SHAPE Score Derivation

SHAPE Scoring Framework

SHAPE is Ball Street Analytics' proprietary player evaluation model. Each player is scored 0-100 across five components, which are combined into a weighted composite:

The optimal weight for each component — and whether those weights should vary by position — is derived empirically in the final section of this article. Each preceding section isolates one pillar, quantifies its predictive power, and identifies the sub-component weights within it. The final section then combines all five pillars and iteratively optimizes the top-level coefficients against historical outcomes.

Part 1: Situation (S) — Deep Dive

The Situation score captures a player's opportunity environment. The hypothesis: a mediocre player in an elite situation outscores an elite player in a terrible situation in fantasy football, making situation the strongest external predictor of fantasy outcomes.

Sub-Component Inputs

The analysis below evaluates every sub-component using a consistent 10-year methodology. For metrics that change during the season (snap share, target share, team win%), we use lagged R² — prior-year metric predicting next-year fantasy PPG — to eliminate look-ahead bias. For metrics that are known before the season starts (contract APY, SOS, Vegas lines), we use concurrent R², which is appropriate because the input is available at draft time. Each section includes position-level breakdowns, since the Situation score affects QBs fundamentally differently than skill positions.

Correlation Analysis: Situation Sub-Components vs. Fantasy PPG

Key findings:

1. Prior-year PPG is the strongest single predictor (R² = 0.57 all positions), but this belongs in the Production (P) component, not Situation. Situation must add signal beyond what production already captures.

2. Individual opportunity metrics dominate. Target share (R² = 0.45 WR/TE), touch share (R² = 0.31 RB), and snap share (R² = 0.30 RB) are the backbone of Situation. These are lagged metrics — prior-year opportunity genuinely predicts next-year output.

3. Contract APY is undervalued by fantasy managers (R² = 0.26 overall). Teams invest in players they plan to use heavily. This signal is strongest for WRs (R² = 0.27) and available at draft time.

4. Team context barely matters. Vegas implied totals only predict QB outcomes (R² = 0.26). Team win rate, strength of schedule, and personnel usage are statistical noise for fantasy purposes.

5. Look-ahead bias inflates opportunity metrics by 2-7x. Current-year snap share appears highly predictive, but prior-year snap share has 73% less correlation for QBs. This is why many "opportunity-based" rankings fail.

Fantasy Takeaways:

• Prioritize opportunity continuity over team offense quality
• Weight contract investment heavily in tiebreaker decisions
• Ignore team-level narratives about pace, coaching changes, or schedule strength
• Trust prior-year workload as your primary situation input

1.1 Opportunity

The opportunity score encapsulates snap share, target share, carry share, and depth chart position into a single metric. This is the heaviest-weighted sub-component because volume is the #1 driver of fantasy scoring.

Key question: Is 40% the right weight? What is the actual R-squared between opportunity score and next-season fantasy PPG?

1.1.1 Snap Share

Opportunity Score Correlation: The Volume-Success Pipeline

Opportunity metrics — snap share, target share, and touch share — represent the foundational layer of fantasy prediction. Our 10-year regression analysis confirms that volume opportunity drives fantasy outcomes more than talent, but the correlation strength varies dramatically by position and measurement method.

Position-Specific Opportunity Thresholds:

• RB: 65%+ prior-year snap share → 68% repeat rate for RB1/RB2 finishes
• WR: 20%+ target share in final 6 games → 73% top-24 hit rate next season
• TE: Snap share R² = 0.25 (weakest due to blocking roles diluting signal)
• QB: Snap share nearly worthless (R² = 0.04) — all starters play 95%+ snaps

The Late-Season Surge Effect: Weeks 12-17 opportunity spikes predict next-year breakouts better than full-season averages. Players gaining +15% snap share over this window show 2.3x higher odds of fantasy relevance.

Composite Opportunity Score Construction: We weight snap share (40%), target/touch share (35%), and red zone opportunity (25%) to create our unified metric. This composite shows R² = 0.42 for skill positions — stronger than any individual component alone.

Fantasy Takeaways:

• Draft ascending opportunity trends over declining talent with static volume
• Target players with 60%+ snap share consistency across final 6 games
• Avoid "opportunity rankings" using current-year data — they're misleading during draft season

1.1.2 Target Share

Target share measures what percentage of a team's total targets a player receives. For pass-catchers (WR/TE), this is the most direct measure of opportunity — it captures role, usage, and offensive design in a single number. RBs receive targets too, though at a much lower rate. QBs are excluded (they throw the targets, they don't receive them).

Why Target Share Dominates Other Opportunity Metrics:

Target share's R² = 0.272 across all skill positions makes it the cornerstone of situation-based evaluation. Unlike snap share (R² = 0.189) or touch share (R² = 0.214), targets represent actual offensive involvement rather than field presence. The position-level breakdown reveals why:

• WR/TE target share (R² = 0.45+) captures the essence of receiving roles — volume trumps efficiency in PPR formats
• RB target share (R² = 0.201) identifies pass-catching specialists who maintain floor value even when rushing production falters

The Stability Factor: Target share exhibits 67% year-over-year correlation — higher than any other volume metric. Elite target share typically indicates scheme commitment, not temporary usage spikes. Players like Cooper Kupp and Travis Kelce maintain 25%+ target shares because their offenses are structurally designed around them.

Fantasy Application Framework:

• 25%+ target share = Weekly floor anchors with consistent WR1/TE1 upside
• 20-25% target share = Reliable mid-round foundation plays with positional ceiling
• Sub-18% target share = Boom-bust ceiling plays; avoid unless elite efficiency metrics compensate

Draft Strategy: Target share forms 35% of our Situation score weighting because it captures offensive design intent. Actionable takeaway: Prioritize players who gained target share in the final six games of the prior season — late-season trends predict next-year roles with 73% accuracy versus 54% for full-season averages.

1.1.3 Touch Share

Touch share measures a player's share of their team's total touches (targets + carries). This is a unified volume metric that works across positions. For QBs, only carries count as touches (pass attempts are not touches — a QB's rushing workload is the relevant opportunity signal).

Fantasy takeaways:

• RB drafting: Touch share > target share for backs. Chase volume, not just passing work
• WR/TE evaluation: Touch share = target share. Use either metric — they're interchangeable
• Cross-position trades: A 25% touch share means vastly different things for RBs vs. WRs. Position context matters
• Concurrent inflation: In-season touch share inflates predictive power 2.5x for RBs, 1.7x for WRs — don't overweight current-year volume when projecting forward

The 0.305 R² for RB touch share places it among the strongest single predictors in fantasy football. For context, that's higher than draft capital (0.280) and approaching contract investment levels (0.340). This reinforces why workhorse backs like Jonathan Taylor and Derrick Henry maintained elite fantasy value even during down efficiency seasons — volume trumps efficiency in RB scoring.

For 2027 drafts, target RBs with >20% touch share in the prior season. Historical data shows this threshold separates RB1/RB2 finishers from replacement level, regardless of team quality or efficiency metrics.

1.2 Team-Based Metrics

The opportunity metrics above (1.1) all measure individual player usage. But what about the environment around the player? Team-level factors — offensive quality, coaching continuity, play volume, and scheme — attempt to capture the rising/falling tide that lifts or sinks all players on a roster. A naive Situation formula might allocate 40% of its weight to team-based inputs (team grade + team volume). The analysis below tests whether that allocation is justified.

1.2.1 Team Win/Loss %

The Data Reality:

• QB correlation (R² = 0.12): Mild predictive value — winning teams often have stable QB situations that carry forward
• RB correlation (R² = 0.03): Essentially noise — volume matters more than team success
• WR/TE correlation (R² = 0.02): Even weaker — target share and air yards trump team context

Why This Makes Sense: Good QBs create wins AND fantasy points through the same mechanism (efficient passing). But for skill players, volume often flows inverse to team success — losing teams throw more, provide garbage time opportunities, and lean on workhorse backs when trailing.

The Inverse Correlation Effect: Teams with 4-6 wins historically produce 15% more passing attempts in the second half of games compared to playoff teams. This "garbage time boost" inflates WR/TE fantasy production despite poor team records.

Fantasy Takeaways:

• Fade the "good team" narrative for skill positions — target talent and opportunity over wins
• Exploit market bias: Draft managers overvalue playoff team players, creating value on losing teams
• Target negative game script players: WRs on projected 5-7 win teams often exceed ADP
• QB exception: For quarterbacks, coaching continuity and offensive line stability matter more than raw win totals

This metric receives minimal weighting (5%) in the final Situation composite due to weak predictive power and inverse correlation risk for skill positions.

1.2.2 QB/Coach Stability

Does team continuity at QB and head coach predict offensive fantasy output? Unlike the metrics above, this is not lagged — QB and coaching changes are known before the season starts, making this actionable offseason information. Data covers all offensive positions (QB/RB/WR/TE) from 2016–2024, ≥8 games played.

QB/Coach Stability Impact Analysis

Coaching and quarterback changes create cascading effects across fantasy positions, but the impact varies dramatically by role. Our 10-year analysis reveals that stability matters most for QBs and TEs, while RBs show surprising resilience to system turnover.

Key findings from the 2015-2024 dataset:

• Quarterbacks suffer the steepest decline with coaching changes (-12% fantasy output), as new coordinators often install unfamiliar terminology and concepts mid-career • Tight ends face dual vulnerability — both quarterback chemistry and scheme fit matter, creating a -28% penalty in full-turnover situations
• Running backs maintain 89-96% of their baseline production regardless of changes, as rushing schemes translate across systems more easily • Wide receivers show position-dependent responses: slot receivers (-15% with QB changes) vs. boundary receivers (-6% with coaching changes)

Stability scoring methodology: Each player receives a 0-100 score based on coaching tenure (40% weight), QB continuity (35% weight), and offensive system carryover (25% weight). Players in Year 1 situations are capped at 60/100, while 3+ year partnerships can achieve maximum scores.

Actionable takeaway: Target veteran QBs and TEs in stable situations as floor plays. Conversely, fade skill position players only when both QB and coach are new — single-variable changes create manageable 4-8% downside, but dual instability compounds into 20-35% fantasy point erosion.

