WHY
Decision logic
No recommendation yet.
2026 NFL SURVIVOR
Command Center
RECOMMENDATION
Three-entry portfolio
Sync the real schedule and odds, then run the optimizer. Nothing is prefilled.
FIELD RISK
Upset Kill Zones
No field data yet.
LIVE PERIOD
Games & calibrated probabilities
ALTERNATIVES
Portfolio rankings
| # | Portfolio | Any survives | Expected alive | Pool equity | Phase-end survival | Future preserved | Field leverage |
|---|
FOOTBALL INTELLIGENCE
Intelligence Center
Market odds are the baseline. Your selected LLM researches current football context and can only make a bounded, auditable adjustment.
ACTIVE PROVIDER
Not configured
Configure an LLM provider in Settings. The optimizer will remain market-only until then.
MAX AI ADJUSTMENT±5.0 ptsConfidence-weighted and ignored if stale after a material market move.
No football intelligence has been run for this period.
MULTI-MODEL CHECK
Not runWar Room
Runs every configured provider independently on the same slate and exposes disagreement. It is advisory and never silently averages itself into the optimizer.
Configure two or three provider keys, then run War Room when you want a second and third opinion.
METHOD
How the intelligence layer works
1. MarketNo-vig sportsbook consensus
→
2. ResearchInjuries, QB, weather, matchup, news, rest, line movement
→
3. LLMEvidence-backed risk assessment
→
4. CalibrationBounded probability adjustment
→
5. SurvivorInventory + field + Monte Carlo portfolio optimization
AUDIT THE MODEL
Team Lab
Click any team to see market inputs, AI intelligence, future value, field ownership and team-inventory provenance.
REAL PEOPLE ONLY
Your pool
The app creates only the three coalition entries. Every opponent below exists because you added them.
Your coalition
Opponents
PERIOD ENTRY
Picks & history
Preseason inventory and regular-season inventory are completely separate.
Current period
Full pick history
| Entry | Phase | Week | Team | Result |
|---|
Configuration
Coalition names
Football intelligence engine
Vercel environment variables
Secrets are server-side only. Survivor TAZ reads them from the deployment environment and never creates or edits a .env file.
Set or rotate values in Vercel → Project Settings → Environment Variables, then redeploy. Provider and model selection below remains app-configurable; only credentials live in Vercel.
Sports data controls
Credential label: ODDS_API_KEY. The browser never receives the secret value.
Supabase
On Vercel, Supabase is the durable system of record and SQLite is only an ephemeral serverless cache. Connection credentials are injected by Vercel and are never editable from the browser.
Vercel labels:
SUPABASE_TAZ_URL, SUPABASE_TAZ_SECRET_KEY, SURVIVOR_TAZ_WORKSPACE_ID, and optionally SUPABASE_SYNC_ENABLED=true. Run the included schema once in Supabase SQL Editor.Cloud history
Refresh to load the append-only Supabase event history.
What actually decides a pick?
Market baseline: sportsbook moneylines from The Odds API. Each book is converted to implied probability, de-vigged, then averaged.
LLM role: researches current football context and proposes a bounded adjustment to that baseline. It cannot invent a win probability when market odds are missing.
Future matchup strength: direct sportsbook odds where available, otherwise consensus Super Bowl futures-derived team strength.
Future Value: deterministic calculation over the strongest remaining regular-season spots.
Field ownership: your actual opponents only. Submitted picks are exact; projections require real prior history.
Upset Kill Zone: field ownership × calibrated upset probability.
Final choice: portfolio optimization + Monte Carlo. The LLM informs game risk; it does not choose your survivor portfolio.
Clean reset
Deletes all picks, opponents and synced data, then returns to exactly three blank coalition entries. Vercel environment credentials are untouched.