Methodology Explained and Verified

Systemic Risk Channels in Digital Finance — Literature Search and Composite Scoring Pipeline

1 Overview

This document explains and verifies the pipeline used to rank 14 candidate systemic risk channels in digital finance. The goal is to select channels for an academic taxonomy paper based on three measurable criteria: literature prominence, scholarly impact, and crisis history.

What is this pipeline?

A systematic literature search combined with a composite scoring methodology. The pipeline ingests Boolean search queries, retrieves academic papers from the OpenAlex API, links channels to historical crisis events, and produces a weighted composite score for each channel.

Purpose

To objectively determine which of 14 candidate channels deserve inclusion in a scholarly taxonomy of systemic risk in digital finance. The ranking reflects the intersection of (a) how much the academic literature covers a channel, (b) how influential that literature is by citation count, and (c) how frequently real-world crisis events have activated the channel.

14 Channels
search_queries.json

→

OpenAlex API
2009–2026, ≥5 cites

→

Scoring
3 dimensions

→

Composite Ranking
channel_rankings.json

2 Step 1 — Define Channels

Fourteen candidate systemic risk channels are defined in search_queries.json. Each channel has a unique identifier, a human-readable name, 3–4 Boolean search queries designed to capture its literature footprint, and optional theory anchors (foundational references from traditional finance).

#	Channel ID	Display Name	Queries	Theory Anchors
1	network_contagion	Network Contagion	4	Allen & Gale (2000); Acemoglu et al. (2015)
2	liquidity_spirals	Liquidity Spirals	4	Brunnermeier & Pedersen (2009)
3	stablecoin_runs	Stablecoin Runs	4	Diamond & Dybvig (1983)
4	oracle_manipulation	Oracle Manipulation	4	—
5	composability_risk	Composability Risk	4	—
6	liquidation_cascades	Liquidation Cascades	4	Brunnermeier (2009)
7	counterparty_concentration	Counterparty Concentration	4	Upper (2011)
8	regulatory_contagion	Regulatory Contagion	3	—
9	gateway_risk	Gateway Risk (Fiat-Crypto Bridging)	4	—
10	governance_failure	Governance Failure	4	—
11	information_asymmetry	Information Asymmetry and Opacity	3	Akerlof (1970)
12	rwa_transmission	Real-World Asset Transmission	3	—
13	bridge_vulnerability	Cross-Chain Bridge Vulnerability	3	—
14	validator_concentration	Validator/Miner Concentration	3	—

Example: network_contagion queries

systemic risk contagion cryptocurrency network
DeFi interconnectedness systemic risk
crypto exchange network contagion
financial network topology cryptocurrency

Theory anchors: Allen and Gale (2000), Financial Contagion, Journal of Political Economy; Acemoglu, Ozdaglar, and Tahbaz-Salehi (2015), Systemic Risk and Stability in Financial Networks, American Economic Review.

3 Step 2 — Collect Papers

API Parameters

Source	OpenAlex API
Year range	2009–2026
Min. citations	5
Sort order	cited_by_count:desc
Per-channel limit	200 papers
Language	English

Collection Process

For each of the 14 channels, each of its 3–4 queries is submitted to OpenAlex. Results are paginated (up to 5 pages total: 1 initial + 4 additional), deduplicated by paper ID within each channel, and capped at 200 papers per channel.

Per-Channel Paper Counts

Capped at 200 10 channels

Channel	Papers
bridge_vulnerability	200
composability_risk	200
counterparty_concentration	200
governance_failure	200
information_asymmetry	200
liquidation_cascades	200
network_contagion	200
regulatory_contagion	200
rwa_transmission	200
validator_concentration	200

Exceeded 200 4 channels

Channel	Papers
gateway_risk	348
liquidity_spirals	263
stablecoin_runs	218
oracle_manipulation	208

Corpus Summary

Total channel-paper assignments	3,037
Multi-channel papers	363
Final unique papers	2,433

Channel membership distribution

Papers can belong to multiple channels. Distribution of how many channels each paper appears in:

Channels per paper	Number of papers
1	2,070
2	238
3	63
4	36
5	9
6	8
7	7
8	2

Check: 2,070 + 238 + 63 + 36 + 9 + 8 + 7 + 2 = 2,433 unique papers. Weighted sum: 2,070×1 + 238×2 + 63×3 + 36×4 + 9×5 + 8×6 + 7×7 + 2×8 = 3,037 channel-paper assignments.