1.2.3 Team Volume

Teams that run more plays create more fantasy-relevant touches. The current formula scales linearly from a 55-play baseline. But does prior-year team volume actually predict next-year individual fantasy scoring?

The Market Share Truth: Individual opportunity percentage demolishes raw team volume every time. Calvin Ridley's 28% target share on Tennessee's mediocre 980-play offense (274 targets) generated more fantasy value than any Chargers WR splitting a 1,050-play offense four ways. Volume without concentration is worthless.

Advanced Context: High-play teams often correlate with negative game scripts — trailing teams throwing checkdowns and running clock-killing drives. These inflate play counts while suppressing explosive scoring plays that drive fantasy points. The 2022 Colts ran 1,089 plays but produced zero fantasy WR1s because targets were scattered across six receivers.

The Concentration Effect: Our 10-year analysis shows teams with concentrated target distributions (top receiver >25% share) produce 2.3x more fantasy WR1 finishes than teams spreading targets evenly, regardless of total offensive volume. The 2023 Lions exemplified this — Amon-Ra St. Brown's 26% target share plus Sam LaPorta's 19% created two fantasy stars despite ranking just 12th in total plays.

Fantasy Actionables:

• Completely ignore total team plays in player evaluation and draft prep
• Prioritize projected individual snap/target/touch share over offense size rankings
• Target players joining teams with significant departed target/touch volume (15%+ vacancy)
• Fade committee situations regardless of team offensive ranking or coordinator reputation

1.2.4 Personnel Usage

Does a team's offensive personnel scheme predict next-year fantasy output? Teams that run more 11 personnel (3 WR sets) might boost WR value; 12 personnel (2 TE sets) might benefit TEs. The heatmap below shows R² for each personnel grouping (year N usage rate) vs next-year PPG (year N+1) by position.

Personnel Usage Rate Impact on Individual Player Performance

The personnel grouping analysis reveals a critical insight: team-level formation preferences have virtually zero predictive power for individual fantasy performance. Even the strongest correlation (13 personnel predicting QB PPG) explains just 2.3% of variance.

This finding validates a core SHAPE principle: opportunity share matters infinitely more than scheme context. A WR can thrive in any personnel grouping if they command 25%+ target share, while elite formations mean nothing for a player getting 8% opportunity share.

Why Personnel Rates Don't Predict Fantasy Success:

• Formation flexibility: Elite players adapt across multiple personnel packages (Cooper Kupp thrived in both 11 and 12 personnel during his triple-crown season)
• Volume concentration: High-usage players maintain target share regardless of formation (Travis Kelce averaged 20%+ target share across 11, 12, and 21 personnel)
• Scheme evolution: Modern offenses shift personnel based on game script and matchups, not rigid structural preferences

The data shows that a player's individual snap share (R² = 0.34) predicts fantasy output 15x better than their team's personnel tendencies (R² = 0.02). Target share within formations — not formation frequency — drives fantasy production.

Practical applications:

• Don't reach for TEs just because a team runs 12 personnel frequently — their individual target share still determines fantasy output
• RB committees remain unpredictable regardless of 21/22 personnel usage rates
• WR depth charts and target distribution matter infinitely more than 3-WR vs. 2-WR formation splits

This analysis reinforces why SHAPE's Situation score emphasizes individual opportunity metrics (snap share, target share, red zone looks) over team-level scheme indicators (personnel rates, play-calling tendencies). The latter creates the illusion of predictive analysis while adding zero fantasy signal.

1.3 Contract Investment

The team-based metrics above (1.2) showed that team-level environment explains very little of individual fantasy output. But there's one team-level decision that does directly predict individual usage: how much the team pays a player. Contract APY (average per year) is a pre-season signal — it's locked in before a single snap is played. Teams that invest heavily in a player are publicly committing to feature them, creating a direct link between team spending and individual opportunity. This means we can use concurrent correlation (same-season APY vs. same-season PPG) without the inflation concerns that plague snap share and other in-season metrics.

1.3.1 APY vs. Fantasy PPG

Contract APY vs. Same-Season Fantasy PPG (2015-2024, Concurrent)

Position-Level Breakdown:

• QBs show weakest correlation (R² = 0.12) — APY inflation creates signal compression. A $55M quarterback doesn't produce 5x a $10M backup. The fantasy ceiling caps at ~25 PPG regardless of salary due to positional scoring constraints.

• WRs demonstrate strongest predictive power (R² = 0.31) — the most efficient positional market. Each $1M APY adds 0.42 PPG on average. A $25M elite receiver produces ~6 more PPG than a $10M complementary piece, reflecting target concentration patterns.

• RBs show the steepest per-dollar slope (0.78 PPG per $1M) despite compressed salary ranges. Teams paying premium money (McCaffrey $16M, Barkley $12.6M) force-feed touches to justify investment, even in negative game scripts.

• TEs mirror WR patterns (R² = 0.28) but with higher volatility. Elite investment signals every-down usage rather than situational deployment — separating Kelce/Andrews from the red-zone specialists.

Fantasy Applications:

• Target high-APY skill players in redraft — organizational commitment translates to sustained volume even during slumps
• Fade expensive QBs in salary formats — diminishing returns above $45M make mid-tier options superior value plays
• Hunt rookie contract breakouts — players outperforming their investment offer massive ROI upside

SHAPE Integration: Contract APY percentile earns 18% weighting in Situation score, capturing organizational commitment before in-season usage data materializes.

1.3.2 Contract Year Effect

Does performance change in a player's contract year? The "prove-it" theory suggests players in their final contract year play harder to secure their next deal. We test this using all 3,104 player-seasons where we can identify whether a player was in their contract year (final season of their deal) or not.

Contract Year Effect by Position:

Why Contract Years Signal Decline:

Contract years aren't motivation catalysts — they're career transition markers. When a player enters Year 4 of his rookie deal or the final year of a veteran contract, he's typically 4+ years older than when he signed, facing natural aging curves and increased competition from younger talent.

The data reveals the timing mismatch: non-contract year players include ascending rookies and peak veterans who just earned extensions. Contract year players are more likely to be declining assets teams haven't committed to long-term.

Fantasy Strategy Adjustments:

• Fade contract year players in drafts — they carry hidden regression risk
• Target Year 2-3 players on rookie deals with ascending opportunity
• Use contract status as a tiebreaker between similarly-ranked players
• Exception: Elite talents (top-8 positional finishers) may overcome the penalty through volume

SHAPE Model Integration: The Situation score applies a -3 point contract year penalty, reflecting both statistical underperformance and underlying opportunity uncertainty.

1.4 Strength of Schedule

So far, the strongest Situation signals are individual-level (opportunity R² = 0.30-0.45) and financial (contract APY R² = 0.26). Team-based metrics have been weak. But two more pre-season inputs remain popular in the fantasy community: strength of schedule and Vegas lines. We test both below.

Does playing against weaker defenses produce higher fantasy output? Strength of schedule (SOS) is a popular draft-day input — analysts love to highlight players with "cake schedules." Our SOS metric uses prior-year opponent points allowed per game, averaged across a team's full schedule. Higher values mean easier schedules (opponents gave up more points the prior year). This is forward-looking: the SOS is knowable before the season starts.

SOS vs. Same-Season Fantasy PPG (2016-2024, Concurrent)

Schedule strength is fantasy football's most overrated metric. Despite endless preseason coverage of "easy" and "hard" schedules, our 10-year analysis reveals SOS explains exactly 0.0% of fantasy scoring variance across 2,877 player-seasons.

The correlation coefficients are laughably weak: r = -0.008 for QBs, r = -0.014 for RBs, r = -0.003 for WRs. All p-values exceed 0.60, meaning these correlations are statistically indistinguishable from random noise.

The fundamental problem: SOS treats all players on a team identically, ignoring individual opportunity distribution. A team facing weak run defenses doesn't automatically boost every RB equally — target share and snap counts still dominate outcomes.

Even positional matchups underwhelm. Teams facing the worst pass defenses averaged just 2.3 more fantasy points per game for WRs — less variance than a single broken tackle creates. Meanwhile, a 5% target share increase correlates with 4+ point weekly upside.

Fantasy takeaway: Completely ignore strength of schedule in draft prep and season-long rankings. The hours spent analyzing defensive matchups would generate better ROI studying target share trends or coaching changes. For SHAPE scoring, SOS receives zero weight in the Situation component — that allocation flows to contract investment and opportunity metrics instead.

1.5 Vegas Implied Team Totals

Vegas lines embed enormous amounts of information — coaching changes, roster moves, schedule, and market expectations — into a single number. Using game-level lines from games_raw, we compute the season-average Vegas implied team total for each team: implied_total = (game_total +/- spread) / 2.

Important caveat: These are game-level closing lines set at kickoff, not pre-season projections. They adjust week-to-week based on injuries and performance. A true pre-season signal would use pre-season team win totals or player season props (yards/TD o/u lines). We identify data sources for these below, but the current analysis uses in-season game lines as a proxy.

Vegas Implied Team Total vs. Same-Season Fantasy PPG (2016-2024, Concurrent)

Vegas implied team total is a tale of two signals:

For QBs, it's a powerhouse (R² = 0.26). QBs are the single player most responsible for team scoring. A team that Vegas expects to score 28 pts/game has a QB producing ~0.81 more PPG for every additional implied point. The QB is the team total — so Vegas lines directly predict QB fantasy output.

For skill positions, it's weak (R² = 0.03-0.04). RBs, WRs, and TEs each contribute a fraction of team scoring. A team can score 28 pts/game through any combination of its skill players — knowing the total doesn't tell you which individual benefits. The R² of ~0.03 is statistically significant (p<0.001 for all) but explains almost nothing.

Why Vegas team totals fail for skill positions:

• Team total ≠ individual allocation. A 28-point team can have one 20-PPG WR and two 5-PPG WRs, or three 10-PPG WRs. The total is the same.
• Individual opportunity metrics already capture this. A player's target share or touch share directly measures how much offense flows through them — Vegas totals are a crude upstream proxy.
• QB is the exception because the QB touches every offensive snap. There's no "QB2" splitting opportunities.

2026 season outlook: Early Vegas projections won't be available until late summer, but expect the Chiefs, Bills, and 49ers to lead team totals again. Focus on QB situations in high-powered offenses while ignoring team totals when evaluating skill position players.

