Position-relative inputs
Quarterbacks, receivers, linemen, defenders, and specialists are judged with different statistical contracts. Raw fantasy points do not become an overall rating.
METHODOLOGY
Every grade needs
a receipt.
NFSMELL is a position-relative forecast system, not a reputation poll. Its statistical core may include a separately labeled, independently reviewed assignment-execution component when strict publication gates pass. Public pages expose the basis, sample, rank eligibility, and model version needed to challenge a result.
Models: player-forecast-v41 / player-performance-v18 / team-strength-v39Context: coaching-context-v1 Data updated: October 9, 2026 at 8:29 PM EDTVerify live snapshot Verify release manifest Verify deployment
PLAYER ENGINE
How a player rating is built
Percentile scoring
Each qualifying metric is compared with the distribution for that position and season. Lower-is-better inputs, such as turnover or missed-tackle rate, are inverted before weighting.
Sample shrinkage
Small samples move toward a conservative position baseline. More qualifying plays allow observed performance to carry more of the score.
Reviewed assignment evidence
Charted execution can become a capped player-only component after canonical play verification, independent review, assignment-facet coverage, confidence, agreement, and same-position cohort gates. Missing or unreviewed film evidence adds zero points.
Forecast and resume
QB Index blends a validated QBR projection, opponent-adjusted current form, and a maturity-shrunk NFL resume. Non-QB headline weights are selected separately by position; when added complexity fails the holdout gate, the simpler forecast order wins.
One public overall
Each non-QB position uses the pre-registered blend tested against a statistical production target in the untouched 2024-2025 release gate. Every receipt shows observed production, rate-based signals, and forecast weights. That does not independently validate assignment execution, blocking or coverage quality. Current-team impact remains separate, and limited execution evidence takes priority over exact quality-order claims.
PRODUCTION CONTEXT
Every position has a different football contract
Player OVR answers how good the player is relative to his position. This separate layer asks how the staff, assignment, opponent, health, workload, and likely game script change what that position can produce. Only effects that pass their own validation gate move the central forecast.
Offensive play caller, pass rate, protection, designed rushes, and red-zone philosophy
Season: team pass environment · position schedule · verified starter status · designed-rush role
Weekly: opponent pass defense · pressure and protection · projected pass rate · score state · weather
Health and role: throwing or lower-body limitation · starter certainty · designed-rush restriction
Gate: QB play-caller usage was tested, but improvement was not stable in every walk-forward fold, so it changes uncertainty rather than the central forecast.
Offensive play caller, run design, backfield rotation, target share, and goal-line usage
Season: position schedule · verified touch plan · committee depth · goal-line role
Weekly: opponent run defense · box environment · projected score state · target opportunity · weather
Health and role: touch cap · managed workload · return-to-play ramp · active backup allocation
Gate: RB play-caller usage was tested, but player-level gains were not stable in every fold. Verified workload receipts can still move opportunity directly.
Offensive play caller, personnel grouping, blocking assignment, and route deployment
Season: personnel package role · active roster status · verified touch plan
Weekly: personnel grouping · short-yardage plan · opponent front · score state
Health and role: package availability · blocking assignment changes · backfield injuries
Gate: Public data does not yet support a stable player-level fullback coaching adjustment.
Offensive play caller, route participation, target hierarchy, formation, and motion
Season: position schedule · verified starting role · route share · target competition
Weekly: opponent coverage strength · likely shadow coverage · pressure-to-target effect · score state · weather
Health and role: route restriction · snap ramp · lower-body limitation · teammate target vacancy
Gate: WR play-caller usage was tested and did not beat the player baseline consistently enough to move the central forecast.
Offensive play caller, route-versus-block assignment, target share, personnel rate, and red-zone usage
Season: validated play-caller TE usage · position schedule · route role · target competition
Weekly: opponent TE coverage · linebacker and safety matchup · route expectation · score state · weather
Health and role: route restriction · blocking-only deployment risk · snap ramp · target vacancy
Gate: The bounded TE play-caller usage prior improved retained-player PPR error in every available walk-forward fold.
Offensive play caller, line coach, protection system, run concept, and assignment continuity
Season: starter assignment · unit continuity · availability · scheme fit
Weekly: opponent front · pressure package · run-front structure · protection help · game script
Health and role: starter availability · position switch · snap limitation · adjacent-line continuity
Gate: Assignment-quality public coverage is too thin for a central coaching adjustment; the context remains explicit and uncertainty-only.
