Skip to content

intervention_datasets.md

Latest commit

 

History

History
135 lines (100 loc) · 4.27 KB

intervention_datasets.md

File metadata and controls

135 lines (100 loc) · 4.27 KB

LIF Intervention Datasets

Last Updated: 2026-02-13
Version: 1.0

Intervention Dataset Overview

The LIF intervention datasets track emergency response mechanisms in blockchain protocols.

Dataset Structure

High-Fidelity Metrics (lif_intervention_metrics.csv)

Records: 52 (includes proactive cases)

Contains detailed timing and effectiveness data for intervention cases:

  • Detection time (minutes)
  • Containment time (minutes)
  • Success percentage
  • Scope and authority classification
  • Source attribution with confidence levels

All Interventions (lif_all_interventions.csv)

Records: 130

Complete database of exploit-linked cases where intervention mechanisms were activated, including:

  • Exploit responses
  • Governance actions
  • Emergency pauses

Intervention Classification

By Scope

Scope Count Description
Protocol 60 Entire protocol intervention
Account 46 Individual account freeze
Network 9 Network-wide intervention
Module 9 Specific module/function pause
Asset 6 Token/asset-level action

By Authority

Authority Count Description
Delegated Body 48 Council/committee
Signer Set 47 Multisig emergency signers
Governance 35 Community vote

By Success Tier

Tier Count Description
Full (100%) Prevented all losses
Partial (1-99%) Prevented some losses
Reactive (0%) Response after losses
N/A No intervention

Case Types

Exploit Interventions

Cases where protocols responded to active exploits:

  • Example: Curve Finance vyper bug response
  • Example: BNB Chain bridge exploit pause
  • Count: 46 cases in metrics dataset

Proactive Interventions

Cases where protocols intervened without active exploit:

  • Example: MakerDAO PSM pause during USDC depeg
  • Example: Tether PDVSA wallet freeze
  • Count: 7 cases in metrics dataset

Prevented Definitions

The project uses multiple valid but different “prevented loss” aggregates. To avoid ambiguity, always specify which dataset (and therefore which definition) you are using.

Prevented (All Interventions)

  • Source: data/refined/lif_all_interventions.csv
  • Definition: sum(loss_prevented_usd) across exploit-linked intervention events.
  • Use for: Website/report headlines about prevented losses across the 130 exploit-linked intervention cases.

Prevented (Metrics Subset)

  • Source: data/refined/lif_intervention_metrics.csv
  • Definition: sum(loss_prevented_usd) across the 52-case curated, high-fidelity subset.
  • Use for: Claims that explicitly reference the curated metrics subset (timing + prevention-confidence).

Prevented (Eligible + Intervened subset)

  • Source: data/refined/lif_exploits_final.csv (filter: is_lif_relevant==True and is_intervention==True)
  • Definition: sum(loss_prevented_usd) across intervention-eligible exploits where an intervention occurred.
  • Use for: Market-coverage style analysis where the unit of analysis is the “intervention-eligible exploit market.”

Important note (paper model vs dataset totals)

  • The paper-style market decomposition (e.g., coverage vs effectiveness layers) is a separate modeling/aggregation layer and should not be conflated with dataset-aggregate prevented sums.

Data Quality Standards

All intervention cases meet these criteria:

  1. Primary source documentation
  2. Clear timeline of events
  3. Quantified losses (actual or prevented)
  4. Identified intervention mechanism
  5. Confidence level assessment

Usage

For Researchers

Use lif_intervention_metrics.csv for:

  • Effectiveness analysis
  • Response time studies
  • Authority performance comparison
  • Scope effectiveness research

For Protocol Developers

Reference lif_all_interventions.csv for:

  • Incident response precedents
  • Mechanism design patterns
  • Failure mode analysis

Methodology Notes

  • Detection Time: From exploit start to protocol awareness
  • Containment Time: From awareness to successful intervention
  • Success %: Funds prevented / Total funds at risk
  • Confidence: High (official source), Medium (multiple sources), Low (single source)