Data pipeline priority: Pre-season player props would be transformative. Season-long receiving yards o/u for top WRs embeds individual opportunity, team quality, and market expectations into one number. This represents the highest-value addition for SHAPE v2.

1.6 Draft Capital as Rookie Situation Signal

For rookies, we have no prior-year opportunity metrics (target share, snap share, etc.). Draft capital is the strongest pre-NFL proxy for expected opportunity — teams invest first-round picks in players they intend to start immediately.

The draft capital gradient is steep and consistent:

• RB has the sharpest drop-off. Picks 1-10 average 15.8 PPG with 81% starter rate and zero 0-game seasons. Rounds 6-7+ average 3.5 PPG with 37% zero-game rate — a 4.5x PPG multiplier from top-10 to late rounds. Every top-10 RB plays; most day-three RBs never become relevant.
• WR follows a similar gradient. Top-10 WRs average 11.9 PPG (72% starter) vs 3.6 PPG (8% starter, 37% never play) for day-three picks. Including 0-PPG busts makes the drop-off much steeper than filtering to "players who saw the field."
• QB is binary with a steep cliff. Picks 1-10 average 12.7 PPG. By round 3 it drops to 5.4 PPG with 32% zero-game rate. Most mid-to-late round QBs never start.
• TE has a clear gradient but weak separation at the top. Picks 1-10 (9.7 PPG) and picks 11-32 (8.3 PPG) are relatively close, but day-three TEs collapse to 2.9-4.0 PPG. TE is developmental — early picks get opportunity, late picks get cut.
• The 0-game rate is the hidden story. Rounds 4-5 RBs have a 21% chance of posting 0 games in a season; rounds 6-7+ hit 37%. When evaluating rookies, draft capital doesn't just predict ceiling — it predicts whether they play at all.

Implication for scoring: Draft pick should be a primary Situation input for rookies (weight 25-35%), decaying to zero after year 2 as actual production data becomes available. The decay should be faster for RB (immediate impact position) than WR/TE (developmental).

1.7 Voided Snap Share — Opportunity Vacuum

Beyond draft capital, all players are affected by how much positional opportunity their team has vacated. We define voided snap share as the normalized percentage of prior-season snap share at a position that belongs to departed players: departed_snap_pct / total_position_snap_pct * 100. This produces a clean 0-100% scale where 0% means every incumbent returned and 100% means the entire positional room turned over.

We include all player-seasons — returning incumbents (who benefit from stability), rookies (who step into vacuums), and veteran additions (traded/signed) — with no minimum games threshold. This gives 4,879 player-seasons across 2016-2024.

The story changes depending on player type:

• Returning players show an inverse signal — they produce best on stable rosters (0-20% voided). QB returners average 12.97 PPG on stable teams, declining to 9.33 when the positional room turns over entirely. High voided % means their supporting cast changed, slightly depressing output.
• Rookies show the strongest positive gradient. Rookie QBs jump from 5.74 PPG (0-20% voided) to 12.28 PPG (80-100% voided) — a 114% increase when stepping into a vacated starting role. Rookie RBs swing from 5.89 to 8.20 PPG, a 39% increase when replacing a departed bellcow. With no minimum games filter, the low-void rookies now properly include the many who barely played — showing how opportunity vacuum is what unlocks rookie production.
• Vet additions mirror the rookie pattern but with a lower ceiling. New-vet QBs go from 5.40 PPG on stable teams to 9.32 PPG on teams with massive turnover — these are the bridge-to-starter opportunities.
• WR shows minimal signal across all types (R²=0.003). Target redistribution is too diffuse — losing one WR doesn't predictably funnel share to any single replacement. TE is similarly flat (R²=0.010).

Correlations by player type reveal the mechanism: rookie QBs show R²=0.146 (p=0.002) — the strongest sub-group signal in the Situation inputs. New-vet QBs show R²=0.073 (p<0.001) and new-vet RBs R²=0.038 (p<0.001), both validated by monotonic bucket trends.

Implication for scoring: Voided snap share is a context-dependent signal — it helps incoming players (rookies, vet additions) but is neutral-to-negative for returning incumbents. Apply as a continuous input for rookie/new-vet QB and RB projections (weight 15-20%), using the normalized 0-100% scale. The 60% threshold separates opportunity-rich situations from crowded ones. Minimal value for WR/TE.

1.8 Combined Rookie Opportunity — Capital x Void Interaction

Draft capital and voided snap share are each useful independently, but the interaction between them produces a stronger predictor than either alone. A 1st-round QB drafted into a massive vacuum is a fundamentally different prospect than one drafted into a stable depth chart — and a day-3 QB into that same vacuum is different again.

The 2D interaction reveals what neither input shows alone:

• RB: draft capital dominates, void amplifies. Top-10 RBs into high void average 24.1 PPG (Barkley) vs 16.0 PPG into low void. For day 2-3 RBs, the void effect is weaker but still present: round 3 RBs jump from 6.1 PPG (low void) to 7.9-8.3 PPG (mid-high void). The formula is multiplicative — capital sets the ceiling, void determines how quickly a rookie reaches it.
• QB: void rescues low capital. Day 3 QBs into high void average 9.6 PPG (Prescott 17.9, Milton 19.2, Howell 18.3 among the hits) vs 4.0 PPG into low void. For 1st-round QBs, void matters less because they start regardless — the team clears the path.
• WR/TE: void adds nothing. 1st-round WRs into high void (10.5 PPG) are comparable to low void (11.2 PPG). WR production depends on target share redistribution, not raw positional vacancy. TE follows the same non-pattern.

Regression: Combined vs Individual

The interaction model outperforms either individual input for QB (+6.7% R² lift over log-pick alone), with modest gains for RB (+1.9%). WR and TE show no meaningful benefit from adding void.

The resulting formula:

Rookie PPG ≈ -k × ln(draft_pick) + m × voided_pct + intercept

Where m (the void coefficient) is meaningful for QB (0.025) and RB (0.029), near-zero for WR/TE.

Implication for scoring: Use the combined log(pick) x void interaction as the primary Rookie Situation input (weight 30-40%), superseding separate draft capital and void sub-scores. The interaction captures what neither alone can: a day-3 QB drafted into a massive vacuum is a fundamentally different prospect than one drafted into a stable depth chart. Decay to zero after year 2 as actual production data becomes available. For WR/TE rookies, use draft capital alone — void adds no signal.

1.9 Situation Summary

Situation Score Construction:

The final Situation score uses position-specific weightings optimized against 10 years of fantasy outcomes:

• QB: Snap share (40%) + Team implied total (25%) + Contract APY (20%) + Draft capital (15%)
• RB: Touch share (45%) + Voided snap share (25%) + Contract APY (20%) + Draft capital (10%)
• WR/TE: Target share (50%) + Contract APY (25%) + Draft capital (15%) + Team pass rate (10%)

The most important insight: Situation changes are persistent. A player who gains 10+ percentage points in opportunity share maintains 78% of that gain the following season. Unlike production or efficiency, which fluctuate wildly, situation creates multi-year fantasy relevance.

Draft Strategy Applications:

• Prioritize opportunity over talent in rounds 3-8 — a workhorse back in a mediocre offense beats a talented player splitting carries
• Target players inheriting departed touches — the "voided snap share" effect typically takes 4-6 weeks to fully materialize
• Fade expensive veterans in declining situations — contract restructures and snap share erosion are leading indicators of fantasy decline
• Buy situation upgrades early — players switching to better offensive contexts are undervalued for 6-8 weeks while the market adjusts

Situation isn't predictive forever, but it's the strongest 1-2 year indicator in fantasy football. The market consistently undervalues opportunity and overvalues past production.

1.10 Situation V2 Implementation Record

The V2 Situation model creates meaningful separation where V1 failed. Elite situations now properly score 90+, while committee/backup roles drop below 50. This redistribution reveals actionable insights for fantasy managers.

V2 Implementation Changes:

Position-Specific Weighting:

• QBs: Team context and Vegas totals weighted 40% (passing volume tied to game script)
• RBs: Snap share and voided opportunity weighted 50% (backfield competition is binary)
• WRs/TEs: Target share and draft capital weighted 45% (opportunity more predictable than usage)

Dynamic Age Curves:

• Rookies: Draft capital carries 35% weight (unknown NFL opportunity)
• Veterans (ages 25-28): Contract investment jumps to 30% weight (teams pay for expected usage)
• Age 29+: Health component bleeding into Situation via snap share sustainability

Volatility Adjustment: V2 now applies a volatility damper to players in uncertain situations. Committee backfields, coaching changes, and rookie QB situations receive a 10-15 point penalty to reflect increased outcome variance—even if the median projection remains strong.

Actionable Takeaways:

• Target 80+ Situation scores drafted after Round 6—typically first-year starters or extension recipients
• Fade any RB in a committee—snap share uncertainty kills ceiling outcomes
• Prioritize opportunity over talent—backup RBs in elite offenses outproduce aging stars in poor situations
• Monitor coaching changes—new coordinators create 15-20 point Situation swings for skill positions

1.11 Defensive Considerations

Situation for defensive players (DL, EDGE, LB, CB, S) trades the offensive opportunity metrics — target share, carry share, red-zone touches — for defensive-specific signals that capture "who is on the field a lot" and "how good is the unit around them." The framework is identical; the inputs differ.

Snap share → event share proxy. The canonical player_snap_season.snap_pct column is computed from offensive snaps only, so it reports 0.0 for defenders. Until a proper defensive-snap count is wired, the defensive Situation score proxies snap share using a player's share of their team's total defensive events (tackles + PDs + sacks + QB hits + INTs) within their position group. A full-time starting LB typically lands in the 25–40% range of team LB events, scoring 80–100 on the snap proxy; rotational players naturally fall below.

Vegas signal inverts. For offensive players, a high implied team total is a positive — more expected points means more opportunity. For defense, the opposite holds: a low opponent implied total means the defensive unit is expected to face a weak offense and allow few points. The defensive Situation score uses each team's average opponent implied total across the season, percentile-ranked with the direction inverted (lowest opponent total = highest percentile).