Defensive coordinator, front, alignment, rush-plan role, and rotation
Season: alignment · rush share · rotation · front structure
Weekly: opposing tackle and protection · dropback volume · time to throw · chip rate · score state
Health and role: rush snap cap · rotation change · hand or lower-body limitation · teammate pressure vacancy
Gate: Coordinator identity is tracked, but player-level pressure and sack outcomes have not cleared a cutoff-safe DC transition test.
Defensive coordinator, front technique, interior rush role, run-fit responsibility, and rotation
Season: technique and alignment · pass-rush share · run-down share · rotation
Weekly: opposing interior line · double-team environment · dropback volume · run rate · score state
Health and role: snap cap · rotation change · return ramp · adjacent-line availability
Gate: Interior role is position-specific, but DC transition has not yet earned a central player-production multiplier.
Defensive coordinator, base package, coverage assignment, blitz rate, and tackle opportunity
Season: every-down role · package participation · coverage-versus-rush assignment · green-dot role
Weekly: opponent play volume · run rate · middle-field targets · blitz assignment · score state
Health and role: sub-package role loss · snap cap · return ramp · teammate tackle vacancy
Gate: Public role assignment is not complete enough to convert a DC change into stable individual tackle or coverage production.
Defensive coordinator, coverage family, man-zone mix, alignment, and matchup assignment
Season: outside-slot alignment · coverage family · shadow role · target deterrence
Weekly: opposing receiver alignment · target expectation · route tree · pass volume · safety help
Health and role: snap restriction · alignment change · secondary injuries · return ramp
Gate: Coverage responsibility and target causality remain too inferred for a direct DC-based central adjustment.
Defensive coordinator, shell, box rate, slot role, blitz usage, and deep responsibility
Season: box-versus-deep role · coverage shell · slot participation · blitz share
Weekly: opponent route distribution · run rate · tight-end usage · deep-pass rate · score state
Health and role: alignment change · snap restriction · secondary injuries · return ramp
Gate: Safety roles are highly scheme-sensitive, but public assignment evidence has not cleared a DC transition gate.
Head coach decision policy, offense scoring environment, venue, and field-goal opportunity
Season: team scoring projection · job security · coach decision policy · venue mix
Weekly: team total · red-zone stall rate · wind and precipitation · surface · game script
Health and role: active job status · leg limitation · replacement competition
Gate: Exact kicker ordering remains near baseline, so coaching context cannot create a precise central rank.
Head coach decision policy, offense field position, game script, and venue
Season: job security · team drive environment · venue mix
Weekly: expected drive outcomes · score state · wind · field position
Health and role: active job status · leg limitation · replacement competition
Gate: Punter ordering is experimental and receives no direct coaching multiplier.
Special-teams operation, unit continuity, and active roster status
Season: job security · operation continuity · availability
Weekly: weather · surface · special-teams operation
Health and role: active job status · replacement operation
Gate: Long-snapper value is modeled through availability and operation integrity rather than fantasy-style volume.
TEAM ENGINE
Roster strength becomes matchup strength
The team model weights the positions that influence offense, defense, and special teams instead of taking a flat average. Quarterback receives the largest share while specialists carry far less. Injury status can reduce a player's current-team contribution without rewriting his underlying ability rating.
Current roster value blends 80% canonical player forecast with 20% latest production, then calibrates that position-weighted roster around the live league distribution. The team grade uses a 56% roster / 44% results base contract selected on 2018-2022 and evaluated on 2023-2025 across 256 reconstructed team-seasons; current-season results are sample-attenuated until more games are complete, so Week 1 cannot replace the roster baseline with one noisy result.
Postseason games are scored as a separate receipt against regular-season opponent strength inside the prior-results signal. A 2001-2025 walk-forward test selected a 15% playoff extension cap: next-season rank correlation improved from 0.420 without playoffs to 0.424 with the capped extension across 795 team-seasons. A long playoff run matters, but it cannot become the entire results signal.
Current reserve status, prior-season reserve status, and player-specific return-to-play evidence can apply a temporary availability deduction. Major-injury recovery fades only after verified current-season games establish workload; calendar time alone does not clear it.
Current head coach, offensive coordinator, actual offensive play caller, and defensive coordinator are compared with the prior season on all 32 teams. Staff changes determine whether the previous style profile is a strong, partial, or weak transfer prior and expose transition uncertainty. The broad coaching-context contract contributes zero player OVR, team-rating, and betting points. A separate player-coaching-usage test earned a bounded tight-end production adjustment after improving every available walk-forward fold; all other staff effects remain uncertainty-only until their own position-level test validates.
