inferred from GitHub
headroom mp
Grid capacity for AI data centres in Madhya Pradesh, built from the state's own published data. Where can a 50-100 MW AI data centre connect, and will it have power at 2 a.m.?
Core trick
Parses published grid documents into substation capacity records and marks unknown fields explicitly instead of estimating them.
Evidence
README matched "Claude Build Day": Built for Claude Build Day, September 2026.
Tags
README excerpt
# Headroom MP **Where can a 50–100 MW AI data centre connect in Madhya Pradesh, and will it have power at 2 a.m.?** A grid situational-awareness tool built from Madhya Pradesh's own published documents. It shows where the state's EHV network is under strain, where it has capacity, where an AI data centre could land — and, deliberately, **where it does not know**. Built for Claude Build Day, September 2026. --- ## The problem Madhya Pradesh wants AI data centres. A 1 GW MoU is signed and one purpose-built 80 MW site exists in Indore. An AI data centre is a **flat 24×7 load**. The MP grid is shaped around two things that are neither flat nor 24×7: winter irrigation, which drove an all-time peak of **19,902 MW in January 2026**, and midday solar, which produces nothing after sunset. By July 2026 peak demand was **13,818 MW** — a seasonal swing of roughly 6 GW that a flat load could sit inside. Night supply today is thermal. The state's first grid-scale battery tenders (625 MW / 1,250 MWh at Bina, Sendhwa and Birsinghpur) only closed bids in July 2026. So the question is not "where is there capacity" but **"where is there capacity at night"** — and answering it currently means reading transmission PDFs, regulator orders and tender notices by hand. ## What this actually is Three views over one record: | | | |---|---| | **Where this comes from** | Claude reads the published PDFs and turns messy multi-row tables into structured records, each carrying its source document a