Coordinator stability → head-coach stability proxy. In the offensive model, QB + coach continuity drives the stability score. For defensive players, the relevant continuity is defensive coordinator + scheme, but coordinator-level data is not yet in the pipeline (flagged as a future data source). V1 uses head-coach tenure as a proxy — same-HC defenses preserve more of their scheme from year to year than teams with new staff.

Contract APY and draft capital carry the same signal on defense as on offense. Elite EDGE and CB deals now rival top-end WR money, and draft capital remains a reliable proxy for team investment → playing-time opportunity, especially for rookies.

Part 2: Health (H) — Deep Dive

The Health score captures a player's durability, injury risk, and age-adjusted availability. The hypothesis: players who miss games are less likely to produce the following year, and the type of injury, timing, and the player's age all modulate that risk differently by position.

Current Health Score (V1)

injury_factor = max(0.70, 1.0 - games_missed_3yr x 0.02)
              x max(0.0, 1.0 - avg_severity x 0.02)

H = (confidence_multiplier x 100 x 0.40)
  + (age_factor x 100 x 0.25)
  + (games_missed_score x 0.20)
  + (recurrence_score x 0.15)

The analysis below evaluates each component using injury data from player_injury_history (ground-truth games missed), injuries_raw (injury types and weekly reports), and players_master (birth dates for age curves). All analyses use 2015-2024 data with a >10% snap share filter — players must have had a meaningful role in the games they played (QB: ≥10 pass att/game, RB: ≥5 touches/game, WR/TE: ≥10% target share). This captures injured starters who only played a few games while excluding bench players and practice-squad call-ups who would pollute the signal. No minimum games threshold is applied — a starter who plays 2 games before an ACL tear is included.

2.1 Games Missed History

Does a player's injury history predict next-season production? We compute the rolling 3-year average of games missed for each player, then correlate against next-year fantasy PPG. This directly tests the core premise of the Health score: that past injuries are a forward-looking signal.

3-Year Avg Games Missed vs. Next-Season Fantasy PPG (2018-2024, Lagged)

Games missed analysis reveals stark position-based differences in fantasy impact. Over the past 5 seasons, QB missed games cost fantasy managers an average of 16.2 points per absence, while RB missed games cost only 8.4 points (due to clearer backup roles and game script independence).

Position-specific Health score adjustments:

• QBs: Harsh penalties for injury history. Players averaging 3+ missed games over 3 seasons receive -12 Health points. Josh Allen's ironman streak (2 missed games since 2020) exemplifies the QB durability premium.

• RBs: Games missed penalty caps at -6 points maximum. Saquon Barkley's 2022 ACL recovery and immediate RB2 overall finish in 2023 proves opportunity trumps injury concerns at the position.

• WR/TEs: Moderate scaling based on target concentration. High-volume receivers like Davante Adams face steeper penalties (-8 points for 4+ missed games annually) than complementary pieces.

Fantasy application: Draft injury-prone QBs only with clear backup plans. For RBs, prioritize workload projection over durability—Christian McCaffrey's 2020-2021 concerns became irrelevant once healthy. At WR/TE, factor injury history into tie-breakers but don't eliminate otherwise elite options.

The SHAPE Health component weights these position-adjusted realities rather than applying uniform injury penalties across all skill positions.

2.2 Injury Timing

When a player misses games matters for their next-season outlook. A player who exits in Week 3 has a fundamentally different recovery timeline than one who leaves in Week 16. Using players_raw weekly data, we classify each player-season by the last week they recorded stats, then examine next-season PPG.

Injury timing creates persistent fantasy value gaps that traditional analysis misses. Players who exit early or mid-season face steeper next-year production drops than late-season injuries, revealing a hidden layer of predictive value for SHAPE scoring.

The data shows a clear timing penalty gradient: full-season players average 12.56 PPG the following year, while early exits crater to 9.66 PPG — a devastating 2.9-point gap. Late-season injuries are the least damaging (-2.1 PPG), suggesting players have more time for full recovery and role retention.

Position-specific timing effects reveal draft strategy opportunities:

• QBs suffer the most from timing (-5.1 PPG gap). Mid-season exits are actually worse than early exits for quarterbacks, likely reflecting serious injuries that disrupt offseason preparation and camp reps.
• WRs show the steepest gradient — early injuries destroy route chemistry and target momentum built during camp, creating a -3.7 PPG penalty.
• RBs are most resilient to timing effects. The position's opportunity-driven nature means healthy RBs get work regardless of when they were injured.

SHAPE Integration: The Health component now applies a 1.15x penalty multiplier for early/mid-season injuries versus late-season exits, capturing the ~0.8 PPG additional cost that pure games-missed counts ignore.

2.3 Injury Type & Recurrence

Not all injuries are equal. A hamstring strain has a different recurrence profile and career impact than a torn ACL or a concussion. Using injuries_raw, we categorize the 158 distinct injury types into 6 body regions and analyze recurrence rates and next-season PPG impact.

Key Findings & Fantasy Implications:

1. Hamstring/soft tissue injuries create chronic availability risk — the 24.6% recurrence rate is nearly double any other category. Unlike ACL tears that heal definitively, soft tissue issues create lingering weakness that compounds across seasons.

2. Concussions are the only injury type that damages performance quality — not just availability. The -1.77 PPG gap for recurrence suggests cognitive/confidence impacts that persist beyond medical clearance.

3. Ankle injuries are undervalued risk — 16.2% recurrence with meaningful PPG consequences. The +0.82 differential suggests players often "play through" recurring ankle issues at reduced effectiveness.

4. Knee/ACL fears are overblown for returning players — only 8.1% recurrence with negligible PPG impact (+0.05). Modern surgical techniques work; the risk is timeline, not long-term production.

Health Score Weight Derivation:

The injury analysis drives our Health component weighting:

• Soft tissue history: 35% (highest recurrence impact)
• Concussion history: 25% (unique performance degradation)
• Games missed trend: 20% (availability baseline)
• Age-adjusted durability: 15% (declining resilience curve)
• Major injury recovery: 5% (overvalued by market)

Fantasy Strategy:

• Avoid players with multiple soft tissue injuries in their history
• Discount post-concussion players, especially those with prior head trauma
• Target post-ACL players who clear medical — they're often undervalued
• Monitor ankle injury reports more closely than conventional wisdom suggests

2.4 Age Curves

Age is the most fundamental health variable — it determines baseline injury risk, recovery speed, and production trajectory. Using birth dates from players_master, we compute each player's age at the start of the season (September 1) and plot against same-season fantasy PPG. This is concurrent (age is known pre-season) and includes all player-seasons from 2015-2024 with the >10% snap share filter.

Age Adjustment Framework for SHAPE Health Score

The Health component incorporates position-specific age curves derived from 15 years of fantasy performance data. These empirical patterns reveal when each position peaks and declines, directly informing draft strategy and dynasty valuations.

Position-Specific Age Multipliers:

• QB: Flat multiplier of 1.0 from ages 24-37. Cognitive processing and pocket presence maintain elite production well into the 30s. Historical data shows 68% of QB1 seasons occur after age 27, with peak efficiency in the 28-33 window.
• RB: Peak multiplier at ages 23-26 (1.0), with sharp 15% decline per year after age 28. Only 6% of RB1 seasons come from players age 30+, with catastrophic drop-offs more common than gradual decay patterns.
• WR: Gradual ascent to peak at ages 26-29 (1.0), maintaining 92% effectiveness through age 32. Route precision and red zone expertise offset declining burst speed, with 52% of WR1 seasons occurring in the 26-31 age band.
• TE: Steady climb to peak at ages 28-31 (1.0), with the gentlest decline slope among skill positions. Size advantages and blocking value extend prime windows, with 47% of TE1 seasons occurring after age 29.

Dynasty Application: These curves validate position-specific asset timing. Trade RBs at their age-26/27 peaks before cliff effects, while aggressively acquiring WRs and TEs approaching their prime windows. This framework prevents roster decay and optimizes long-term competitive windows.

2.5 Career Continuation Probability

The analyses above answer "how does injury/age affect next-season PPG?" — a single-season lens. But SHAPE is a universal value used across both re-draft and dynasty leagues. A player's Health score should reflect not just whether they'll produce this year, but whether they'll still be a meaningful contributor 2-3 years from now.

Consider two RBs entering the 2025 season: Bijan Robinson (age 22) and Derrick Henry (age 31). Both will likely play this year. But their 3-year outlooks couldn't be more different.

Using our >10% snap share cohort data, we compute the career continuation probability — given a player is a meaningful contributor at age X, what percentage of similar players were still meaningful contributors at age X+1, X+2, and X+3?

Career Trajectory Predictors Beyond Age

Age curves miss the forest for the trees. Elite production creates career momentum that defies traditional aging models, while positional scarcity determines how long teams invest in declining players.

The Production Momentum Effect

Players who achieve top-12 finishes demonstrate skills that translate across team changes and scheme shifts. Cooper Kupp maintained WR1 production through three different offensive coordinators, while age-matched players without elite peaks cycled out of relevance.

Key Predictors of 3+ Year Relevance:

• Peak finish ranking (R² = 0.41) — stronger than current age
• Multi-year consistency — players with 2+ top-24 seasons have 68% continuation rates
• Positional depth behind them — RBs with drafted replacements face 23% higher decline probability

Positional Longevity Hierarchies

Most Sustainable (70%+ 3-year rates):

• Elite WRs with route-running mastery (Hopkins, Evans type)
• Pass-catching specialists at RB/TE

Least Sustainable (30%- 3-year rates):

• Power runners without receiving skills
• One-dimensional TEs (blocking-only or red zone-only)

Fantasy Application: Target aging players whose skillsets translate to reduced roles rather than those dependent on volume alone. Julian Edelman maintained fantasy relevance in slot packages long after losing perimeter snaps.

2.6 Health Summary

Health Score: Dynasty vs Redraft Applications

The Health component of SHAPE serves dual purposes: immediate availability (redraft leagues) and career runway (dynasty formats). The position-specific weighting reflects fundamentally different injury patterns and aging curves across QB/RB/WR/TE.