The public outlook keeps that NFSMELL team grade independent, then blends in a de-vigged season win-total signal. A one-sportsbook snapshot starts at 15%; a multi-book consensus may start at 20%. After seven days the market share decays linearly with freshness, and the unused share returns proportionally to the independent roster/results core. Market rank and model rank remain visible so disagreement is never hidden.
The matchup model starts from those roster inputs, then adds home field, availability, rest, weather, and market context where those inputs are available.
Spread, total, and moneyline settlement probabilities use a discrete outcome lattice generated from 4,431 completed 2009-2025 nflverse regular-season and playoff games. The lattice is shifted to the projected matchup mean while preserving NFL key margins, integer scores, ties, and pushes. Push probability is returned to the bettor and never counted as a win or loss.
Head-to-head streaks are recency-, coach-, and quarterback-decayed. A 4,619-game training test selected zero projection weight, and an 816-game holdout did not justify overriding that decision. Rivalry history is shown as explanatory context, never hidden inside the pick.
Every observed spread, total, moneyline, and available sportsbook quote is timestamped as a distinct market state. The betting receipt identifies movement direction, key-number crossings, reversals, volatility, freshness, and source breadth. "Opening" always means NFSMELL's first observed price unless a provider explicitly supplies the true origin line.
Five seasons of cutoff-safe nflverse closing lines provide ATS, over/under, favorite/underdog, venue, recent-form, and comparable-spread context. Historical market trends currently receive zero direct model points; weights will be selected only through walk-forward CLV and prediction testing.
Verified player contract terms are stored separately from football ability. Exact incentives, per-game active bonuses, extension windows, and option decisions can become role or late-season fantasy context, but generic contract-year motivation receives zero points. A contract feature cannot alter a projection until its terms, progress, coaching mechanism, and incremental walk-forward lift are all verified.
The historical backtest tool is a rolling points-form proxy because equivalent historical pregame roster snapshots do not yet exist. It can test cutoff safety and reject weak baseline ideas, but its results are not advertised as performance of the public roster-and-context model.
Prediction quality is evaluated only from snapshots captured before kickoff. Records captured at or after kickoff are rejected to prevent future-data leakage.
DATA SOURCES
What feeds the model
nflverse play data
Play-level efficiency, success, usage, regular season, and separately weighted postseason evidence.
ESPN context
Total QBR, team news, schedule, standings, roster identity, and injury context when available.
FTN charting
Charted quarterback mistakes and context, including interception-worthy throws and QB-fault sacks when joined coverage is available.
Sportsbook markets
Dated spreads, totals, moneylines, per-book quotes, and first-observed movement states support market receipts and closing-line-value evaluation. Listed hold is removed where prices allow it.
Historical schedules
Cutoff-safe nflverse game history supplies recent series records, streaks, venue, playoff meetings, and personnel-continuity decay.
Early-status newswire
When configured, recent search is restricted to official teams, named national reporters, an explicit local-beat allowlist, and corroboration-only rapid aggregators. Anonymous rumor accounts are excluded. Every eligible post receives a source tier and flags a verification lead only; social posts never move a pick directly.
Verified contract watch
First-party announcements and directly attributable contract reporting supply disclosed incentive thresholds, option decisions, and extension windows. Restricted bulk contract databases are not scraped into the public product.
Coaching and play callers
Dated all-team staff guides and first-party team reporting identify the head coach, coordinators, and actual offensive play caller. Title changes and play-calling changes are tracked separately because they do not imply the same level of scheme turnover.
Model receipts
Version IDs, component scores, qualifying samples, and source notes are retained with the finalized rating snapshot.
Reviewed assignment charting
NFSMELL stores derived labels and review receipts, not copied game film. Unverified plays, single-review labels, incomplete positional facets, and undersized peer cohorts remain research-only and cannot change a public grade.
IMPLEMENTATION STATUS
What is live, shadowed, or still research
A named idea is not automatically a production feature. Every major capability below states its current impact and the gate required before it can change a grade, forecast, or decision.
Current performance and separate player forecasts
Current impact: Qualified current-season player grades/ranks; separate forecast OVR and team value
Evidence: player-performance-v18 supports position-relative current production. player-forecast-v41 supplies the separate evidence-shrunk forecast. Current percentiles are not injected into team strength as a second form overlay. These are not all-snap film grades or cross-position WAR.
