DeepCrease and ICR surfaced
A new India‑made analytics stack—DeepCrease for phase/shot contexts and an ICR rating that claims to explain ~70% of venue/role win‑variance—was demoed publicly as an example of homegrown cricket analytics maturing. These tools point to faster, match‑phase aware scouting and tactical reporting for IPL/ISL-style analysis pipelines. (x.com)
DeepCrease was published as a Streamlit/Hugging Face space under the handle omkarthecricketest and accepts ball‑by‑ball CSVs (including bowler names, dismissal types and field coordinates) to produce interactive Plotly visualizations for phase and shot contexts. (huggingface.co) The Integrated Contextual Rating (ICR) and a bowling variant BICR are documented in a supporting GitHub repository titled “MIT‑SSAC‑2026‑Unseen‑Game,” which includes ICR_project_code.py and computed ICR/BICR results for IPL 2025 players. (github.com) Public commentary from the author states the ICR explains “over 70%” of variance in team win percentages versus the next best metric at under 35%, a claim the project notes as central to its value proposition for role/venue adjustments. (substack.com) The demo plus repository output files (ICR/BICR CSVs and visual widgets) are positioned as tools to shorten the timeline from raw ball‑by‑ball ingestion to match‑phase aware scouting reports and auction‑grade player lists used in IPL‑style decision pipelines. (github.com) Author posts tie ICR outputs to concrete roster/auction narratives, citing an ICR percentile example for Dewald Brevis at 82.09 and referencing Royal Challengers Bengaluru’s 2025 squad decisions amid the Jeddah mega‑auction (November 24–25, 2024) as an empirical validation point. (t.co) The project’s public artifacts (ICR_project_code.py and IPL 2025 CSV) are ready to clone for student capstone work, and canonical open datasets such as the IPL complete dataset on Kaggle (updated to 2024) are listed as suitable inputs for reproducing DeepCrease pipelines and deploying a Streamlit/Hugging Face demo. (github.com)