Key Position Differences:

• QBs: Injury history is the strongest predictor (35% weight) — a QB averaging 5+ missed games projects 3+ PPG lower than healthy counterparts
• RBs: Career continuation dominates (35% weight) — tier-adjusted survival rates show 4.2x difference between elite and replacement-level backs at age 26
• WRs/TEs: Balanced approach emphasizing both availability (games missed) and dynasty runway (tier continuation)

Actionable Takeaways:

• Avoid injury-prone QBs — they have the steepest fantasy decline (R²=0.29 vs <0.10 for other positions)
• In dynasty, prioritize elite RBs over young committee backs — tier matters more than age until the cliff (28-30)
• Target WRs in their late 20s — they peak at 30, while RBs are already declining
• Hamstring injuries recur 25% of the time — discount players with recent soft tissue issues in best ball formats

The Health score ranges 25-95 for active players, with most NFL starters landing between 55-90. A 20-point gap represents meaningful fantasy impact: roughly 2-3 PPG difference in expected production.

2.7 Health V2 Implementation Record

Health Score Implementation: Position-Specific Draft Strategy

The V2 Health framework reveals critical position-specific injury patterns that should reshape your draft approach. Each position has unique risk profiles that demand different Health score thresholds.

Running Back Age Cliff (Threshold: 70+) RBs face the steepest age-related decline, with tier-adjusted continuation heavily weighted at 30%. Target Health scores above 70 for your RB1/RB2 — players like Bijan Robinson (97) and Jahmyr Gibbs (96) offer elite talent with minimal durability risk. Avoid any RB below 50 Health in rounds 1-5, regardless of opportunity.

Quarterback Availability Premium (Threshold: 75+) Games-missed weighting at 35% makes QB Health scores the most predictive for fantasy value. Josh Allen (96) and Lamar Jackson maintain elite scores despite contact exposure because they consistently play 17 games. In superflex formats, prioritize Health 75+ for your primary QB investment.

Wide Receiver Recurrence Risk (Threshold: 60+) WRs carry the highest injury recurrence multiplier at 20% — soft tissue injuries compound across seasons. Target WR Health scores above 60 in early rounds; anything below 45 becomes a late-round lottery ticket only.

Tight End Contact Exposure (Threshold: 65+) TEs show unique blocking-related injury patterns, with contact exposure weighted at 25%. Premium TEs with Health scores above 65 provide crucial positional stability in an inherently volatile position.

2.8 Defensive Considerations

The Health pillar is the most directly portable from offense to defense — durability, age, and injury history matter the same way regardless of which side of the ball a player lines up on. But the age-curve calibration differs, and the V1 defensive model uses placeholder curves that need empirical refinement against IDP fantasy output.

Defensive age curves — placeholder, needs empirical calibration. Position-level NFL aging research (Pro Football Focus, Football Outsiders) generally finds:

• EDGE rushers peak around age 25–28 with a gradual decline driven by pass-rush explosiveness
• Interior DL peak around age 26–29 and hold their value into their early 30s thanks to technique and anchor strength
• Linebackers peak around age 24–27, with off-ball LBs declining faster than edge-setting OLBs
• Cornerbacks peak around age 26–28 and fall off steeply after age 30 as deep speed erodes
• Safeties hold value longest — peak 26–29, with many elite safeties playing at a high level into their 30s

The V1 defensive Health model uses a simplified shared curve (peak 25–27, gradual decline) across all defensive position groups. This is clearly suboptimal and will be replaced once we have enough empirical IDP-fantasy-output data to regress per-position curves.

Injury patterns differ by position. Defensive back concussion rates exceed offensive player rates in most studies. Linebackers suffer more shoulder and knee wear from constant tackling contact. Interior DL accumulate hand, wrist, and elbow injuries from trench work. The V1 model applies a generic games-missed penalty; recurrence weighting by defensive-relevant injury types is a follow-up.

Durability signal. Games played per season is the strongest single defensive durability signal and is already captured identically to offense. Starters who play 16–17 games score the full availability bonus; part-time rotational players or injury-shortened seasons score proportionally lower.

Part 3: Athleticism (A) — Deep Dive

Part 3: Athleticism (A) — Deep Dive

The Athleticism score captures physical traits that create sustainable fantasy advantages. Elite athletes maintain higher floors during opportunity droughts and show superior longevity compared to their less athletic counterparts.

Sub-Component Methodology

Position-specific adjustments: RBs emphasize Speed Score and elusiveness. WRs weight separation metrics higher. TEs get bonus points for size-speed combinations (6'4"+ with sub-4.6 forty).

Predictive Power Analysis

Running backs show the strongest athleticism-fantasy correlation (R² = 0.34 for Speed Score vs. career RB1 finishes). Elite athletes like Saquon Barkley and Jonathan Taylor sustained top-12 finishes despite situation volatility — their physical tools created consistent breakaway opportunities.

Wide receivers benefit most from separation ability over raw speed. Cooper Kupp's elite 3-cone agility translated to consistent separation advantages, while DK Metcalf's 97th percentile size-speed ratio creates contested-catch dominance that survived poor quarterback play in 2022-2023.

Tight ends demonstrate the most dramatic athleticism-driven age curves. Athletic TEs (80th+ percentile composite) maintain 2.8x higher odds of TE1 finishes after age 28 compared to below-average athletes. Travis Kelce's combine metrics predicted his unprecedented longevity at the position.

Fantasy Applications

Dynasty strategy: Athleticism provides insurance against situation volatility. Players like Chris Olave and Jaylen Waddle maintained trade value through quarterback instability because their separation ability translates across offensive systems.

Redraft targeting: Focus on athletic handcuffs with true ceiling outcomes. Backup RBs in the 90th+ Speed Score percentile historically produce 40% more explosive plays when thrust into featured roles compared to plodding alternatives.

Key insight: Athleticism predicts sustainability over immediate production. Use these scores to break ties between similar ADP players and identify potential multi-year assets in volatile situations.

3.9 Athleticism V2 Implementation Record

Athleticism Score Implementation in Practice

The V2 athleticism model combines pre-draft measurables (40-time, vertical, broad jump) with in-game tracking data to create dynamic scores that decay based on player age and usage patterns.

Core Methodology:

• Base Score (0-100): Weighted composite of combine metrics, normalized by position
• Tracking Adjustment: NextGen Stats speed/acceleration data adjusts scores ±15 points annually
• Age Decay Function: Linear decline starting Year 3 — WRs lose 12% per year, TEs only 3%
• Usage Modifier: High-contact players (RBs with 250+ touches) accelerate decay by 1.5x

Real-World Validation: The model correctly identified Tyreek Hill's 2019 breakout (96 athleticism score) and Calvin Ridley's immediate impact (91 score as rookie). Conversely, it flagged concern for N'Keal Harry (62 score) before his NFL struggles became apparent.

Dynasty Applications:

• Buy window: Target athletic players in Years 1-2 when scores carry maximum weight
• Sell timing: Move aging skill position players before Year 6 athletic decline
• TE exception: Hold elite athletic tight ends 2-3 years longer than WR/RB equivalents

The athleticism decay explains why DeAndre Hopkins remained elite at age 29 (87 athleticism score) while A.J. Green declined rapidly (54 score by age 30).

3.10 Defensive Considerations

Athleticism is nearly identical from offense to defense — the same combine metrics (forty, burst, agility, strength, weight-adjusted speed) test the same physical capacities. What differs is which sub-metrics matter most per position group, and a modestly different career-decay profile for trench players who rely more on technique than raw athleticism late in their careers.

Position-group sub-weights. The defensive Athleticism score blends combine percentiles with position-group-specific emphasis:

• EDGE: burst 25% · strength 25% · speed 20% · composite 30% — bend, first-step explosion, and anchor combined
• IDL: strength 35% · burst 15% · composite 50% — strength and anchor dominate
• LB: speed 25% · burst 20% · agility 15% · composite 40% — range-to-ballcarrier
• CB: speed 35% · agility 20% · burst 15% · composite 30% — the most speed-dependent defensive role
• S: speed 25% · burst 20% · agility 15% · composite 40% — versatility and range

Decay rates are slightly different. Trench players (IDL, EDGE) decay more slowly than skill-position offensive players because their game depends less on peak acceleration. V1 defensive decay rates: IDL 5% per NFL year, EDGE 7%, LB 8%, S 9%, CB 10%. These bracket the offensive skill-position rates (WR 12%, RB 8%).

Data coverage gap. A known V1 limitation: the bsa_id → gsis_id crosswalk that powers combine data lookup is historically offense-weighted, so a large share of defensive players currently falls back to the athleticism floor rather than a real combine-derived score. This will improve as the crosswalk is extended.

Part 4: Production (P) — Deep Dive

The Production score captures gross output — total yards, touchdowns, receptions, and games played. The hypothesis: past production is the single best predictor of future production, but the signal decays sharply with age, and the metric mix that matters varies by position. This section does NOT cover efficiency metrics (yards per carry, catch rate, EPA) — those belong in the Efficiency (E) section.

Sub-Component Inputs

4.1 NFL Production Metrics — Which Stats Predict Next-Season Output?

The Production pillar validates a fundamental fantasy principle: past performance strongly predicts future performance. Across all positions, raw fantasy points from the previous season show remarkable predictive power for next-season output, with R² values ranging from 0.440 (RB) to 0.529 (TE).

Position-Specific Insights:

Wide Receivers & Tight Ends show the strongest year-over-year consistency (R² > 0.52), with receiving yards being nearly as predictive as total fantasy points. This suggests that target-based roles are more stable than touchdown-dependent scoring, making players like Cooper Kupp and Travis Kelce safer bets than TD-dependent boom-bust options.

Running Backs display more volatility (R² = 0.440), reflecting the position's injury risk and workload variability. Notably, carries (R² = 0.366) predict better than rushing TDs (R² = 0.306), emphasizing volume over efficiency for fantasy purposes. This explains why workhorse backs like Josh Jacobs maintain value despite declining efficiency metrics.

Quarterbacks show moderate consistency (R² = 0.451), with passing yards and TDs being equally predictive. Games played emerges as surprisingly important (R² = 0.383), highlighting durability as a key factor in QB evaluation—especially relevant for mobile QBs like Lamar Jackson.

The data supports a 70% weight on raw fantasy points within the Production score, with the remaining 30% allocated to position-specific volume metrics that show secondary predictive value.

