAL. A FIRST-ROUND PLAYER AT #40 CAN STILL WIN HEAD-TO-HEAD COMPARISONS AGAINST PLAYERS RANKED LOWER OR LEFT OFF ENTIRELY.

FOR THE MATH NERDS

EVERY PAIR OF FIRST-ROUND PICKS IS ONE YES/NO CALL: DID THE ANALYST ORDER THE TWO PLAYERS THE WAY THE DRAFT DID? THE SCORE IS SIMPLY THE SHARE OF CORRECT CALLS OUT OF ALL C(32,2) = 496 PAIRS.

RelativeOrder(A,y) = CORRECT HEAD-TO-HEAD CALLS / 496

CORRECT = THE EARLIER-DRAFTED PLAYER RANKED HIGHER BY THE ANALYST. LISTED BEATS UNLISTED. BOTH UNLISTED = NO CREDIT.

3TIER ALIGNMENT

TIER ALIGNMENT ASKS: DID THE ANALYST HAVE THE RIGHT NAMES IN THE RIGHT PART OF THE FIRST ROUND? WE CHECK EVERY CUTOFF FROM 1 TO 32. HOW MANY OF THE ACTUAL TOP 5 WERE IN THE ANALYST’S TOP 5? HOW MANY OF THE ACTUAL TOP 10 WERE IN THE ANALYST’S TOP 10? HOW MANY OF THE ACTUAL TOP 32 WERE IN THE ANALYST’S TOP 32? THEN WE AVERAGE ALL 32 CHECKS.

FOR THE MATH NERDS

THIS IS THE “SHAPE OF THE BOARD” TEST. FOR EVERY K FROM 1 TO 32, WE COMPARE THE ANALYST’S TOP K TO THE ACTUAL TOP K, THEN AVERAGE THOSE 32 RESULTS.

SO WE CHECK: HOW MANY OF THE REAL TOP 5 WERE IN THE ANALYST’S TOP 5? HOW MANY OF THE REAL TOP 10 WERE IN THE ANALYST’S TOP 10? HOW MANY OF THE REAL TOP 20 WERE IN THE ANALYST’S TOP 20? AND SO ON, ALL THE WAY THROUGH THE FIRST ROUND.

EVERY CUTOFF COUNTS ONCE — TOP 1, TOP 2, TOP 3, ALL THE WAY TO TOP 32. THE TOP OF THE BOARD STILL MATTERS MORE IN PRACTICE, BECAUSE A TOP-3 HIT HELPS AGAIN AND AGAIN ACROSS LATER CUTOFFS.

TierK(A,y) = | ANALYST TOP K ∩ ACTUAL TOP K | / K
TierAlignment(A,y) = AVERAGE OF TierK FOR K = 1 THROUGH 32

3EXAMPLE: SCORING AN ANALYST

A REAL EXAMPLE: COREY PRONMAN’S 2023 MOCK DRAFT — THE HIGHEST SINGLE-YEAR MOCK SCORE IN THE 2022–2025 SAMPLE.

SLOT ALIGNMENT

91.1%

+

RELATIVE ORDER

88.1%

+

TIER ALIGNMENT

90.7%

÷3 =

ALIGNMENT SCORE

90.0%

SEE EVERY PICK: THE FULL WALKTHROUGH

COREY PRONMAN 2023 FIRST ROUND MOCK DRAFT — PICK-BY-PICK SCORING

DRAFT PICK	SLOT ALIGNMENT: 91.1%	RELATIVE ORDER: 88.1%	TIER ALIGNMENT: 90.7%

4THE CONSENSUS APPROACH

THE CONSENSUS IS NOT AN ANALYST — IT’S THE PUBLIC-EXPERT BENCHMARK: THE STANDARD ANY INDIVIDUAL ANALYST HAS TO BEAT TO SHOW THEY ADD SIGNAL.

THE CONSENSUS USES A FIXED PANEL OF REPEAT, HIGH-SIGNAL SOURCES RATHER THAN EVERY LIST WE TRACK. THE GOAL IS TO CREATE A STABLE BENCHMARK THAT IS COMPARABLE ACROSS DRAFT YEARS, NOT A ONE-YEAR AVERAGE THAT CHANGES BASED ON WHO HAPPENED TO PUBLISH.