4 Step 3 — Score Channels

Each channel is scored on three dimensions, each normalized to the [0, 1] interval.

Dimension 1: Literature Volume weight = 0.35

Formula

lit_volume(ch) = paper_count(ch) ÷ max_paper_count

Normalization denominator

max_paper_count = 348 (gateway_risk)

Because 10 of 14 channels hit the 200-paper cap, they all receive an identical score of 200 ÷ 348 = 0.5747. Only the four uncapped channels differentiate on this dimension.

Channel	Papers	Calculation	Score
gateway_risk	348	348 / 348	1.0000
liquidity_spirals	263	263 / 348	0.7557
stablecoin_runs	218	218 / 348	0.6264
oracle_manipulation	208	208 / 348	0.5977
10 capped channels	200	200 / 348	0.5747

Dimension 2: Citation Impact weight = 0.35

Formula

cit_impact(ch) = [ Σ top-10 citation counts ÷ 10 ] ÷ max_mean_top10

The denominator is always 10, even if a channel has fewer than 10 papers. This penalizes channels with a very small literature base.

Normalization denominator

max_mean_top10 = 5,228.8 (bridge_vulnerability)

Channel	Mean Top-10	Calculation	Score
bridge_vulnerability	5,228.8	5228.8 / 5228.8	1.0000
validator_concentration	5,041.1	5041.1 / 5228.8	0.9641
composability_risk	3,024.5	3024.5 / 5228.8	0.5784
gateway_risk	1,699.5	1699.5 / 5228.8	0.3250
liquidity_spirals	1,603.6	1603.6 / 5228.8	0.3067
governance_failure	1,213.3	1213.3 / 5228.8	0.2320
liquidation_cascades	1,151.3	1151.3 / 5228.8	0.2202
rwa_transmission	1,148.4	1148.4 / 5228.8	0.2196
counterparty_concentration	1,105.1	1105.1 / 5228.8	0.2113
regulatory_contagion	937.3	937.3 / 5228.8	0.1793
information_asymmetry	927.2	927.2 / 5228.8	0.1773
network_contagion	854.8	854.8 / 5228.8	0.1635
oracle_manipulation	686.9	686.9 / 5228.8	0.1314
stablecoin_runs	161.4	161.4 / 5228.8	0.0309

Dimension 3: Crisis Evidence weight = 0.30

Formula

crisis_ev(ch) = Σ_{events activating ch} log₁₀(losses_usd) ÷ max_sum

The crisis chronology contains 25 events (2014–2025). Of these, 21 have numeric loss figures and 4 have losses recorded as "undetermined". Undetermined losses are imputed at the median of the 21 known losses: $305,000,000 (log₁₀ = 8.4843).

Normalization denominator

max_crisis_sum = 116.08 (counterparty_concentration, activated by 13 crisis events)

Channel	Events	log₁₀ Sum	Calculation	Score
counterparty_concentration	13	116.08	116.08 / 116.08	1.0000
liquidity_spirals	13	115.58	115.58 / 116.08	0.9957
composability_risk	10	85.94	85.94 / 116.08	0.7403
gateway_risk	9	78.92	78.92 / 116.08	0.6799
information_asymmetry	8	71.68	71.68 / 116.08	0.6175
bridge_vulnerability	7	60.02	60.02 / 116.08	0.5171
network_contagion	5	47.11	47.11 / 116.08	0.4058
liquidation_cascades	5	39.80	39.80 / 116.08	0.3429
stablecoin_runs	3	28.44	28.44 / 116.08	0.2450
governance_failure	3	24.36	24.36 / 116.08	0.2099
oracle_manipulation	2	15.23	15.23 / 116.08	0.1312
validator_concentration	1	8.80	8.80 / 116.08	0.0758
regulatory_contagion	1	8.48	8.48 / 116.08	0.0731
rwa_transmission	0	0.00	0.00 / 116.08	0.0000

