cheeksmagunda
@cheeksmagunda
Public Skills
wnba-oracle-strategy-gap
by cheeksmagunda
Run the WNBA Oracle strategy gap analysis. Compares the model's picks to realized outcomes, identifies systematic biases (archetype hit rates, starter signal accuracy, boost calibration), and surfaces knob candidates for the next shadow cycle. This is the primary tool for iterative model improvement. Use when: operator wants to know where the model is losing EV, which player types are systematically over- or under-weighted, or whether the current starter/boost configuration is tracking reality. Authoritative store: Postgres (wnba_game_logs, slate_labels, frozen_lineups, model_shadow_runs, player_pool).
wnba-oracle-knob-shadow
by cheeksmagunda
Activate, review, and graduate a knob-shadow challenger in WNBA Oracle. A knob shadow runs the incumbent model's head predictions through a hypothetical picker-knob overlay on the same slate data, logs the ranking divergence (rbo_at_5, ndcg_at_5), and lets dayclose backfill the realized value delta once final scores arrive. Use this workflow to test a new default before committing it to production. Use when: operator wants to A/B test a picker knob change without deploying new code or changing the production freeze. Do not use: to change the actual production freeze result during a live contest.
wnba-oracle-slate-review
by cheeksmagunda
Pre-freeze slate review for WNBA Oracle. Reads player pool, archetype labels, stat-leverage scores, and streak quality from Postgres. Surfaces the lineup the model froze (or will freeze) with interpretable player context so the operator can confirm the picks before the contest lock. Use when: operator wants to review today's picks, understand why a player was slotted where they were, or compare production lineup to a challenger. Do not use: to place bets, to fetch live scores from ESPN, or to bypass the freeze rules. The authoritative store is Postgres; do not substitute ESPN public API data for production picks.
wnba-oracle-corpus-status
by cheeksmagunda
Audit the WNBA Oracle training corpus and model artifact. Verifies that the backups branch is current, the live model artifact SHA matches STATUS.md, and the game-log corpus covers the expected date range. Use before a training run, before promoting a new model artifact, or after a production incident to verify corpus integrity. Authoritative stores: Postgres (wnba_game_logs), GitHub backups branch, STATUS.md (artifact SHA pin).
wnba-oracle-ops-runbook
by cheeksmagunda
Operational runbook for WNBA Oracle. Step-by-step procedures for common game-day operations, incident response, service health checks, and controlled configuration changes. Use this as the first reference when something is wrong or when making a production change. This skill cites the authoritative store for every check.