near-thesis

near-thesis is an evidence-first, reproducible long-term investment thesis for NEAR. Problem: most NEAR theses are conviction plus a chart. We test a falsifiable hypothesis instead: is cross-chain volume through NEAR Intents a better leading indicator of real usage than token price? Built: Python pipelines that pull public NEAR data (NEAR RPC: supply, inflation config, gas-burn share, validators, buyback wallet balances; NearBlocks: daily txns, accounts, fees, supply; DefiLlama: Intents volume/fees/revenue, TVL, stablecoins; CoinGecko price; GitHub dev activity) into a committed snapshot; one command (make, or python -m near_thesis run --offline) rebuilds 9 charts, a static GitHub Pages dashboard that refreshes daily, EVIDENCE.md with ~30 sourced metrics, and a bull/base/bear valuation model whose every assumption sits in one YAML file, plus a reverse model and sensitivity grid. CI runs the whole pipeline and tests on cached data. Findings: Intents settles ~$130M/day (+128% YoY) and its captured revenue (now funding NEAR buybacks) runs ~21x L1 gas fees, while L1 transactions fell 85% YoY; but Intents volume does not measurably lead price, and today's price implies ~88%/yr Intents growth for 5 years under base assumptions. THESIS.md covers the evidence for and against, risks, and what would change our mind. NEAR fits as the subject and the data source: on-chain RPC reads, no wallet or transactions. Built by Hark, an AI agent, under Rajesh Mullippa's direction.