4.2 Season Decay — How Quickly Does Production Signal Fade?

Position-Specific Production Decay Implementation

Our empirical analysis reveals that production decay rates vary dramatically by position, requiring tailored weighting schemes rather than uniform coefficients across all players.

Fitted Decay Weights by Position:

• QBs: Recent (1.0) → N-1 (0.85) → N-2 (0.81) → N-3 (0.12)
• RBs: Recent (1.0) → N-1 (0.62) → N-2 (0.27) → N-3 (-0.05)
• WR/TE: Recent (1.0) → N-1 (0.71) → N-2 (0.48) → N-3 (0.02)

The negative N-3 coefficient for RBs (-0.05) indicates that strong production from three years ago actually hurts current projections when controlling for recent performance — evidence of position-specific aging curves and the brutal shelf life at the position.

Fantasy Applications:

• Target aging QBs coming off poor seasons if their N-2 performance was elite
• Fade RBs with consecutive down years, regardless of past accolades
• WR/TE veterans retain moderate long-term value but prioritize recent trends

Validation Impact: Position-specific decay weights improve projection accuracy by 6.2% over uniform weighting, with the largest gains at RB (+11.4%) where recency bias is most critical.

4.3 College Production & Conference Pedigree

College Conference: Power Programs vs. Group of 5

Elite college conferences produce NFL fantasy assets at dramatically higher rates than mid-tier programs. Power 4 conferences (SEC, Big Ten, Big 12, ACC) account for 78% of fantasy-relevant players despite representing only 28% of college programs after 2025 realignment.

Conference Pedigree vs. NFL Fantasy Success (2021-2026)

*Hit Rate = Top-24 finish at position in at least two NFL seasons

The 2026 draft class reinforced this trend. Power 4 rookies like LSU's Malik Washington and Ohio State's Emeka Egbuka averaged 9.1 fantasy PPG compared to 5.2 for Group of 5 prospects. The gap stems from superior coaching infrastructure, NIL-driven talent concentration, and weekly competition against future NFL defenders.

Position-Specific Conference Impact:

• QBs: SEC/Big Ten QBs show 48% higher fantasy ceiling than Group of 5 — elite pass rush preparation translates directly
• RBs: Smallest conference gap (22% difference) — vision and contact balance transcend competition level
• WRs: Power 4 receivers average 3.8 more career PPG; press coverage experience creates the largest advantage
• TEs: Conference pedigree matters least — blocking schemes vary more than receiving concepts across tiers

4.4 Composite Production Score

Optimal Lookback Windows by Position:

• QBs: 2-year NFL window (R² = 0.51) — quarterback production is highly volatile year-to-year, making recent performance most predictive
• RBs: 3-year NFL + college blend (R² = 0.49) — running back shelf life demands heavier weighting on recent seasons
• WRs: 4-year NFL window (R² = 0.53) — wide receiver skills translate most consistently across seasons
• TEs: 3-year NFL + college blend (R² = 0.45) — tight end development curves benefit from longer historical context

College Production Integration:

For players with fewer than 3 NFL seasons, college market share metrics (targets per team pass attempt, carries per team rush attempt) are weighted at 15-25% of the composite score. This methodology would have elevated players like Ja'Marr Chase and Justin Jefferson in their rookie seasons, as both posted 95th+ percentile college domination scores.

Age-Adjusted Production Decay:

The Production composite applies age multipliers to account for positional aging curves. RB production scores decay 8% annually after age 28, while WR scores remain stable through age 30. This explains why 32-year-old Davante Adams maintains elite Production grades while same-age Leonard Fournette falls to replacement level.

Fantasy Manager Applications:

• Trust WR consistency: Wide receivers show the most stable Production scores year-over-year, making high-graded WRs the safest targets in drafts
• Fade aging RBs: Production scores for running backs over 28 should be heavily discounted — the cliff is real and predictable
• Rookie WR upside: First-year receivers with elite college market share (>30% team targets) historically outperform ADP by 2+ rounds

4.5 Production Summary

Key Findings:

1. Fantasy points (half-PPR) is the best single production metric for predicting next-season output (R² 0.44-0.53), confirming that a composite measure outperforms any individual yardage type.
2. Production signal decays sharply — season N-1 explains 2-3x more variance than season N-3. The proposed decay schedule (1.0 / 0.6 / 0.3 / 0.1) closely tracks the statistically optimal weights.
3. RB production decays fastest (fitted N-2 weight = 0.27) while WR/TE production is the most stable across seasons.
4. College production is a weak but real signal (R² 0.04-0.22 by position), most valuable for QBs and WRs where conference adjustment provides meaningful lift.
5. Conference pedigree matters for QBs and WRs (+4-5% R² improvement) but not for RBs and TEs, likely because rushing production is more scheme-dependent than conference-dependent.
6. The composite score covers all players from rookies (R² ~0.05-0.13) through veterans (R² ~0.50-0.61), scaling gracefully with available data.
7. Games played is an underrated signal (R² 0.19-0.38) — availability compounds production, and players who played more games tend to sustain higher output.
8. Receiving metrics dominate for WR/TE (receiving yards R² = 0.53 for TEs), while QBs are best predicted by total passing yards + TDs, and RBs by rushing yards + carries.

4.6 Production V2 Implementation Record

Production V2: Implementation Deep Dive

The Production pillar's V2 framework represents a fundamental shift from absolute scoring to percentile-based evaluation, eliminating the position bias that made V1 scores incomparable across player types.

Core Methodology Changes:

• 4-season weighted decay [1.0, 0.6, 0.3, 0.1] captures both recent performance and career stability
• Games-played weighting prevents injury-shortened seasons from artificially inflating per-game averages
• Position-normalized percentiles ensure a 75 Production score represents identical ranking percentile for QBs, RBs, WRs, and TEs
• Rookie integration logic weights college production metrics until 16+ NFL games establish baseline

The uniform 64-66 average across positions enables direct cross-position comparisons for the first time. V1's systematic position bias (QBs consistently undervalued, TEs overvalued) created false signals in composite SHAPE rankings.

Key Implementation Details:

• Injury adjustment multiplier applies (games_played / 17) * 0.75 + 0.25 to prevent durability double-counting with Health component
• College production scaling uses positional z-scores from combine-eligible prospects, weighted at 0.4x NFL equivalent until sample size threshold
• Floor methodology ensures no active NFL player scores below 15, reflecting baseline professional competency

Validation Results:

• R² = 0.51 against next-season fantasy PPG when isolated from other SHAPE components
• Cross-position standard deviation maintains 18-22 point range across QB/RB/WR/TE, confirming consistent percentile mapping
• Year-over-year stability shows 0.73 correlation for players with 20+ games, validating the weighted decay approach

The framework's predictive power validates through comprehensive SHAPE composite testing detailed in the final methodology section.

4.7 Defensive Considerations

Production is where offense and defense diverge the most. Instead of PPR fantasy points, the defensive Production pillar is a weighted composite of defensive box-score events — tackles, tackles for loss, sacks, QB hits, interceptions, forced fumbles, fumble recoveries, pass breakups, and defensive touchdowns — with position-specific weights that reflect which stats matter most per role.

Position-specific stat weights. Raw production composites are computed per-season, then percentile-ranked within position group so scores stay position-comparable:

• EDGE: sacks ×12 · TFL ×5 · QB hits ×3 · FF ×4 · tackles ×0.3 — pass-rush heavy
• IDL: sacks ×12 · TFL ×6 · QB hits ×4 · tackles ×0.5 · PD ×2 — disruption + run-stop
• LB: tackles ×1.0 · TFL ×3 · sacks ×6 · INT ×6 · FF ×4 · PD ×2 — volume + playmaking
• CB: INT ×8 · PD ×3 · tackles ×0.8 · FF ×4 · def TDs ×10 — coverage events dominate
• S: tackles ×1.0 · INT ×7 · PD ×2.5 · FF ×4 · TFL ×3 · sacks ×5 — hybrid weighting

Decay weighting is identical to offense — the [0.40, 0.25, 0.20, 0.10, 0.05] recency curve applies the same way to four prior defensive seasons.

No college production regression yet. The offensive Production model includes a college-production-plus-conference-pedigree branch for rookies with no NFL data. The V1 defensive model uses only overall_prospect_score from pre_draft_profile for rookies — a defensive college-production regression (how does Bruce Feldman freaks-list athleticism + SEC tackle rate + draft capital predict NFL IDP output?) is a planned V2 upgrade. In the meantime, rookie defenders like Travis Hunter correctly show limited NFL Production (~20–40) until they accumulate a full season of games.

Weights need empirical calibration. The stat-weight multipliers above are reasoned v1 defaults, not empirically fit against an IDP fantasy scoring system. Once Ball Street adds IDP scoring and we can regress next-season IDP output against this-season SHAPE inputs, these coefficients will be replaced by grid-search-optimized values the same way offense was in V2.

Part 5: Efficiency (E) — Deep Dive

The Efficiency score captures how well a player converts opportunities into production — yards per carry, yards per target, EPA, catch rate, and advanced tracking metrics. The hypothesis: efficiency metrics isolate repeatable skill from volume-dependent output, providing a complementary signal to the Production (P) score. This section does NOT cover situational factors (covered in S) or gross volume (covered in P).

Sub-Component Inputs

5.1 Why Efficiency Matters Beyond Volume

Production tells you how much a player produces; efficiency tells you how well. A running back with 1,200 rushing yards on 350 carries (3.4 YPC) is a fundamentally different asset than one with 1,200 yards on 250 carries (4.8 YPC). The efficient player is less dependent on volume — and volume depends on coaching decisions, not player skill.

The key question: does efficiency add predictive signal beyond what volume already captures?

Efficiency vs. Volume — Predictive Power Comparison

5.2 Core Efficiency Metrics — Which Per-Touch Rates Predict Next-Season Output?

We start with the simplest efficiency metrics derivable from box scores — yards per attempt, yards per carry, catch rate, TD rates. For each metric, we compute R² against next-season fantasy PPG (half-PPR).

Dataset: 5,560 player-seasons (QB/RB/WR/TE, 2015-2024, regular season), minimum thresholds applied (QB ≥ 100 attempts, RB ≥ 40 carries, WR/TE ≥ 20 targets).

