THE COMMISSIONER

League Brain Plan

Fast chat, deep research, durable league knowledge

What the brain should become good at

Not a generic football encyclopedia. It should build knowledge that is specifically useful to this group and difficult to reconstruct from a single API call or web search.

• Explain who has historically drafted well, traded well, wasted FAAB, benefited from schedule luck, or collapsed in the playoffs.
• Connect current NFL role changes to this league’s exact scoring, rosters, keeper rules, and waiver pool.
• Track LSU’s season as an evolving story: scheme, depth chart, recruiting, transfers, player development, opponents, and NFL prospects.
• Remember established rivalries, predictions, recurring jokes, public corrections, and shared league moments without turning people into personality dossiers.
• Answer historical questions with evidence rather than reconstructing the league from scratch every time.

Model routing

Exact model names remain configurable so the routing can follow real latency and cost benchmarks. The important rule is that strong models are reserved for synthesis and ambiguity, not routine chat.

Research program: this fantasy league

Canonical history

• Season-by-season standings, playoffs, champions, points, records, draft results, keepers, trades, waivers, and rule changes.
• Head-to-head history, playoff meetings, rivalry records, scoring records, biggest wins, closest losses, streaks, and collapses.
• A dated record book so future answers can distinguish regular-season finish from final playoff placement.

Derived manager and team analysis

• All-play record and schedule luck compared with actual wins.
• Optimal-lineup efficiency, bench points left behind, positional strength, and roster construction.
• Draft value, repeated reaches, keeper value, waiver hit rate, FAAB efficiency, and trade outcomes over time.
• Observable manager tendencies with sample size, dates, and counterexamples.
• Predictions and takes preserved as they were made, so hindsight does not rewrite the evidence.

Research program: NFL

The NFL layer prioritizes the teams represented in this group, then fantasy-relevant players and the general league context needed to understand them.

Priority-team expertise

• Current rosters, depth charts, staffs, schemes, injuries, transactions, schedules, role competitions, cap decisions, and draft choices.
• Weekly snaps, routes, targets, carries, red-zone work, pass protection, defensive usage, and meaningful performance trends.
• Season arcs, franchise history, rivalries, major personnel decisions, and the context behind why fans care about an event.
• Fantasy implications under this league’s scoring, keeper rules, roster ownership, and actual waiver availability.
• A dated “what changed?” section for each team so knowledge compounds instead of becoming repeated weekly summaries.

General NFL coverage

• League-wide developments that affect the four priority teams, rostered fantasy players, major contenders, rules, coaching, or the NFL Draft.
• Durable player profiles covering role, strengths, weaknesses, usage trajectory, injury history, and fantasy relevance.
• Schedule strength, bye weeks, fantasy playoff matchups, and weather when it materially changes a decision.

NFL source hierarchy

1. Official NFL and team injury reports, rosters, schedules, transactions, and releases for primary facts.
2. Pro Football Reference and comparable reputable statistical sources for performance and historical data.
3. The Ringer NFL as a starting point for analysis, narratives, questions, and competing interpretations, not as sole authority for contested facts.
4. Wikipedia for historical orientation and discovery, with important claims checked against stronger sources.
5. Clearly labeled original analysis that connects the evidence to this group and fantasy league.

Research program: LSU football

• Current roster, depth chart, injuries, staff, scheme, recruiting classes, transfers, and role competitions.
• Opponent previews focused on matchup edges, personnel, likely game plans, and meaningful uncertainty.
• Postgame learning: what LSU attempted, what worked, what failed, what changed, and which players gained or lost roles.
• Player development, NFL Draft prospects, LSU alumni in the NFL, and fantasy relevance.
• Major program history that gives current events context rather than dumping trivia.
• Group predictions and opinions stored separately from verified LSU facts.

Research program: shared history and group dynamics

This includes a lightweight CRM for every group member. Its purpose is continuity and useful personalization, not sales software and not surveillance.

• Preferred names, authorized public handles, fantasy-team mappings, self-disclosed location, NFL and college fandom, explicitly stated preferences, public league moments, predictions, rivalries, and recurring bits.
• Objective league history: finishes, records, transactions, draft tendencies, head-to-head results, and notable public moments.
• Distinguish something said once from a repeated joke or established pattern.
• Separate self-reported facts, objective calculations, observations, jokes, and inferences.
• Record dates, sources, confidence, evidence count, counterevidence, volatility, and review dates.
• Let weak social observations decay when they are not reinforced.
• Never infer protected traits, mental health, finances, relationships, politics, private motives, or sensitive personal details.
• Never merge private DMs into group memory or convert playful insults into permanent personality labels.

Initial CRM profile: Gregory Hill

Recurring useful outputs

Questions this should uniquely answer

Who does Andy consistently reach for in drafts, and has it actually worked?

Which manager has been luckiest over the last three seasons after accounting for all-play record?

What is Greg’s record against Andy, including playoffs, and what are the established jokes around that rivalry?

Which waiver pickup created the most value per FAAB dollar?

What trade looked terrible at the time but aged well?

Which NFL usage changes matter for our exact scoring and the players currently available?

What changed in LSU’s offense over the last month, and is it personnel, opponent, or scheme?

Who was right about a preseason prediction based on what was knowable at the time?

Fast-path behavior

• Simple replies use recent context and at most one targeted brain lookup.
• Live league facts use Sleeper directly; no stale wiki detour.
• Deep requests get an immediate natural acknowledgement and a research subagent.
• Long work is published as a link rather than dumped into Telegram.
• No grep output, indexing commentary, or maintenance chatter appears unless a manager asks for diagnostics.

Quality controls

• Every durable claim stores its source, observation date, confidence, type, and volatility.
• Facts, deterministic calculations, opinions, jokes, predictions, and inferred patterns remain separate.
• Time-sensitive claims expire or force a live refresh.
• Unchanged inputs are hashed and skipped; routine runs touch only changed pages.
• Social patterns require repeated public evidence or explicit manager confirmation.
• Managers can correct, exclude, retire, or request review of a compiled claim.
• The constitution, member mapping, and private communications remain outside autonomous brain edits.

Success metrics

• Median simple-response latency under 2 seconds; 95th percentile under 5 seconds.
• Deep-work acknowledgement under 2 seconds.
• At least 95% accuracy on audited league facts and calculations.
• Zero stale live-state answers when an authoritative source is available.
• At least 90% of durable factual claims include source and date metadata.
• Fewer than 5% duplicate or contradictory entries after weekly audit.
• Zero private-DM leakage or unsupported sensitive profiling.
• Track “brain added value”: how often stored context materially improves an answer beyond recent chat and live APIs.

Rollout

1. Apply the league-brain skill and create the canonical schema.
2. Backfill verified Sleeper history and produce the initial record book.
3. Start nightly consolidation in observation mode with a visible audit log.
4. Launch NFL and LSU weekly change-tracking.
5. Benchmark the model routes for latency, cost, extraction accuracy, and synthesis quality.
6. Run a two-week manager audit using real questions, then tighten routing and confidence thresholds.