NFL

Changelog

Open notes: what the system is built from, what every run does, and every change we've made · predictions are published to show the model works. nothing here is betting advice.Last updated: predictions Oct 1, 11:02 AM ET · results graded Sep 28, 11:22 PM ET

What every run does

Runs at 7:00 AM, 11:00 AM, 3:00 PM and 7:00 PM Eastern every day, plus 11:45 AM, 4:10 PM and 8:00 PM on Sundays and 8:00 PM on Mondays and Thursdays. The steps, in order, every time:
  1. Fetch: the latest schedule and results, play-by-play, injury reports and depth charts from nflverse; finals are also picked up from ESPN's scoreboard within minutes of a game ending.
  2. Grade: every finished game with a stored pick gets a result row: straight-up, against the market margin, over/under, and the the market benchmark. Each analyst claim on those games is graded too.
  3. News: public feeds are read and tagged by team; new items are attached to the games they concern.
  4. Analyst: games with new items get one read; claims that pass the filters and aren't already counted become a capped adjustment. Declined claims are kept as shadow claims.
  5. Predict: the model retrains on every completed game since 2009 and re-predicts every game in the next eight days. The last run before kickoff is the pick of record.
  6. Markets: a snapshot of the DraftKings price (via ESPN), Polymarket, Kalshi and the nflverse price, for open-to-close comparison.
  7. Rankings and projections: power ratings and a 5,000-season simulation for wins, playoff and championship prices; then the player rankings.
  8. Write-ups: when a week's last game is final, a review is written and published; a preview is published a couple of days before the next week's first kickoff.

What it's built from

The ingredients, without the exact recipe.
  • Team strength: an Elo rating with margin-of-victory and offseason regression.
  • Efficiency: opponent-adjusted expected points added per play, for offense, defense, passing, rushing and success rate, weighted toward recent games and the current season.
  • Quarterback: a rating of the named starter (recency-weighted EPA per dropback and completion percentage over expected, shrunk toward a prior for small samples) and a flag when the starter changed.
  • Availability: position-weighted listed starters Out or Doubtful, from the official report and depth charts.
  • Situation: rest days, short weeks, byes, division games, home field (zero at neutral sites).
  • Models: a regularized logistic regression for win probability; ridge regressions for margin and total. Retrained weekly. Tested walk-forward, 2010–2025, with a leak audit.
  • Analyst layer: news → typed, capped, source-linked claims, applied in log-prices, total cap ±10 points of probability, graded per claim.

What's new

What changed on the site, by date, in plain language. Changes to the prediction method itself are always noted here and never applied to picks already stored.

October 3, 2026

  • A new look across niftiIQ: calm greys, compact type, and two doors on every page (Maine high school and Pro). The NFL tabs are This week, Record, Rankings, Season and Analyst, with the rest under More.

September 30, 2026 (evening)

  • Every game page now ends with quick plain answers: when the game is, who we pick and at what chance, and after the final, who won and whether the pick was right. Search engines get the same answers.

September 30, 2026 (evening)

  • New Maine art across niftiIQ (a shoreline in every footer, fog on missing pages) and clearer headline numbers on the Record page.

September 30, 2026

  • The niftiIQ has one mascot now: a Maine lobster buoy. It sits in the footer of every page and reacts to our latest week on the Record page.

September 29, 2026 (evening)

  • A new look for the whole site, now for everyone: calm green and white, easier to read on a phone, with our two hosts in the footer.

September 29, 2026

  • A new look for the whole site: a clean sketchbook style, easier to read on a phone. Tap anywhere on a pick row to open that game.

September 25, 2026

  • Games to watch is now pickier: a game makes the list only when our pick is at least 52% to win, not just when we back a different team than the market. Near coin-flips are left out, so the list shows the games where we have a real opinion.

September 20, 2026

  • QB matchup now respects the injury report: when a listed starter is ruled out and a backup is going in, the game page shows the real starter instead of the stale name: e.g. "Bryce Young vs Cooper Rush (Tua Tagovailoa out)" rather than showing Tua as the starter. It reads the same injury report and in-week reporting the pick already uses, so the headline QB and the prediction no longer disagree.