5 Step 4 — Composite Ranking

Composite Score Formula

composite(ch) = 0.35 × lit_volume(ch) + 0.35 × cit_impact(ch) + 0.30 × crisis_evidence(ch)

Worked Example: bridge_vulnerability (Rank 1)

Step	Calculation	Result
Papers retrieved	200 (capped at per_channel_limit)	200
Literature volume	200 ÷ 348	0.5747
Top-10 citations	[8937, 6328, 6112, 5982, 5761, 4666, 3760, 3620, 3596, 3526]	—
Mean top-10	52,288 ÷ 10	5,228.8
Citation impact	5,228.8 ÷ 5,228.8	1.0000
7 crisis events	Poly Network (8.786) + Wormhole (8.513) + Ronin (8.796) + Nomad (8.279) + Multichain (8.100) + Bybit (9.176) + WazirX (8.371)	60.02
Crisis evidence	60.02 ÷ 116.08	0.5171
Composite	0.35 × 0.5747 + 0.35 × 1.0000 + 0.30 × 0.5171	0.7063

Breakdown: 0.35 × 0.5747 = 0.2012 (literature) + 0.35 × 1.0000 = 0.3500 (citation) + 0.30 × 0.5171 = 0.1551 (crisis) = 0.7063

Full Ranking (All 14 Channels)

Rank	Channel	Papers	Lit. Volume	Mean Top-10	Cit. Impact	Crisis Events	Crisis Ev.	Composite
1	Cross-Chain Bridge Vulnerability	200	0.5747	5,228.8	1.0000	7	0.5171	0.7063
2	Liquidity Spirals	263	0.7557	1,603.6	0.3067	13	0.9957	0.6705
3	Gateway Risk (Fiat-Crypto Bridging)	348	1.0000	1,699.5	0.3250	9	0.6799	0.6677
4	Composability Risk	200	0.5747	3,024.5	0.5784	10	0.7403	0.6257
5	Counterparty Concentration	200	0.5747	1,105.1	0.2113	13	1.0000	0.5751
6	Validator/Miner Concentration	200	0.5747	5,041.1	0.9641	1	0.0758	0.5613
7	Information Asymmetry and Opacity	200	0.5747	927.2	0.1773	8	0.6175	0.4485
8	Liquidation Cascades	200	0.5747	1,151.3	0.2202	5	0.3429	0.3811
9	Network Contagion	200	0.5747	854.8	0.1635	5	0.4058	0.3801
10	Governance Failure	200	0.5747	1,213.3	0.2320	3	0.2099	0.3453
11	Stablecoin Runs	218	0.6264	161.4	0.0309	3	0.2450	0.3036
12	Oracle Manipulation	208	0.5977	686.9	0.1314	2	0.1312	0.2945
13	Regulatory Contagion	200	0.5747	937.3	0.1793	1	0.0731	0.2858
14	Real-World Asset Transmission	200	0.5747	1,148.4	0.2196	0	0.0000	0.2780

6 Step 5 — Sensitivity Analysis

To test the robustness of the ranking, four weight schemes are compared. If the top channels are stable across schemes, the ranking is not an artifact of the chosen weights.

Weight Schemes

Scheme	Lit. Volume	Cit. Impact	Crisis Ev.
Primary	0.35	0.35	0.30
Equal	0.333	0.333	0.333
Crisis-dominant	0.25	0.25	0.50
Literature-dominant	0.50	0.25	0.25

Full sensitivity table (rank per scheme)

Channel	Primary 35/35/30	Equal 33/33/33	Crisis-dom. 25/25/50	Lit-dom. 50/25/25	Range
bridge_vulnerability	1	1	5	3	1–5
liquidity_spirals	2	2	1	2	1–2
gateway_risk	3	3	3	1	1–3
composability_risk	4	4	4	4	4
counterparty_concentration	5	5	2	5	2–5
validator_concentration	6	6	7	6	6–7
information_asymmetry	7	7	6	7	6–7
liquidation_cascades	8	9	9	9	8–9
network_contagion	9	8	8	8	8–9
governance_failure	10	10	10	10	10
stablecoin_runs	11	11	11	11	11
oracle_manipulation	12	12	12	12	12
regulatory_contagion	13	13	13	13	13
rwa_transmission	14	14	14	14	14