Activation gate: Already active; changes require time-ordered validation
Availability-adjusted team power
Current impact: Team power and matchup inputs
Evidence: team-strength-v39 combines canonical player team values, opponent-adjusted results and availability. Current player display scores remain separate; no sportsbook rank sets this ordering.
Activation gate: Every new central effect requires time-ordered validation and a released receipt contract
Schedule-adjusted season simulation
Current impact: Expected regular-season wins and schedule-adjusted rank
Evidence: season-wins-v1 uses released team-strength inputs and the complete schedule. Game-result variance is not a complete model-confidence interval.
Activation gate: Ranges must remain labeled as game-result variance, not complete model-confidence intervals
Game-specific matchup context
Current impact: Matchup projections and explanatory unit edges
Evidence: Current roster, unit strength, venue, rest, availability, weather readiness, and bounded position-group comparisons
Activation gate: Route, coverage, protection, and tracking-level interaction claims stay withheld until qualified data and holdout lift exist
Pregame market and betting receipts
Current impact: Bet/PASS decisions only after evidence gates
Evidence: Book, line, price, timestamp, de-vigging, closing capture, settlement
Activation gate: No market may publish without attributable pre-kickoff evidence
Format-specific fantasy decisions
Current impact: Fantasy-only rankings and decision values
Evidence: Format-specific fantasy decisions use current team-strength-v39 opponent pass/run inputs, workload, schedule and published role evidence. These are fantasy outputs, not replacement player talent grades.
Activation gate: Fantasy output never rewrites player OVR
Player production context
Current impact: Downstream season production and weekly fantasy decisions; zero player OVR or team-power impact
Evidence: Attributable workload, coaching and schedule context supports downstream production. Current workload and line-clearing probabilities require separate chronological evaluation; coverage is not guaranteed for every profile.
Activation gate: Every central effect must be source-attributed, cutoff-safe, position-specific, and improve held-out player production; otherwise it is uncertainty-only
Week 1 player package comparison
Current impact: Fantasy package comparison only
Evidence: fantasy-trade-v2 uses zero-based value above replacement, heuristic package ranges, duplicate-player rejection, and replacement-level treatment of open roster slots
Activation gate: Rest-of-season or dynasty market claims remain withheld until attributable market history and time-ordered validation exist
Coaching continuity and scheme context
Current impact: Descriptive context and uncertainty only
Evidence: All-team HC, OC, play-caller, and DC transition registry
Activation gate: Cutoff-safe ablation must improve held-out prediction
Verified player contract context
Current impact: Receipt context only
Evidence: Directly attributable incentive, option, and extension records
Activation gate: Terms, progress, mechanism, and incremental lift must all validate
Fantasy acquisition activity
Current impact: Directional acquisition radar only
Evidence: Timestamped Sleeper add/drop activity
Activation gate: Activity counts are never treated as ownership or ADP
Team challenger portfolio
Current impact: None
Evidence: Production, reduced-DT, elite-offense, and combined frozen ledgers
Activation gate: Predeclared prospective season scorecard
Reviewed play-assignment execution
Current impact: None without a published qualified evidence snapshot
Evidence: Ingestion, canonical-play, independent-review, adjudication, and abstention infrastructure exists; no complete public all-snap corpus is published
Activation gate: Qualified corpus plus untouched holdout lift
Role Velocity and Role Acceleration
Current impact: None
Evidence: Point-in-time ingestion, as-of queries, weekly nflverse offensive snap shares, structured-source gates, and explicit abstention
Activation gate: Repeated timestamped route, snap, package, and valuable-touch observations
Attributable fantasy ADP
Current impact: Draft-cost context only; no grade or projection changes
Evidence: Daily Fantasy Football Calculator mock-draft ADP, independently timestamped and matched by scoring, league size, team and player identity
Activation gate: Fresh attributable 1QB redraft feed required; no superflex/dynasty substitution or automatic bargain claims
Injury survival and workload restoration
Current impact: None
Evidence: Forecast and settlement contract; current public adjustments remain sourced workload scenarios
Activation gate: Time-ordered calibration, workload error, and stability tests
Owner and subscriber phone alerts
Current impact: None
Evidence: verified-alerts-v5, private-signals-v4, and subscriber-alerts-v1.2: durable retries, actionable owner notifications, five-minute league injury checks, preseason-only suppression across structured and verified-news availability alerts, material-change suppression, attributable material-news classification, week-aware immutable pick receipts, kickoff and retry expiry, owner ntfy delivery, consent-based browser push, and token-scoped device-state verification
Activation gate: Owner alerts require a private webhook; visitor alerts require explicit browser permission and selected categories
Official X account publishing
Current impact: None
Evidence: x-publisher-v3 uses X API v2 user-context authentication when funded, plus a zero-cost X Web Intent fallback that sends a deduplicated Post to X approval button through the private owner alert channel. Immutable source receipts, stale-pick expiry, verified team-update gates, and polished templates apply in both modes.
