near-thesis: NEAR on the evidence

A data-backed, re-runnable long-term thesis on NEAR, judged on evidence. Every number is rebuilt from keyless public data, mostly NEAR's own chain: archival RPC block headers (supply, burn, staking), the FastNear indexer (per-signer histories, wallet flows) and DIP-4 settlement events on intents.near. At the 2026-10-29 snapshot, organic human-like signers sent 2.17% of transactions in 2026-Q3 (90% interval 0.86% to 3.70%) and NearBlocks' active accounts miss real signers by 1.9x over the last six months, because relayers hide them. NEAR Intents swaps are real, but counted once they are 46.04% of DefiLlama's volume (92.08% two-sided); captured revenue rebuilt from the revenue wallets matches at 98.68%, and 3.11M NEAR has been bought back and held. Holders are still diluted: the float grows 2.12% a year, and offsetting it needs 8.1x today's Intents revenue (4.0x at half today's price, 16.6x against the last 365 days of revenue; 5.1x under HSP-029). treasury.near sent 30.17M NEAR in April-May 2026; 20.95M NEAR reached staking pools within five hops. The verdict is conditional, not bullish. The repo has a static dashboard (nine charts, each saying what it shows and what it does not, ending with a path-to-yes chart of what would confirm the bull case by late 2027), a thesis whose numbers are filled in from the data by the build, a falsifier scorecard, 7 dated predictions frozen with a sha256 in a git tag, and one command, make reproduce, that rebuilds everything offline and tests that it is byte-identical.