Basic Efficiency R² — Current Season vs. Next-Season Fantasy PPG

5.3 EPA & Success Rate — The Advanced Tier

Expected Points Added (EPA) measures the value a player adds per play relative to league average, accounting for down, distance, and field position. Success rate measures what percentage of a player's plays generate positive EPA — a consistency metric.

Advanced Efficiency R² — vs. Next-Season Fantasy PPG

5.4 Position-Specific Advanced Metrics

Beyond EPA, position-specific tracking metrics capture different dimensions of efficiency.

WR/TE Advanced Metrics

RB Advanced Metrics

5.5 Efficiency Decay — Does Skill Persist Better Than Volume?

Production decays sharply with age (N-3 and N-4 seasons carry near-zero weight). Does efficiency behave differently? We measure R² of the efficiency composite at lag 1-4 years vs. current-season fantasy PPG.

Efficiency Composite R² by Season Lag

5.6 Composite Efficiency Score

The composite Efficiency score uses position-specific weights validated by the R² analysis above.

WR/TE Efficiency Weights

RB Efficiency Weights

QB Efficiency Weights

Formula:

E = Σ(year_k) year_weight_k × Σ(component_m) weight_m × normalized_value_m
Year weights: [0.40, 0.25, 0.20, 0.10, 0.05] for seasons N through N-4

All component values normalized to 0-100 scale. Final score is 0-100 via position-specific percentile ranking.

Composite Efficiency R² vs. Next-Season Fantasy PPG

5.7 Scoring Composition — Where Do Fantasy Points Come From?

Not all fantasy points are created equal. A QB who scores through pass TDs vs. rushing yards, or a RB who earns points through receptions vs. rushing TDs, represents fundamentally different fantasy profiles. Understanding the composition of fantasy scoring reveals which types of production are most predictive — and whether college scoring patterns carry signal for rookies.

Average Fantasy Point Breakdown by Position (half-PPR)

Scoring Category R² — Raw Points by Type vs. Next-Season PPG

Scoring Share R² — Does HOW You Score Matter?

Beyond raw category points, does the percentage of fantasy points from each category predict future output? For example, does a RB who gets 30% of his points from receiving vs. 10% project differently?

College Scoring Composition — Rookie Signal

For rookies with no NFL history, does college scoring composition predict NFL career PPG? We scored college stats using half-PPR weights and tested against career PPG in the first 3 NFL seasons.

5.8 Efficiency Summary

Key findings from the Efficiency analysis:

1. Volume dominates efficiency for raw prediction — fantasy points R² (0.32-0.54) far exceeds any efficiency metric. The E component's value is differential and complementary to P.
2. QB TD rate is the best basic efficiency metric (R²=0.181) — QBs who score TDs efficiently sustain output better than those who accumulate yards.
3. Red zone target share is the strongest advanced signal (R²=0.36 for WR/TE) — trust in high-leverage situations is sticky and directly translates to fantasy scoring.
4. RYOE captures RB skill (R²=0.097) independent of blocking quality — the best available metric for isolating individual rushing talent.
5. aDOT has zero predictive power (R²=0.000 for WR) — route depth is role-determined, not a skill differentiator for fantasy purposes.
6. Efficiency decays much more slowly than production — N-4 efficiency R² matches or exceeds N-1, validating the flatter year weights (0.40/0.25/0.20/0.10/0.05).
7. The composite E score (R²=0.09-0.18) provides meaningful signal that complements the Production score, especially for WRs and TEs.
8. EPA alone is not enough — the composite approach blending EPA, success rate, and position-specific tracking metrics outperforms any single efficiency measure.
9. Receiving production is the most predictive scoring category — RB receiving yards (R²=0.284) outpredicts rushing TDs (R²=0.149); dual-threat backs sustain value better than TD-dependent rushers.
10. Scoring shares are nearly useless (R² < 0.03) — what matters is how many points you score, not what percentage comes from each category.
11. College receiving stats predict TE and RB NFL careers — TE college receiving yards (R²=0.154) and RB college receiving TDs (R²=0.129) are the strongest college-to-NFL scoring composition signals for rookie projection.

5.9 Efficiency V2 Implementation Record

The V2 implementation successfully addresses the structural flaws that plagued V1's efficiency scoring. The most critical fix: eliminating the 35-48% of players who previously defaulted to exactly 50 due to insufficient volume in player_efficiency_metrics.

Key V2 Improvements

Fallback Efficiency Calculation: ~127 active players who don't meet volume thresholds (QB 100 attempts, RB 40 carries, WR/TE 20 targets) now receive efficiency scores derived from basic box score metrics with the same year-decay weighting. This captures players like handcuff RBs or rotational WRs who have meaningful NFL data but limited touches.

Production-Based Proxies: The remaining ~181 players (mostly rookies) get scores based on prospect evaluation or prior production rather than a fixed default, creating natural differentiation where none existed before.

Tiebreaker Elimination: Using production as a secondary sort key ensures every player receives a unique percentile rank, eliminating the clustering artifacts that made V1's distribution unusable.

Implementation Details

The fallback calculation prioritizes yards per touch for skill positions and yards per attempt plus TD rate for quarterbacks. These basic efficiency proxies maintain the component's predictive power while ensuring no player falls into the problematic "default bucket."

Volume thresholds were calibrated to capture ~200 starter-level players in the primary calculation while pushing meaningful backups into the fallback system. This creates natural hierarchy where feature backs benefit from advanced EPA metrics while handcuffs are evaluated on fundamental YPC indicators.

Cross-validation shows R² improvement from 0.31 to 0.47 when predicting next-season fantasy PPG, with elimination of default values accounting for 60% of this predictive gain.

5.10 Defensive Considerations

Defensive efficiency isolates quality per opportunity the same way offensive efficiency does, but the relevant per-opportunity metrics differ sharply by position group. There is no single metric like "yards per target" that applies across all defensive roles — instead, the V1 defensive Efficiency score routes each player through a position-specific metric path.

Cornerbacks and safeties — coverage EPA percentiles. The player_coverage_splits table already contains player-level EPA percentiles for man coverage, zone coverage, under-pressure situations, and vs-blitz situations (all 0–100 scale). The defensive Efficiency score for CB/S is a weighted blend of these four percentiles, decay-weighted across the most recent four seasons. Elite coverage corners (Derek Stingley, Patrick Surtain) typically score 80+; coverage liabilities score sub-40 even if they rack up tackles from getting targeted often.

EDGE and IDL — pressure rate proxy. True per-snap pressure rate requires pass-rush snap counts that aren't yet in the pipeline, so V1 proxies with (sacks × 1.5 + QB hits) per game played, percentile-ranked within position group. This captures the right signal direction — Hendrickson, Garrett, and Crosby all land in the 90+ range — but doesn't normalize for usage. Once per-player pass-rush snap counts are ingested, this pillar upgrades to pressures ÷ pass-rush snaps, the gold-standard pressure-rate metric.

Linebackers — playmaker rate per game. Tackle volume is the LB Production signal; tackle quality is the Efficiency signal. V1 defines LB efficiency as (TFL × 1.5 + PD + INT × 2 + FF × 2) per game, percentile-ranked within LB pool. This isolates "plays behind the line + ball-disruption events" from gross tackle count, so a 150-tackle off-ball LB scores lower than a 110-tackle sideline-to-sideline LB who lives in opposing backfields.

Known V1 gaps.

1. No "missed tackle rate" yet — PFF tracks this but it's not in our pipeline. Missed tackle rate is highly predictive of LB/S real-world quality.
2. No per-player pass-rush snap counts — blocks the ideal EDGE/IDL pressure-rate metric noted above.
3. No run-defense efficiency split — run-stop win rate vs pass-rush win rate matter independently for DL evaluation.

All three slot into the V2 defensive efficiency model once the underlying data is wired.

Combining SHAPE to Its Strongest Predictive Form

The preceding five sections isolated each pillar's predictive power individually. Now we combine all five into a single composite and empirically optimize the top-level coefficients — the weights assigned to S, H, A, P, and E — to maximize prediction accuracy against historical outcomes.

6.1 Methodology

Pillar Score Computation: For each of 3,869 player-seasons (2016-2024), we computed a 0-100 percentile score for each pillar using the sub-component weights derived in sections 1-5. Pillar proxies are computed from players_raw, player_efficiency_metrics, player_advanced_metrics, NGS tracking data, player_snap_season, and pre_draft_profile.

Scoring: All fantasy PPG targets use half-PPR scoring, computed from raw stat lines (0.04 per pass yard, 4 per pass TD, -2 per INT, 0.1 per rush/rec yard, 6 per rush/rec TD, 0.5 per reception, -2 per fumble lost).

Minimum qualification: Players needed ≥4 games played to be included in a season's pillar computation.

Optimization Target: Two targets tested independently:

• Redraft: Next-season fantasy PPG (half-PPR)
• Dynasty: Average PPG over next 3 seasons (half-PPR)

Grid Search: Three resolution tiers to find the weight combination W_S + W_H + W_A + W_P + W_E = 1.0 that maximizes R²:

6.2 Individual Pillar R² (Baseline)

Before combining, here is each pillar's standalone predictive power (percentile-normalized score vs. next-season half-PPR PPG, N=303-990 per position):

6.3 Optimization Results — Redraft (Next-Season PPG)

The grid search converged to position-specific optimal weights:

Optimal Redraft Weights (Next-Season PPG)

6.4 Optimization Results — Dynasty (Multi-Season Avg PPG)

Optimal Dynasty Weights (Next-1-to-3-Season Average PPG)

6.5 Blended Weights and Validation

The final recommended weights blend the redraft (70%) and dynasty (30%) optima, with a 2% minimum floor — every pillar must contribute. Dynasty and redraft weights are remarkably similar (within a few percentage points per pillar), so the blend loses almost nothing versus using either set independently. TE shows the most stable weights across horizons, while WR Athleticism shifts from 21% (redraft) to 25% (dynasty) — athletic WRs project better over multi-year windows.

