Five real decks from the archive, page by page, with investigator's notes. Gold-tagged notes = something worth stealing.
The champion's slide 18–19 turned into a machine: pick the buy/sell threshold, watch where you want to be (train Sharpe) versus where you actually land (test Sharpe). Real numbers from our Sprint A replication.
The only moving footage in the archive: Team Datawave walks their 2025 Power BI cash-flow dashboard — the tab structure, slicers and forecast overlays that placed in the AstraZeneca year.
Rendered notebooks — code and outputs inline, no setup needed.
The actual analysis behind the winning deck — cleaning, indicators, XGBoost, simulation.
Open notebook →Top-10 team's PriceCatcher workflow: parquet loading, merging lookups, heatmaps, Prophet.
Open notebook →Our runnable champion replication — every step commented and mapped to the deck pages.
Open source → · raw .pyThe 2024 dataset's 43-column dictionary (gold = columns in the champion's winning pipeline). Filter, then go stare at the real data.
Raw materials: 2024 dataset (CSV) · 2025 AstraZeneca dataset (XLSX) · Datathon 2026 handbook (PDF) · all archived repos