September 20, 2026

  • Share cards for every game: when you post a game link (Facebook, iMessage, X), it now unfurls a clean preview of that matchup, the pick and its win probability in the team's color, the projected score, the market's final price, and, once it's played, the final and whether the pick hit, instead of a generic logo. Built fresh per game, so a shared link always shows the current state.
  • Cleaner "Why <team>" write-up: the reasons paragraph now sticks to the model's own factors. The news the analyst layer read is still shown in full in its own "What the news layer did" section, so nothing is lost: the summary just no longer repeats older, in-week headlines.

September 15, 2026

  • "Notable this week": every result is checked against every NFL game since 2009, and anything rare gets a mention, a team's most points in years, a first win over an opponent since a given season, a game in the top 1% for combined score or upset size, a week among the highest-scoring or best for road teams on record, plus comebacks, the longest plays and field goals, turnovers, weather, and individual games (a career high, a 400-yard passing day). Every ranked item shows the games above it, so "4th-highest since 2009" comes with the top three. Plain arithmetic on the play-by-play record; it appears on the week pages, on team pages, and inside the weekly review.

September 14, 2026

  • New page: "How accurate are NFL prediction models?": the plain answer (about two in three for the best public forecast), what a good number is, why one season proves little, why calibration matters more than a record, and five questions to ask of any prediction site. All the numbers come from our own sixteen-season backtest and live record, and update themselves.
  • Player rankings: a new Players page ranks quarterbacks, running backs, wide receivers and tight ends from play-by-play data: what each player produced per touch, adjusted for the defenses he faced, weighted toward recent games, and shrunk so a few big plays do not outrank a full season. The page says what we do not rank (linemen, defenders, blocking) and why.
  • Team pages: every team now has its own page with its rank history, the reason behind each move, its full schedule with our pick and result for each game, and our record on its games.
  • Rankings: a "What changed" panel lists every rank move since the last snapshot with the reason.
  • Write-ups are automatic: after a week's last game a review is published, and a preview appears a couple of days before the next week's first kickoff. They are written by our system from the data, labeled as such, and no human edits them.
  • Kalshi joins Polymarket, DraftKings (via ESPN) and the nflverse price on the Markets page.
  • Results arrive faster: finals are picked up within minutes of the end of a game, so the Record page updates the same evening.
  • Record page: records by kickoff slot (Thursday, Sunday early, Sunday late, Sunday night, Monday night), by home or away pick, by pick or other side, and a flat-unit thought experiment against the closing prices (clearly labeled as such, not advice).
  • Per-game pages: the pick, the reasons, every pre-kickoff update, every market's open and close, the news layer's claims, and the result.
  • News is read only before kickoff. Anything published during a game is excluded from the analyst layer and from its scorecards.
  • Share previews and canonical links so posts on Facebook and LinkedIn show a picture.
  • Every page now shows when predictions, results and market prices were last updated.

September 13, 2026 · public launch

  • niftiNFL went live at lab.niftiiq.com/sports/nfl, part of niftiIQ, with the shared site header, Sports and Local sections, and the NFL and NBA tabs.
  • Market margin and totals are graded like the winner, with the break-even mark shown.
  • Glossary, FAQ, this changelog page, privacy and terms, a contact form with a subscribe box.
  • Markets grid shows where each market agrees with our pick (green) and where it differs (red).
  • "What moved": every run's prediction is kept, so each week's page lists the games whose probability moved or whose pick flipped, before and after, with the biggest reason.
  • American English throughout.

September 12, 2026 · built

  • Version 0.1: the statistical model (team strength, opponent-adjusted efficiency, the named starting quarterback, rest and home field), a walk-forward backtest of 2010–2025 with a leak audit, and the This week / Record / Rankings / Season / Backtest pages.
  • Version 0.2: the availability layer (listed starters ruled Out or Doubtful). Backtest 65.0% straight-up against the market pick at kickoff's 66.5%.
  • Markets tracker: our probability and each public market's price from open to close.
  • Analyst layer: public news is read into typed, capped, source-linked claims; each claim is graded afterward, and declined claims are kept and graded too so the filters can be judged.
  • The "How this works" page and extra Sunday runs so inactives are counted before kickoff.