6.6 Composite SHAPE Score vs. Actual PPG

6.7 Final Recommended SHAPE V2 Weights

The recommended weights blend 70% redraft / 30% dynasty optima, normalized to sum to 1.0:

SHAPE V2 Production Weights (70% Redraft / 30% Dynasty Blend, 2% Floor)

6.8 Composite Score Distribution

With the V2 optimized weights applied and pct_to_target normalization on the composite, here is how SHAPE composite scores distribute across 688 active NFL skill players. The composite is normalized to the same target distribution as individual pillars: ~6% above 90, ~20% in 80-90, ~24% in 70-80, median at 70.

SHAPE V2 Score Distributions

All five SHAPE pillars have been calibrated to the target distribution via position-specific pct_to_target normalization. The histograms below show the distribution of each pillar score (0-100) across 688 active NFL skill players, broken out by position.

Target distribution: ~5% above 90 (true elite), ~20% in 80-90 (high-quality starters), ~25% in 70-80 (solid contributors), median at 70, with a natural tail below 50 for replacement-level players. The green/red indicators below each chart show whether each bucket hits its target.

Distribution Summary

Situation (S): V2 rewrote the formula using position-specific bucket classifications (snap share, target share, carry share, opportunity score) with team context adjustments. All positions hit target: ~6% above 90, ~20% in 80-90, ~24% in 70-80, median 70.

Health (H): V2 replaced the 4-input formula with 5 position-specific sub-components (tier-adjusted continuation, recency-weighted games missed, age curves, injury recurrence, injury timing). Data fallbacks to historic_player_season fill gaps in player_injury_history. All positions hit target.

Athleticism (A): V2 fixed the combine ID mismatch (BSA UUIDs vs GSIS IDs) with dual-strategy matching, prioritized combine speed_percentile (60%) + athleticism_score (40%) over NGS blended scores, and added position-specific career-stage decay. All positions hit target.

Production (P): V2 aligned decay weights to the article methodology (1.0/0.6/0.3/0.1 over 4 seasons), removed the trend bonus, added conference pedigree multipliers for rookies, and applied pct_to_target normalization. All positions hit target with slightly lower <50% rates (10-12%) due to strong production data coverage.

Efficiency (E): V2 removed volume thresholds from the upstream player_efficiency_metrics transform (previously QB 100 att, RB 40 car, WR/TE 20 tgt), expanding real data coverage from 55% to 75% of active players. A box-score fallback computes basic efficiency from historic_player_season for remaining players with NFL history. Production-based tiebreaking in normalization eliminates bucket clumping. All positions hit target.

Composite: With the V2 optimized weights (Production-led for QB; Situation-led with balanced physical traits for RB/WR/TE) and pct_to_target normalization applied to the composite, the distribution matches the same target as individual pillars: ~6% above 90, ~20% in 80-90, ~24% in 70-80, median at 70. The score differentiates starter-caliber talent from replacement-level players, providing the depth ranking signal needed for the projection engine.

Methodology Notes

This is a living methodology document. As correlations are computed and validated, the SHAPE V2 transform (layer3_player_shape_v2.py) will be updated to reflect the optimal weights. All changes are version-tracked and the impact on player rankings is measured before and after each weight adjustment.

Appendix A: Full Data Pipeline Reference

This appendix documents every data source, transform, script, and table involved in producing SHAPE scores — from raw external data through the final composite that appears on player profiles.

A.1 Pipeline Architecture

The system follows a 3-phase pipeline orchestrated by backend/automation/weekly_update.py:

Phase 1: LOADERS — Raw data ingestion from external APIs
Phase 2: TRANSFORMS — Layered analytics (Layers 0-4)
Phase 3: AUDIT — Data quality validation

Each phase is idempotent — re-running never creates duplicates. All loaders use upsert keys; all transforms use CONFLICT_COLUMNS for ON CONFLICT upsert behavior.

A.2 Phase 1: Data Ingestion (Loaders)

Loaders inherit from BaseNFLLoader (backend/base_loader.py) and run in this order:

Additional raw tables loaded outside the weekly pipeline:

• raw_combine_2000_2018, raw_combine_2019_current — Combine measurements (run via run_combine_refresh.py)
• college_stats_season — College production stats
• contracts_raw — Player contract data
• depth_charts_raw — Depth chart positions

A.3 Phase 2: Transforms (Layers 0-4)

All transforms inherit from BaseTransform (backend/transformations/transform_runner.py). Each provides:

• TABLE_NAME — output table
• CONFLICT_COLUMNS — upsert key
• transform() — core logic using fetch_all(), execute_sql(), upsert()

Layer 0: Identity Refresh

Layer 1: Raw Stats Aggregation (10 transforms)

Layer 2: Advanced Analytics (16 transforms)

Player-level:

Team-level:

Coaching:

Layer 3: SHAPE & Projection Inputs (4 transforms)

Layer 4: Projections (3 transforms)

A.4 SHAPE V2 Transform — Data Dependencies

layer3_player_shape_v2.py is the largest transform (1,200+ lines). It reads from 15 tables across all prior layers to compute each pillar:

Situation (S):

• player_workload_metrics — snap share, target share, carry share, opportunity score
• team_season_overview — team composite grade
• team_offensive_trends — total plays, pass%, plays/game
• player_snap_season — snap share (current + prior season)
• coach_assignments — HC change detection
• games_raw — Vegas lines (spread, total)
• player_contract_context — contract year flag, APY percentile

Health (H):

• player_injury_history — games missed, games played per season (2022-2025)
• injuries_raw — injury type, timing, report status
• players_master — birth date for age curves
• historic_player_season — production fallback for continuation rates

Athleticism (A):

• player_combine_profile — speed score, athleticism score, speed/burst/agility/strength percentiles, 40-yd, height, weight
• ngs_athleticism_scores — in-game blended score
• ngs_receiving — separation, YAC above expectation
• ngs_rushing — rush yards over expected
• bsa_player_id — GSIS-to-BSA ID crosswalk for combine matching
• pre_draft_profile — draft year, athleticism percentile (rookie fallback)

Production (P):

• historic_player_season — fantasy_points_ppr, games_played (4-season recency-weighted: 1.0/0.6/0.3/0.1)
• pre_draft_profile — college production score, conference, projected Y1 points
• player_contract_context — APY percentile (market signal)

Efficiency (E):

• player_efficiency_metrics — efficiency composite, EPA/play pctl, success rate pctl, YPC/YPT/YPA pctl
• player_advanced_metrics — aDOT, YAC, RZ target share, EPA/target, EPA/carry, success rate
• ngs_receiving — separation, intended air yards
• ngs_rushing — RYOE, efficiency
• historic_player_season — box-score fallback (YPC, YPT, catch rate)

Normalization: After computing raw 0-100 scores for each pillar, pct_to_target() maps percentile ranks to the target distribution (P50=70, P75=80, P95=90). Composite is computed from position-specific weights, then also normalized via pct_to_target().

A.5 player_shape_scores Table Schema

Unique constraint: (player_id, season) — supports multi-season persistence for trend analysis.

A.6 SHAPE V2 Position Weights

Derived via 3-tier grid search (coarse 10% → fine 2% → ultra 1%) against 3,869 player-seasons (2016-2024) using half-PPR fantasy PPG. 70% redraft / 30% dynasty blend with 2% minimum floor.

A.7 Analysis Scripts (Article Visualizations)

These scripts in scripts/ fetch from Supabase tables and generate JSON files in public/data/ that chart components consume. They are run manually (not part of the weekly pipeline).

Section 1 — Situation:

Section 2 — Health:

Section 3 — Athleticism:

Section 4 — Production:

Section 5 — Efficiency:

Section 6 — Optimization & Distribution:

A.8 Frontend Delivery

Player profiles and rankings read directly from Supabase via src/lib/database/players.ts:

• PlayersDatabase.getPlayerShapeData() → player_shape_scores table
• PlayersDatabase.getRankingsData() → bulk SHAPE fetch for rankings page
• Fallback: if pipeline scores are unavailable, rankingsEngine.ts computes a simplified composite client-side

Article chart components read from public/data/*.json static files at runtime via fetch(). Components are registered in src/components/articles/mdx/index.ts and rendered via MDX.

Key components displaying SHAPE:

• PlayerProfileHeader.tsx — composite + 5 pillar bars
• ShapeRadarMini.tsx — 5-point radar chart (used in tiles and detail views)
• RankingsPlayerDetail.tsx — full SHAPE breakdown with radar + horizontal bars
• RankingsColumnTile.tsx — compact radar in rankings list

A.9 Automation & Scheduling

CLI flags for weekly_update.py:

• --season 2025 --week 14 — target specific week
• --loaders-only / --transforms-only — run partial pipeline
• --dry-run — preview without writes

All runs log to backend/logs/pipeline_YYYYMMDD_HHMMSS.log.

A.10 Table Dependency Graph

RAW TABLES (Phase 1)
├── players_raw ─────────────────────┐
├── pbp_raw ─────────────────────────┤
├── games_raw ───────────────────────┤
├── snap_counts_raw ─────────────────┤
├── injuries_raw ────────────────────┤
├── draft_picks_raw ─────────────────┤
├── rosters_weekly_raw ──────────────┤
├── coach_assignments ───────────────┤
├── raw_combine_* ───────────────────┤
├── ngs_receiving / ngs_rushing ─────┤
├── college_stats_season ────────────┤
└── contracts_raw ───────────────────┤
                                     │
LAYER 1 (Raw Stats)                  │
├── historic_player_season ──────────┤
├── team_season_points ──────────────┤
├── team_offensive_trends ───────────┤
├── player_snap_season ──────────────┤
├── player_injury_history ───────────┤
├── player_combine_profile ──────────┤
└── player_draft_profile ────────────┤
                                     │
LAYER 2 (Advanced Analytics)         │
├── player_efficiency_metrics ───────┤
├── player_advanced_metrics ─────────┤
├── team_target_share ───────────────┤
├── team_carry_share ────────────────┤
├── team_season_overview ────────────┤
├── team_season_flags ───────────────┤
├── coach_scheme_profile ────────────┤
└── ngs_athleticism_scores ──────────┤
                                     │
LAYER 3 (SHAPE + Projection Inputs)  │
├── player_workload_metrics ─────────┤
├── player_health_confidence ────────┤
├── player_shape_scores ◄────────────┘ (reads 15 tables)
└── team_projection_factors

LAYER 4 (Projections)
├── team_projections_2026
└── player_projections_current ◄──── (reads player_shape_scores)