Evidence Matrix

BSAHI — every headline number, classified by how it was actually obtained. Produced: 2026-09-17 · Checked by tools/research/integrity_audit.py

The three layers

Layer	Definition	What it means for a reader
Observed	Read from a node, or from a frozen primary capture with a recorded date	Cite as measurement. Grade A/B.
Reconstructed	Derived by a stated procedure from anchors (interpolation, era means, attribution)	Cite as procedure, not as a measurement. Grade C/D.
Modelled	Assumptions dominate; the output is a scenario, not an observation	Cite as a scenario. Grade C.

Nothing in this project may be presented as Observed if it is Reconstructed or Modelled. The matrix is the enforcement: if a number appears anywhere without a row here, that is a defect.

Matrix

#	Metric	Value / reading (as published)	Layer	Source	Date	Grade
1	SCCR (live)	0.2789 · 139 blocks · 99.28% below 1× · N=32K · T=10yr · spec 2.1.0	Observed	`data/sccr.json` ← mempool.space 24h fee endpoint (GH Actions)	2026-09-16	B
2	SCCR daily history	36 daily readings	Observed	`data/sccr_history.json`	2026-08-02 → 2026-09-16	B
3	SCCR historical (era reconstruction)	era-level SCCR, 2010-2026	Reconstructed	`data/sccr_historical_series.json` (daily aggregates → era legs)	frozen 2026-09	B/C
4	Chain size (bytes to verify)	789.45 GB (2026)	Observed	cumulative frozen `avg-block-size` (unit MB)	2026	B
5	UTXO / chain state	0.5 GB (2013) → 11 GB (2025/26), era table	Reconstructed	anchors + interpolation (no continuous source)	—	D
6	UTXO / chain state (observed)	h671,462 · 72,234,156 UTXOs · 18,633,940 BTC · UTXO DB 4.39 GB	Observed	`gettxoutsetinfo` from a real Bitcoin Core node (`utxo_state_measure.py`)	captured 2026-09-17	A
7	Mining difficulty	127,450,789,715,844	Observed	`data/difficulty_series.json` (frozen primary)	2026-09-13	A
8	Mempool congestion	1,498 daily points, 2016-2026	Observed	`data/mempool_congestion_series.json` (frozen primary)	2026	B*
9	Per-block samples	3 blocks/era (2026: 2)	Observed (spot)	`data/perblock_validation.json` (esplora + blockchain.info)	2026-09-16	B (spot-check, not a survey)
10	Network hashrate	956 TH/s·10⁶ (≈956 EH/s)	Observed	frozen `hash-rate` series (unit TH/s)	2026	B
11	Miners' revenue	41.4M USD/day	Observed	frozen `miners-revenue` series (unit USD)	2026	B
12	Production cost ratio	reported across $0.03–$0.15/kWh	Modelled (observed legs + assumed price/efficiency)	`data/production_cost_ratio.json`	2026-09-17	C
13	Electricity price	$0.05/kWh (scenario $0.03–$0.15)	Modelled (assumption)	documented assumption	—	C
14	ASIC efficiency (network avg)	25 J/TH (2026), era table	Modelled (estimate)	documented table	—	C
15	Node census (N)	≥32,000 (lower bound)	Observed	`getnodeaddresses` census (`data/node_census*.json`)	2026-08-02	B
16	Pool attribution	7/7 pools coherent ≤1.5σ; network-total diff 2.34%	Observed	`data/pool_attribution_validation.json`	2026-09-15	C/B
17	Mining concentration	HHI/Gini/N_eff per window	Observed	`data/mining_concentration.json`	2026-09-15	B
18	BIP-110 signaling	lock-in h963,648 (2026-08-23 00:48:47 UTC); snapshot h966,270	Observed	`data/bip110.json` (GH Actions) + blockstream.info	2026-09-09	A
19	Fee history / block interval	95-point intraday series	Observed	`data/fee_history.json`, `data/block_interval.json`	2026-09-17	B
20	Verification Cost Index	trend (headline): sync time flat, value-relative cost ~17× cheaper	Modelled	`data/verify_cost_index.json`	2026-09-17	C
21	VCI chain leg	total bytes to verify	Observed	frozen `avg-block-size`	2026	B
22	VCI state leg	reconstructed UTXO table	Reconstructed	(falls back to row 5)	—	D
23	VCI throughput	assumed commodity archive-sync	Modelled	documented table	—	C
24	Node operator labour	$25/h, labour share	Modelled	documented assumption	—	D
25	Cost-to-flood leverage	L ≈ 3.0× (dust 0.9×)	Modelled	`research/cost-to-flood.md`	2026-08-10	C
26	Bandwidth bound	52.6 GB/yr	Modelled	`data/bandwidth_bound.json`	2026-08-04	C
27	Pruning externality	download+verify unavoidable	Modelled	`research/pruning_externality_analysis.md`	2026-08-10	C

Grades

A — direct primary measurement (node RPC, on-chain fact), dated.
B — frozen primary capture with a recorded unit and date.
C — model/estimate with documented assumptions; direction is the claim.
D — reconstruction by interpolation with no continuous source.
* — B* = primary series with a documented gap (e.g. pre-2016 congestion).

Standing checks (run with the matrix)

tools/research/integrity_audit.py re-checks, on every run:

Heights/dates monotonic — every (height, timestamp) pair increases together.
Anchors — a height's chain date matches the known anchor; the capture time is recorded separately (they differ by years while the node is in IBD).
Units — derived unit fields agree with raw ones (hashrate_ehs == hashrate_ths/1e6; power TH/s × J/TH = W; GB is decimal 1e9).
Provenance — each dataset carries schema + generated_at + a source.
Layers — each dataset is classified Observed / Reconstructed / Modelled.

Known open items (surfaced by the audit, not hidden)

UTXO state is Observed for one height only; the 2013-2026 series remains Reconstructed (grade D) until the node reaches the tip and the series accrues.
Several auxiliary datasets still lack an explicit layer field (ops/health and a few research files); the matrix row is the interim classification.
Electricity price, ASIC efficiency, wage and throughput remain assumptions.

How to use this page

Before publishing any number, find its row. If the layer is Observed, cite it with its date. If it is Reconstructed or Modelled, cite the procedure and the grade — never the level as if it were measured.

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

← All research · ← Back to Learn