Activation gate: Full unattended publishing requires official credentials, a positive daily cap, a positive monthly budget, and explicit autopublish enablement. Manual one-tap approval remains available at zero API cost.
Play-level expected opportunity
Current impact: None
Evidence: Architecture defined; no calibrated route-level xTargets/xTD model
Activation gate: Point-in-time route and play-design dataset plus held-out calibration
Correlated fantasy game distributions
Current impact: None
Evidence: Current production publishes point estimates and heuristic planning ranges, not Monte Carlo player quantiles or a joint game distribution
Activation gate: Reconciled opportunity shares, correlated stat simulation, rolling-origin calibration, and empirical interval coverage
Team process and sustainability decomposition
Current impact: None
Evidence: Current prior-results input is opponent-adjusted points-based strength; a separate turnover, conversion, pressure, coverage, and regression dashboard is not live
Activation gate: Cutoff-safe play and drive features, opponent adjustment, stability testing, and a public process-versus-results receipt
Phase-level special-teams value
Current impact: None beyond the current coarse roster unit input
Evidence: Kicker and punter player proxies are live, but field-goal, kickoff, punt, return, coverage, and operation EPA are not independently modeled
Activation gate: Opportunity-adjusted phase data, weather and venue controls, and held-out team-forecast lift
Causal coaching and decision value
Current impact: None
Evidence: Staff identity and transition uncertainty are live context; fourth-down, clock, sequencing, adjustment, and development value are not causal production grades
Activation gate: Effective-date decisions, roster and opponent controls, partial pooling, and cutoff-safe ablation
Cross-position player value above replacement
Current impact: None
Evidence: Current-team contribution is a roster-model input, not a standardized player WAR or contract-value metric
Activation gate: Counterfactual replacement model, position conversion, uncertainty, and future team-outcome validation
Tracking, matchup geometry, and interaction graphs
Current impact: None
Evidence: Research blueprint only
Activation gate: Permitted tracking data, reproducible features, and independent validation
MODEL BOUNDARIES
What the current release does not claim
NFSMELL separates production features from the richer research architecture. Missing inputs stay visible and carry zero hidden weight.
Forecast, not universal film truth
Player OVR separates recent production, public-data ability proxies, forward projection, and current-team impact. It is not an all-snap execution grade, licensed tracking model, standardized WAR, or medical availability probability.
Power, not every organizational layer
Team-strength-v35 is a frozen availability-adjusted roster and opponent-adjusted prior-results model. A full process-versus-luck sustainability score, phase-level special teams model, and causal coaching grade are not live.
Decision range, not calibrated distribution
Fantasy-decision-v1.5 applies scoring, opponent, game script, workload, availability, and staff-transition context. Player-production-context-v1 keeps ability OVR separate while carrying those verified downstream effects onto every player page. The TE play-caller usage adjustment is live after walk-forward improvement; other position and defensive-coordinator effects remain uncertainty-only until they validate. Planning ranges are heuristic, not calibrated quantile intervals or a correlated player-game simulation.
HYPOTHESIS GATE
Interesting is not the same as predictive
NFSMELL is designed to test overlooked football ideas, including body measurements, role changes, scheme interactions, weather, travel, matchup geometry, and market movement. A signal must have a timestamp, a football mechanism, and an out-of-sample test.
Quarterback hand size is the current example. In 70 qualifying 2008-2022 quarterbacks, adding hand size produced a tiny improvement in held-out EPA/play error and a more interesting fumble-rate signal, but worsened CPOE and turnover-rate error. Small-hand status therefore remains a visible tag with zero rating or betting weight.
This gate is deliberate. NFSMELL can investigate unusual claims faster than mainstream rankings without quietly converting a coincidence into a confident number.