← BACK TO SAM Uncle Sam hat
Sports Analysis Machine · SAM XAI 2.1

How a prediction actually gets made.

Every pick starts the same way: a real question, real live stats, and a real simulation — never a betting line, and never a matchup SAM can't actually verify. Here's exactly what happens behind the scenes from the moment you sign in and ask.

Sign-in gate checks your session
Signed cookie, verified against your approved account — no valid session, no access
↓
You ask a question
e.g. "Washington vs Michigan" · SAM 1.5.8 or SAM XAI 2.1, whichever you picked
↓
The model identifies the matchup
Pulls the two sides and the sport from your message
↓
Checks Airtable for this exact matchup
Logged today? You get that identical result, instantly. Logged on an earlier day? SAM reruns it fresh with today's stats instead of handing you stale numbers.
↓
Looks up the real schedule
ESPN scoreboard (or the league's own source) confirms this is an actual, currently scheduled matchup — exact date, time, venue, never guessed
↓
Can't verify it's real? SAM stops here.
No schedule match, no real player/roster data, no live stats — SAM tells you plainly it can't process the matchup instead of inventing one. No fake or hypothetical matchups ever get simulated or logged.
↓
Pulling the stats — the path depends on the sport
Team sports
NFL / NBA / WNBA / MLB / NHL / Soccer / College
Real current rosters pulled automatically · home-field/court/ice edge factored in · injury counts scale the numbers · MLB probable starting pitcher, and NFL/CFB/NBA/WNBA/CBB/NHL's real key player, folded directly into the math
Individual / field sports
UFC · Bowling · Darts · Tennis · WTA · ITF · NASCAR
Verified against real fighter/player data first (ESPN tale-of-the-tape, PBA, PDC tour cards, ESPN rankings) — unverified names get declined, never estimated. Horse Racing is disabled: no data source exists to verify a race is real.
↓
Pulls your track record
Up to 1,000 of your own graded picks, by sport — used to calibrate confidence, never to invent a number
↓
Runs the real simulation
25,000-trial Monte Carlo · bivariate Poisson or normal distribution · home-field edge, injuries, starting pitcher/key player quality all shift the actual math, not just the writeup · conservatism shrinkage pulls raw win% back toward 50/50 · no odds, ever
↓
Builds the reply
Key player breakdown (only players confirmed on the real current roster), the simulation's win %, plain-English analysis — and an override if the track record has a real reason to disagree
↓
Logged to Airtable, tied to your account
Matchup, win %, model version, and your Bettor record — starts as Pending. A lock keyed to the exact matchup and date stops the same pick from ever getting double-logged.
↓
Sent to you
Your answer
Returned right away — no extra wait
If you want a second opinion
Click "get a second opinion" on any answer
On-demand only, not automatic — the other model runs the same question through its own simulation, logged separately, still tied to your account
↓
Auto-graded once the game ends
A recurring check against ESPN's scoreboard · writes Win, Loss, or Push — nothing stays Pending forever
↓
Feeds back into the next pick
Your graded track record is what the next simulation calibrates against — the loop closes

Nothing here ever looks at a spread or a betting line, and nothing here ever runs on a matchup that couldn't be verified as real — every number on the board came out of a real simulation run against real stats for a real, currently scheduled game.

account / session
data / model step
logged pick
declined — unverified
your answer
second opinion

Proprietary Technology built by GRIMALDI.TV