Composite scores by scheme

Channel	Primary	Equal	Crisis-dom.	Lit-dom.
bridge_vulnerability	0.7063	0.6972	0.6522	0.6666
liquidity_spirals	0.6705	0.6860	0.7634	0.7035
gateway_risk	0.6677	0.6682	0.6712	0.7512
composability_risk	0.6257	0.6311	0.6585	0.6170
counterparty_concentration	0.5751	0.5953	0.6965	0.5902
validator_concentration	0.5613	0.5381	0.4226	0.5473
information_asymmetry	0.4485	0.4565	0.4967	0.4861
liquidation_cascades	0.3811	0.3792	0.3702	0.4281
network_contagion	0.3801	0.3813	0.3874	0.4297
governance_failure	0.3453	0.3388	0.3066	0.3978
stablecoin_runs	0.3036	0.3007	0.2868	0.3822
oracle_manipulation	0.2945	0.2867	0.2479	0.3645
regulatory_contagion	0.2858	0.2757	0.2250	0.3504
rwa_transmission	0.2780	0.2648	0.1986	0.3423

Key Findings

Top-5 core is stable: bridge_vulnerability, liquidity_spirals, gateway_risk, composability_risk, and counterparty_concentration appear in every scheme's top 5 (though their internal ordering shifts).
composability_risk holds rank 4 in all four schemes — the most stable single position.
Ranks 10–14 are locked: governance_failure, stablecoin_runs, oracle_manipulation, regulatory_contagion, and rwa_transmission never leave positions 10–14 across all schemes.
Fragile pairs:
- Ranks 2–3 (liquidity_spirals vs. gateway_risk): gap of 0.003 in primary scheme. Under crisis-dominant weights, liquidity_spirals surges to rank 1.
- Ranks 8–9 (liquidation_cascades vs. network_contagion): gap of 0.001 in primary scheme. They swap under equal and crisis-dominant weights.
bridge_vulnerability drops from rank 1 to rank 5 under crisis-dominant weights, because its citation advantage (1.0) is compressed while counterparty_concentration's crisis dominance (1.0) is amplified.

7 Verification — Independent Recomputation

All 42 sub-scores (3 dimensions × 14 channels) and 14 composite scores were independently recomputed from raw data and compared against the stored values in channel_rankings.json.

Full verification table (stored vs. computed)

Channel	Literature Volume			Citation Impact			Crisis Evidence			Composite			Rank
	Stored	Computed	Δ	Stored	Computed	Δ	Stored	Computed	Δ	Stored	Computed	Δ	Match
bridge_vulnerability	0.5747	0.5747	0.0000	1.0000	1.0000	0.0000	0.5171	0.5171	0.0000	0.7063	0.7063	0.0000	Yes
liquidity_spirals	0.7557	0.7557	0.0000	0.3067	0.3067	0.0000	0.9957	0.9957	0.0000	0.6705	0.6705	0.0000	Yes
gateway_risk	1.0000	1.0000	0.0000	0.3250	0.3250	0.0000	0.6799	0.6799	0.0000	0.6677	0.6677	0.0000	Yes
composability_risk	0.5747	0.5747	0.0000	0.5784	0.5784	0.0000	0.7403	0.7403	0.0000	0.6257	0.6257	0.0000	Yes
counterparty_concentration	0.5747	0.5747	0.0000	0.2113	0.2113	0.0000	1.0000	1.0000	0.0000	0.5751	0.5751	0.0000	Yes
validator_concentration	0.5747	0.5747	0.0000	0.9641	0.9641	0.0000	0.0758	0.0758	0.0000	0.5613	0.5613	0.0000	Yes
information_asymmetry	0.5747	0.5747	0.0000	0.1773	0.1773	0.0000	0.6175	0.6175	0.0000	0.4485	0.4485	0.0000	Yes
liquidation_cascades	0.5747	0.5747	0.0000	0.2202	0.2202	0.0000	0.3429	0.3429	0.0000	0.3811	0.3811	0.0000	Yes
network_contagion	0.5747	0.5747	0.0000	0.1635	0.1635	0.0000	0.4058	0.4058	0.0000	0.3801	0.3801	0.0000	Yes
governance_failure	0.5747	0.5747	0.0000	0.2320	0.2320	0.0000	0.2099	0.2099	0.0000	0.3453	0.3453	0.0000	Yes
stablecoin_runs	0.6264	0.6264	0.0000	0.0309	0.0309	0.0000	0.2450	0.2450	0.0000	0.3036	0.3036	0.0000	Yes
oracle_manipulation	0.5977	0.5977	0.0000	0.1314	0.1314	0.0000	0.1312	0.1312	0.0000	0.2945	0.2945	0.0000	Yes
regulatory_contagion	0.5747	0.5747	0.0000	0.1793	0.1793	0.0000	0.0731	0.0731	0.0000	0.2858	0.2858	0.0000	Yes
rwa_transmission	0.5747	0.5747	0.0000	0.2196	0.2196	0.0000	0.0000	0.0000	0.0000	0.2780	0.2780	0.0000	Yes

