Choose a game
Start with a real NFL matchup.
NFL PICKS. PERSONAL INSIGHT.
Think you know ball?
Put it to the test.
Move the sliders to choose what matters in a matchup. See your prediction change—then compare your read with the final stats.
Test your football instincts. Learn your tendencies. No betting.
Choose what you think will matter
Use the sliders to choose what you think will have the most influence on the game.
Save it. See how you did.
Return after the game for your personal report.
WEEK 3 · EASTERN TIME
Choose your game.
UPDATINGChecking schedules and predictions
NFL WEEK 3THU SEP 24 · 8:15 PM ET
FALCONS at PACKERS
Lambeau Field · Green Bay · WI
LINE PENDING · awaiting verified marketGUIDED OR BLIND?Guided shows the starting numbers.
The starting pointThe starting estimate changes as you move the sliders.
Betting marketPendingCurrent market estimate
ESPN predictorPendingFallback when available
Your starting pointGB 50.0%Neutral · awaiting data
How is my starting point chosen?
We use the market’s win probability with the bookmaker margin removed. ESPN is a comparison, or the fallback if market odds are unavailable. With neither source, we show a neutral 50%. Sliders express your opinion relative to that starting point.
Your prediction+0.0 pts for GB
GREEN BAY · HOME
vsPACKERS
50%ATLANTA · AWAY
FALCONS
50%GB —Estimated chance to winATL —
What do you think will decide this game?
BASIC · 5 KEY FACTORSBest for a quick prediction.
Your callGB 50%ATL 50%
LowHigh
LowHigh
LowHigh
LowHigh
LowHigh
What does my prediction mean for the spread?
MARKET SPREADWaiting for an update
Your cover lean will appear when a verified spread is available.
This is a simple scenario estimate: about 2.5 win-probability points equals one spread point. It is a way to express your opinion, not betting advice.
Use these sliders for other games
COPY SLIDERSSelect one or more matchups.
Your slider importance carries over. Each game keeps its own team data.
Save your prediction before kickoff. Get feedback after the final stats arrive.
Advanced feedback
Your emphasis, beside the final-game stats. This comparison does not change your Reasoning Score or leaderboard rank.
Save a prediction to start. Comparisons appear after final stats are published.
02
AFTER THE FINAL WHISTLE
Your take meets
what happened.
Compare what you thought would matter with our estimate from the final stats. See whether each expected team advantage held or reversed.
WAITING FOR YOUR FIRST CALLSAVE A CALL ABOVEYour review appears after the final stats arrive.
HOW IT WILL WORKYour locked weights stay frozen. Final passing, rushing, defensive disruption and turnover data becomes the comparison.
Waiting for the final whistle. Your original prediction and pregame advantages stay saved.
FEEDBACK CALC · v0.6The same five categories before and after kickoff.
Quarterback 25% · passing offense 20% · rushing offense 20% · overall defense 15% · turnovers 20%. Your 0–10 importance rating is compared with each category’s realized 0–10 game signal; the weighted gap becomes a 0–100 Reasoning Score.
Final stats must be present for both teams. Until then, the report says waiting or processing—never a fabricated grade.1LOCK
→Your weights and prediction freeze before kickoff.
2MEASURE
→Verified final team-game stats are scored against the pregame expectation.
3COMPARE
→Your guessed importance is matched to the realized game signal.
4REFRESH
The completed game enters each team’s next rolling rating.
03
MODEL v0.5 · REFERENCE + USER CALL
How the prediction
actually moves.
The reference uses the outside signal that performed best in testing. Your slider result is a game-specific belief recorded for later feedback—not a claim that the app has found a better probability.
REFERENCEDe-vigged betting marketOur provisional historical test found that adding ESPN did not lower probability error. ESPN remains beside it as a useful comparison.
+YOUR SCENARIOEdge × importance × factor capYour combined belief can move the reference by up to ±10 points.
=YOUR CALLA testable football takeSaved before kickoff so the report card can grade your reasoning.
Quarterback
Passing EPA per attempt + completion percentage over expectation
nflverse team-game statsTeam-level QB proxyPassing offense
Season-to-date passing EPA per attempt before this game
nflverse team-game statsOverlaps with QBRushing offense
Season-to-date rushing EPA per carry before this game
nflverse team-game statsGame-script sensitivePass defense
Pregame: opponent passing EPA suppressed · Postgame: sacks, hits + interceptions
nflverse team-game statsCorrelated with opponent offenseRun defense
Opponent rushing EPA per carry suppressed before this game
nflverse team-game statsCorrelated with opponent rushingTurnovers
Takeaways minus giveaways per completed game
nflverse team-game statsHighly volatilePass protection
Sacks allowed per quarterback dropback
nflverse team-game statsQB behavior also affects sacksExplosive plays
20+ yard passes and 10+ yard runs / (pass attempts + sacks + carries)
nflverse team-game statsTeam proxy, not individual player talentRecent form
Net passing + rushing EPA per game, last up to five completed games this season
nflverse team-game statsNot opponent-adjusted; overlaps with offense/defenseDiscipline
Penalty yards per game; fewer is better
nflverse team-game statsNot a coaching or decision-quality gradeKicking & punting
Equal-weight league-relative kicking conversion and net punt average
nflverse team-game statsNot adjusted for kick distance or field positionHome field
Fixed home prior; zero for neutral-site games
nflverse scheduleVenue calibration comes laterRest & travel
Current model uses rest-day differential
nflverse scheduleTravel and weather come laterADVANCED DATA
Pass protection, explosive plays, recent form, penalty discipline and kicking/punting use completed games before the matchup’s week. Each measure is compared with the league, capped at two standard deviations, then reduced by sample size: games / (games + 2). Kicking and punting are normalized separately and weighted equally. Missing inputs disable only the affected measure. These are exploratory, overlapping proxies—not newly backtested probability improvements. The app checks on visits, when you return to the tab, and every five minutes while open. Source refreshes are cached for 30 minutes; they depend on the source publishing final stats. Injuries and playoff stakes remain inactive.