THE PANEL WAS CHOSEN ON THREE FACTORS:

1. REPEATABILITY

THE SOURCE HAS FINAL PUBLISHED LISTS ACROSS MULTIPLE DRAFT YEARS.

2. CREDIBILITY

THE SOURCE IS AN ESTABLISHED PUBLIC DRAFT ANALYST OR OUTLET WITH A REAL TRACK RECORD.

3. HISTORICAL SIGNAL

THE SOURCE SHOWED STRONG ALIGNMENT WITH ACTUAL RESULTS IN EARLIER DRAFTS, BEFORE THE YEARS BEING SCORED.

RANKINGS PANEL: COREY PRONMAN, SAM COSENTINO, CHRIS PETERS, JASON BUKALA, CAM ROBINSON
MOCKS PANEL: COREY PRONMAN, SCOTT WHEELER, CHRIS PETERS, SAM COSENTINO

HERE’S HOW IT’S BUILT:

1. CONVERT

EVERY RANK IS CONVERTED INTO POINTS USING A DRAFT-VALUE CURVE ADAPTED FROM DOM LUSZCZYSZYN’S GSVA WORK (PICK 1 = 100, PICK 2 = 64.2, PICK 3 = 50.3 ...). NHL DRAFT VALUE IS NOT FLAT — THE TOP PICKS CARRY MORE EXPECTED VALUE.

2. ADD IT UP

EACH PLAYER’S POINTS ARE AGGREGATED ACROSS THE PANEL. STRONG AGREEMENT AT THE TOP OF BOARDS COUNTS FOR MORE THAN SCATTERED MENTIONS LOWER DOWN.

3. SORT

PLAYERS ARE SORTED BY TOTAL POINTS TO CREATE THE CONSENSUS RANKING OR CONSENSUS MOCK — BUILT SEPARATELY FOR EACH, EVERY DRAFT YEAR.

WHY NOT JUST AVERAGE RANKS? A SIMPLE AVERAGE TREATS EVERY GAP THE SAME — THE GAP BETWEEN #1 AND #5 COUNTS THE SAME AS THE GAP BETWEEN #16 AND #20. THE VALUE-WEIGHTED APPROACH GIVES MORE WEIGHT TO HIGH-END CONSENSUS.

THE CONSENSUS IS THEN SCORED WITH THE SAME DRAFT ALIGNMENT SCORE AS EVERY ANALYST.

5THE RESULTS: WHAT THE DATA SHOWS

THE PUNCHLINE: THE CONSENSUS WINS THE MULTI-YEAR RANKINGS LEADERBOARD. IN MOCK DRAFTS, COREY PRONMAN NARROWLY EDGES THE CONSENSUS — BY LESS THAN ONE-TENTH OF A PERCENTAGE POINT.

FIELD AVERAGE BY YEAR

FIELD RANGE BY YEAR

PUBLIC DRAFT ALIGNMENT HAS IMPROVED OVERALL FROM 2022 TO 2025. RANKING SCORES ROSE EACH YEAR IN THE SCORED SAMPLE. MOCK DRAFT SCORES ALSO IMPROVED FROM 2022 TO 2025, DESPITE A SMALL DIP ALONG THE WAY.

THE FIELD IS ALSO GETTING MORE CLUSTERED. THE CLEAREST SIGNAL IS THE RISING FLOOR: THE LOWER-SCORING LISTS HAVE MOVED UP MATERIALLY SINCE 2022, ESPECIALLY IN RANKINGS. AFTER A WIDE-OPEN 2023 DRAFT, BOTH RANKINGS AND MOCKS TIGHTENED IN 2024 AND 2025. BY 2025, THE BEST-TO-WORST GAP WAS SMALLER THAN IT WAS IN 2022 FOR BOTH RANKINGS AND MOCKS.

MOCK DRAFTS CLUSTER MORE TIGHTLY THAN RANKINGS. THAT MAKES SENSE: FINAL MOCKS ARE TRYING TO PREDICT THE SAME DRAFT-ORDER MARKET, WHILE RANKINGS LEAVE MORE ROOM FOR ANALYST-SPECIFIC PLAYER EVALUATION.

SEE THE LEADERBOARD >

METHODOLOGY