The Bitcoin Sahi Research Archive

The founding question: does Bitcoin's fee market pay for the storage it permanently creates? We built a live data platform with a Bitcoin Core node and five research agents to find out.

A one-sentence frame for the whole program: Bitcoin's fee market prices competition for the next block, but not the long-term resource costs of permanently recorded data. Here's the question, stated plainly.

The story of internal validation: a duplicated time-horizon term inflated storage cost by 10×. We corrected it, regenerated every value, and confirmed the conclusion survived. Honesty is the methodology.

From cypherpunk origins to Ordinals — a block-space-economics reading of Bitcoin's history, and why the SegWit witness discount became the pricing structure inscriptions use today.

We cleared the blackboard and read the whitepaper as a freshly discovered manuscript. The master pattern: every mechanism makes correct action cheap and incorrect action expensive.

A reviewer asked: "a nation state could flood your computer with garbage." We priced the attack. Filling every block costs ~$65M/yr and imposes ~$197M of storage cost — a 3× leverage ratio. Dust is the surprise: its real threat is RAM, not storage.

Verified verbatim from the Satoshi Nakamoto Institute: "storage should not be a problem," the node-equilibrium prediction, and the block-size cap as a "circuit breaker." Primary sources, quoted exactly.

Validation is the one cost every node pays that can't be pruned. Our order-of-magnitude survey: ~$0.5–5/yr per node, with a falsifiable claim — under $100/yr, central estimate $1–2. The four-resource model gets its third leg.

The witness discount made witness data 4× cheaper per byte in block weight — a 2017 malleability fix that became 2023's inscription subsidy. Whether it prices permanence is the research question.

Hashrate concentration isn't protocol control. UASF/URSF are coordination mechanisms, not proof of who holds power. BIP-110's mandatory-signaling window is a natural experiment in the governance boundary — and we're recording it hourly.

The UTXO set is the index every node carries. We wired `utxo_size_inc` end-to-end and measured it live: ~29.9 KB of net permanent state per block, 165 backfilled heights. The fourth resource leg is now data, not assumption.

Pruned nodes still download and verify every block. Storage is avoidable; bandwidth and CPU are not. The pruning note answers the "just prune" objection: the unavoidable burden is real, though small today.