Result: All 42 sub-scores and 14 composite scores match exactly (Δ = 0.0000). All 14 rank assignments match. Zero discrepancies in the primary scoring.

Sensitivity recomputation discrepancy

One micro-discrepancy was found in the sensitivity analysis (not the primary scoring):

Scheme	Channel	Metric	Stored	Computed	Δ
Equal (33/33/33)	liquidation_cascades	composite	0.3793	0.3792	0.0001

Explanation: The equal-weight scheme uses 1/3 for each weight. The stored value was computed with the exact fraction 1/3 (0.33333...); the recomputed value used a rounded representation (e.g., 0.3333). This yields a Δ of 0.0001 — well within floating-point tolerance and does not affect the rank. This is the only discrepancy across all 4 schemes × 14 channels = 56 comparisons.

8 Known Limitations

#	Limitation	Impact
1	200-paper cap compresses literature_volume. 10 of 14 channels hit the cap and receive an identical score (0.5747). This reduces the discriminating power of the literature volume dimension to only 4 channels.	Moderate — the dimension differentiates among only 4 uncapped channels; 10 are tied.
2	Citation-count sorting favors older, established work. The OpenAlex query sorts by cited_by_count:desc, so the top-10 papers per channel tend to be highly-cited older papers. Recent, rapidly-growing papers may be underrepresented in the top-10 used for the citation impact dimension.	Low-to-moderate — mitigated by the 2009–2026 year range restriction.
3	Crisis chronology is manually curated. A single coder selected 25 events from the 2014–2025 period. The event set is neither exhaustive nor independently validated. Smaller DeFi exploits (<$10M) are excluded.	Moderate — crisis_evidence scores depend entirely on which events are included.
4	Terra/Luna ($45B) disproportionately weights crisis_evidence. The Terra/Luna collapse activates 5 channels and contributes log₁₀(45,000,000,000) = 10.65 per activation, the largest single weight. Channels linked to Terra/Luna receive a substantial boost.	High for affected channels — counterparty_concentration, liquidity_spirals, composability_risk, stablecoin_runs, and network_contagion all benefit.
5	Weights (35/35/30) are analytical judgment, not empirically derived. The primary weighting scheme was chosen to balance literature presence and scholarly impact equally, with a slightly lower weight for crisis history (which has fewer data points).	Low — sensitivity analysis shows the top-5 is robust across all four tested schemes.
6	API non-determinism. OpenAlex results may vary slightly across retrieval dates (new papers indexed, citations updated). The exact paper set is a snapshot, not a stable reference.	Low — the raw data files are archived for reproducibility.
7	4 crisis losses imputed at the median ($305M). Events with "undetermined" losses (Black Thursday, SushiSwap Vampire Attack, SVB/USDC De-Peg, BTC-e Seizure) receive the median log₁₀ weight of 8.4843. The true losses may differ substantially.	Low — log-scale compression limits the impact of imputation errors.

Generated from methodology_verification.json and channel_rankings.json. All numerical values sourced directly from pipeline output files.