Skip to content

ulissesflores/cyberfinancial-resilience-lrblstm

Repository files navigation

Cyber-Financial Resilience (LR-BLSTM)

Reproducible research software for studying cyber-financial resilience in crypto markets using high-frequency public data, Little's Law-inspired observability proxies, and a roadmap toward Bayesian sequence modeling.

This repository is designed as an audit-grade scientific artifact. Phase 1 is complete and focuses on data collection, exploratory analysis, and reproducible run management. Later phases extend the project toward Bayesian LSTM modeling and resilience-oriented inference under market stress and non-stationarity.

At a glance

What this repository is

  • a reproducible pipeline for collecting high-frequency public market data
  • a run-based artifact system with manifests, checksums, and deterministic traceability
  • an exploratory analysis layer for volatility, stress, and flow-intensity proxies
  • the software foundation for future Bayesian LSTM work on cyber-financial resilience

Why it matters

Crypto-financial markets behave like complex socio-technical systems:

  • they are non-stationary
  • they exhibit burstiness and clustered volatility
  • they show regime shifts and stress propagation
  • they offer only partial observability through public market data

This project uses queueing intuition inspired by Little's Law to construct proxy-observable signals while keeping strict epistemic guardrails: the outputs are descriptive and reproducible, not causal claims disguised as certainty.

Current project status

  • Phase 1: data pipeline and EDA complete
  • Phase 2: Bayesian LSTM modeling planned
  • Phase 3: resilience metrics and stress propagation planned

Key outputs

  • immutable run directories under runs/<RUN_ID>/
  • machine-readable manifests and SHA-256 checksums
  • generated figures for volatility, drawdowns, trade intensity, and inter-arrival behavior
  • multilingual scientific documentation for reproducibility and modeling scope

Data and observability

  • Exchange: Binance public API
  • Instrument: BTC/USDT
  • Granularity: 1-minute OHLCV plus public trades
  • Access model: public REST only

The repository does not use proprietary, private, or user-identifiable data.

Quick reproduction

1. Create the environment

git clone https://github.com/ulissesflores/cyberfinancial-resilience-lrblstm.git
cd cyberfinancial-resilience-lrblstm

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Initialize a run

python scripts/make_run.py --note "phase-1 baseline"

Use the generated run_id printed by the script in the next steps.

3. Collect market data

python scripts/collect_data.py \
  --run_id <RUN_ID> \
  --exchange binance \
  --symbol BTC/USDT \
  --timeframe 1m \
  --ohlcv_days 90 \
  --with_trades \
  --trades_days 14

4. Generate EDA outputs

python scripts/eda_generate_figures.py --run_id <RUN_ID>

All generated artifacts are written under runs/<RUN_ID>/.

Scientific guardrails

  • queue-inspired variables are proxy observables, not direct queue measurements
  • public exchange APIs provide partial visibility
  • results are period- and regime-dependent
  • the repository makes descriptive and methodological claims, not direct causal claims

Multilingual documentation

Additional documentation is also available in pt-BR and es.

Citation and release

If you use this work, cite the DOI-backed archival release rather than an unversioned repository snapshot.

Repository layout

configs/   run and data configuration
docs/      scientific documentation in multiple languages
schema/    formal schemas for datasets and manifests
scripts/   reproducible data collection and EDA pipelines
runs/      immutable run outputs created during execution

About

Research software for cyber-financial resilience modeling with stochastic systems, Little\'s Law, and LSTM forecasting.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages