Architecture
The Front Office is not one AI guessing at your roster. It is a layered decision system that turns league data into recommendations you can understand.
Data Sources
Platform Database
Prediction Models
Intelligence Engines
Specialist Agents
Recommendations
Manager Decisions
Decision Capture
Data Flow
Flow thickness reflects relative data volume at each layer.
Data flows through layers until it becomes a recommendation. Each layer adds context, analysis or action.
The raw information the platform needs to understand your league.
NBA game logs, player statistics, schedules, injury reports, league settings, roster configurations, waiver wire availability, opponent teams, trades and transactions.
Good decisions start with complete context. Incomplete data produces incomplete recommendations.
The database organises raw information into a consistent view of your league.
Your league settings, roster rules, player eligibility, lineup constraints and team context are all stored here. This is what makes recommendations specific to your league, not a generic one.
Your league settings, roster rules and player eligibility all affect what a good decision looks like.
Models estimate what is likely to happen next.
The platform projects player production, games played, minutes, opportunity, availability, durability and category output based on rolling performance, schedule, team context and situational factors.
Fantasy decisions are about future value, not just last week's box score.
Engines are specialist analysts. Each engine answers one specific question.
There are eight engines: player projections, player identity, roster construction, league management, schedule context, lineup constraints, championship odds and decision capture. Each engine consumes data and model outputs to answer its question.
No single metric can tell you what to do. Breaking the problem into parts produces better answers.
Agents are digital workers. They consume engine outputs and execute a specific job on your behalf.
The Trade Agent evaluates trade proposals. The Waiver Agent finds fit-specific pickups. The Matchup Agent builds your weekly game plan. The Opponent Scout analyses the team across from you. The GM War Room assembles multiple perspectives on a difficult decision.
Agents turn analysis into action. They do the work of gathering, interpreting and prioritising information so you do not have to.
A recommendation is the final output. It should tell you what to do, why it matters, what impact it may have and what trade-offs exist.
Recommendations include the action, the reasoning behind it, the expected category or odds impact, the confidence level and any significant trade-offs. The goal is a decision you can make and defend.
The goal is not more data. The goal is a better decision.
The platform records recommendations and tracks whether they were followed.
DCE1 (Decision Capture Engine) logs every recommendation generated, whether it was viewed, whether the manager followed it and what happened afterwards. Over time this creates a measurable record of decision quality.
The Front Office is accountable. It does not just recommend moves — it measures whether they worked.
Each engine answers one part of the decision. Together they cover the full picture.
What it does
Projects player production, games played, minutes, opportunity, durability and role using rolling performance, schedule and team context.
Why you care
Helps you avoid overreacting to short-term hot streaks or cold stretches. The signal matters more than the last three games.
Agents are digital workers with specific jobs. They use the engines to decide what to recommend.
Strategy Agent
Define what your team should become.
Helps you decide which categories to attack, protect or ignore based on your roster build and league position.
Trade Agent
Find realistic trades that improve your team without creating long-term risk.
Looks beyond who wins the trade and considers roster fit, league context and whether the other manager has a reason to say yes.
Waiver Agent
Find the best available move for your team.
Ranks waiver options by roster fit, category need, schedule window, long-term value and defensive value against rivals.
Matchup Agent
Build your weekly game plan.
Identifies lock categories, battlegrounds, conceded categories and the actions most likely to improve your weekly result.
Opponent Scout
Understand the manager across from you.
Identifies their strengths, weaknesses, likely win path and roster vulnerabilities before the matchup begins.
GM War Room
Bring multiple specialist viewpoints into difficult decisions.
When the answer is not obvious, the room gives you a recommendation with reasoning, not a guess.
A single question flows through the full system before becoming a recommendation.
Question
Should I trade Myles Turner for Nikola Vučević?
How the system processes it
Recommendation
Decline.
Reasoning
The trade improves points production but weakens blocks and reduces your playoff matchup advantage. Turner's BLK contribution is categorically scarce in your league and would be difficult to replace via waivers.
Continuous Improvement
Before a week begins, The Front Office records what it believes will happen. When the week ends, that forecast is compared with reality and added to its evidence history. Over time, repeated patterns can reveal where the system may be improved.
What did TFO believe?
The pre-week forecast is frozen before the outcome is known, creating a permanent record of what the system actually predicted.
Is the miss repeating?
Forecasts are compared with reality across players, categories and weeks. One unusual result is evidence — not a reason to immediately change the model.
Does the improvement hold up?
Candidate improvements must prove themselves against unseen historical weeks rather than simply explaining the data that revealed the problem.
Has it earned its place?
Only improvements that pass validation can progress into the forecasting system and influence future decisions.
A single bad forecast should not rewrite the system. The Front Office separates evidence, pattern detection, validation and deployment so that changes must demonstrate they improve forecasting before they can affect future recommendations.
The Front Office is designed to show its work. Every recommendation is traceable back to the data, engines and assumptions that produced it — and every forecast is recorded and measured against reality. You do not have to blindly trust the output: you can see the reasoning behind it, and how well its forecasts have held up.