POSTGAME IMPACT · LIVE
Turn final stats into the same five core categories.
Passing EPA, rushing EPA, CPOE, defensive disruption and net turnovers are combined into quarterback, passing, rushing, overall defense and turnover signals. Each is scaled to 0–10 importance.
YOUR FEEDBACK · LIVE
See what you over- and undervalued.
A locked 8/10 rating versus a 4.5/10 realized signal is “overvalued by 35 points” on the 0–100 importance scale. Winner accuracy stays separate, so a lucky pick does not become proof of strong reasoning.
CONNECTEDnflverse team stats
Passing/rushing EPA, CPOE and turnovers
New finals feed the next week automaticallyCONNECTEDnflverse schedules
Game ID, kickoff ET, venue, rest, odds and results
Validated on refresh; stable IDs prevent duplicatesNEXTDedicated odds provider
Named-book consensus and exact quote timestamps
Needed for a production-grade market historyCONNECTEDESPN predictor
GameProjection from ESPN’s structured matchup feed
Cached with ESPN’s update time and checked every 30 minutesDATA RULES
Only information available before kickoff can enter a prediction. Current team ratings use 2026 games completed before that matchup and are standardized across the league. Final team-game rows require a unique game/team key; missing final stats produce “processing,” never a fabricated grade.
The backtest evaluates the automated factor model. Human slider judgments are graded as beliefs, not treated as scientifically proven effects.04
PREDICTION MODEL · WALK-FORWARD BACKTEST
The market stays
the anchor.
We tested every 5-point market/ESPN mix plus four football-metric sets. Choices were made on 2024, then evaluated on 255 held-out 2025 games. Market-only produced the lowest probability error, so ESPN is now a comparison—not a forced ingredient.
TRAIN2022–2024766 eligible games
→HOLDOUT2025255 unseen games
→RULEPAST WEEKS ONLYFeatures use earlier weeks
WHAT IS A BRIER SCORE?
It asks one simple question: how close were the predicted percentages to the final results? Lower is better, and a confident wrong prediction is penalized more than a cautious one. The friend leaderboard turns it into an easier 0–100 Confidence Score, where higher is better.
2025 CALIBRATION · HOME WIN CHANCEMarket predictions stayed closer to observed results.
Each row is an equal-count probability bin. Perfect calibration means the predicted chance and actual win rate match.
MARKET26%30% actual
MARKET42%43% actual
MARKET56%49% actual
MARKET68%69% actual
MARKET82%80% actual
WHAT IT MEANS
01
More winner picks did not mean better probabilities. The 80/20 blend picked one additional winner out of 255 games, but its confidence was less accurate. Probability error is the primary score.
02
Better football metrics were tested. Opponent passing EPA, pass protection, pressure, explosiveness and discipline joined the original set. A validation-selected ±2-point model scored 0.2145 on holdout—still behind 0.2137.
03
Your ±10 range stays. It is deliberately larger than the automated evidence supported so your opinion is visible, memorable and gradeable.
04
Feedback—not fake precision—is the payoff. The report card compares what you emphasized with postgame signals and tracks what you repeatedly over- or undervalue.
REPRODUCIBLE
The checked-in notebook contains eligibility rules, feature construction, all searches, holdout metrics and limitations. ESPN exposes retained historical projections—not a revision-by-revision archive—so source-blend results are useful but provisional. Week 1, ties, missing odds and games without prior stats are excluded.
05
FRIENDS LEADERBOARD
Who really
knows ball?
Compare both picking winners and using believable percentages—not betting profit. Rankings use Confidence Score first, then Correct Picks, and become official after three graded games.
EMAIL REQUIRED
Save the outcome. Build your reputation.
Your verified sign-in email connects every locked call, postgame report and friend comparison to one account. It is never shown on the leaderboard.
06Don’t just count wins.
YOUR FILM ROOM
Don’t just count wins.
Learn your game.
Your record shows how you think—not only whether your pick won. As finals arrive, it tracks the factors you read best, what you repeatedly over- or undervalue, and whether your judgment is improving.
SUNDAY SENSE REPORT0 LOCKED
NO CALLS LOCKED YETYour first prediction starts the record.
Build a matchup, reveal your prediction, and save it before kickoff. Your learning profile grows after each final.
ABOUT THE CREATOR
Built to test
football conviction.
Sunday Sense is a football prediction game. Choose what matters with sliders, save your call, and learn from the final stats—not just whether your team won.
This evolving portfolio project explores sports data, probability, user judgment and honest postgame feedback.
gavinstanesa@yahoo.com