An independent research project measuring the economics of Bitcoin block space.
Bitcoin has a market price for block space, but no explicit market price for the long-lived resources a confirmed transaction commits the network to. Bitcoin Sahi measures one of those resources — replicated storage — and asks how much of its modeled cost transaction fees actually cover.
N and SCCR are dated measurements, not constants — see the changelog for every correction.
Bitcoin Resource Accounting is the research program: quantifying the long-lived resources Bitcoin consumes (replicated storage, UTXO state, validation, relay, bandwidth) and how much of each cost the fee market internalizes. The first metric is the Storage Cost Coverage Ratio (SCCR):
A ratio below 1.0 means the fees paid do not cover the modeled cost of storing that data across the network. SCCR is fee-market-driven and time-varying: it is published as a dated measurement, not a fixed constant.
The measurement endpoints are open and stable:
/data/sccr.json — latest SCCR reading/data/sccr_history.json — daily SCCR series/data/perblock_validation.json — per-block validation samples/data/difficulty_series.json — difficulty, 2009–present/data/mempool_congestion_series.json — mempool congestion, 2016–presentPost archive coming here as it is added.
We audited all 47 live pages for stale numbers. The reproduction protocol promised 0.2406 when the kit produces 0.2896 — and one fix we had already shipped had been silently reverted.
accuracyWe tested the obvious shortcut — geolocating mining pools' stratum endpoints. 10 of 14 addresses are CDN-fronted, so the country you get belongs to the edge, not the pool.
miningValidating every block in order on one machine: 23 ms/block in 2009 rising to 928 ms/block in 2018 — a 40× rise — while per-transaction cost stayed near 130 µs.
validation costNon-listening nodes are invisible to every crawler. We borrowed a public IPv6 and ran a PROXY-protocol relay — then found the replies leaving by the wrong interface.
private nodesThe model ran on N = 32,000 — a truncated address sample. The measured reachable count is 26,586, so every SCCR value rises by 1.2036×.
correctionResearch and correction queries: beta@bitcoinsahi.com. Also on LinkedIn. Falsification attempts are particularly welcome.
Research, data, and code are published under a Creative Commons Attribution 4.0 International (CC BY 4.0) license. Reuse with attribution.