Per-block validation strip — rows 1 (fee market) + 4 (block/state size)

Date: 2026-09-16 (2025-2026 eras completed 2026-09-17) · Instrument: tools/research/perblock_validation.py Output: data/perblock_validation.json

Purpose

The SCCR historical reconstruction derives era-level fee and block-size legs from daily aggregates (fee_btc_per_day, avg-block-size, blocks_per_day_est). That derivation is grade B: correct in aggregate but unverified at the per-block level. This strip samples actual per-block measurements across 2010-2026 so the aggregate legs can be checked against direct observations, and so the SegWit-era size/weight divergence (a block/state-size boundary question, row 4) is quantified.

Sources (this environment)

Source	Endpoint	Gives	Status
Blockstream esplora	`/api/block-height/:h` + `/api/block/:hash`	`weight`, `size`, `tx_count`, `timestamp` per block	Reachable; ~1 req/s sustained, ~500 credits/hr (429 beyond)
blockchain.info	`/rawblock/:hash` top-level `fee` (sat)	block-level total fees	Reachable; ~5 MB/block, infrequent, not rate-limited in practice

Notes:

mempool.space /api/v1/block/height/:h (has feeTotal) times out from here for historical heights; esplora's block object has no fee field. blockchain.info rawblock carries the block-level fee, so fees were sampled there.
blockchair /bitcoin/blocks remains HTTP 430 (rate-limited) for the continuous per-block scan — unchanged since utxo-series-reachability.md.

Method

For each era 2010-2026, three target days (~Jan 1, ~Jun, ~Dec) are converted to a block height by binary search over the main chain (greatest height with timestamp <= target), via esplora. For each sampled height the block summary is cached; the mid-era sample additionally gets a rawblock fetch for the block-level fee. fee_btc (sample) is compared to the era's reconstruction reference fee_btc_per_day / blocks_per_day_est.

Rate-limit handling: per-request retries with 90 s backoff on 429; per-era checkpointing so a quota-limited run resumes exactly where it stopped.

Findings (table complete through 2026)

Row 4 — size/weight: early eras (2010-2013) sample 0-32 kWU blocks; the chain runs near-full 3993 kWU from ~2018 onward. Mean sample size 0.75 MB (2017) → 1.6-2.0 MB (2023-26) against a weight ceiling of 4 MWU — the SegWit divergence is visible: post-2017 weight ≈ 4× size, pre-SegWit weight ≈ 4× size holds except deflationary 2015-16 blocks. Ordinals/Runes-era tx counts confirm congestion: 2023 sampled 2,924 txs/block, 2024 4,246 txs/block (vs ~1,400-1,700 in 2021-22), and the recent eras stay near-full — 2025 1.95 MB / 2,373 txs, 2026 1.68 MB / 5,522 txs (the highest sampled tx density).
Row 1 — fee leg: single-block fee samples are highly noisy vs an era annual mean, and the dispersion is itself the finding: 2017 spike 2.36× (4.194 vs 1.774 BTC/block), 2018 0.06× (0.031 vs 0.484), 2022 0.02× (0.002 vs 0.102). Per-block fees cluster around fee events, not the annual mean — the aggregation hides a heavy-tailed per-block distribution. The daily-aggregate leg (B) is therefore the right grade for era means; per-block reconstruction (B→A) is genuinely a different object and stays blocked (see verdict). The cooling fee market is visible at block level in the tail: 2025 lands near parity (1.05×, 0.035 vs 0.033) while 2026 falls to 0.31× (0.006 vs 0.019).
Grades: row 1 historical stays B, row 4 historical stays B — each now carries a per-block validation strip (documented sampling bound) instead of an unverified daily-aggregate derivation. Not A: a 3-block-per-era sample is a check, not a reconstruction.

Reachability verdict

Full per-block reconstruction (B→A for rows 1, 4) is DOCUMENTED as blocked on a bulk continuous source: blockchair 430, mempool block-endpoint timeout, esplora no-fee. Two real completion paths: a synced local node (same R5 gate), or blockchair from a rate-limit-tolerant host.

Data source: data/perblock_validation.json — deterministic, frozen-input instrument; re-runnable offline.

← All research · ← Back to Learn