Build an inspectable network model
Represent the response system as an auditable directed graph.
PROJECT DOSSIER · VERSION 1.0.0
A reproducible digital twin for studying disaster-response reliability
This page documents the complete v1.0.0 research chain.
ABOUT THE PROJECT
The research problem
After a major disruption, adding more responders does not necessarily restore service if the road network itself has fragmented.
Scientific boundary
The severity variable is synthetic and must not be interpreted as earthquake magnitude.
RESEARCH OBJECTIVES
Separate allocation from topology
Measure algorithmic efficiency separately from network reachability.
Quantify a reliability boundary
Estimate the 80/80 service-reliability boundary with uncertainty.
DATA AND PROVENANCE
Public data enters a reproducible evidence chain
AFAD
Public earthquake catalogue access and normalization.
OpenStreetMap
Road-network geometry and emergency-facility context.
SHA-256
Frozen evidence is accompanied by integrity hashes.
SYSTEM ARCHITECTURE
From public data to frozen evidence
1DataAFAD + OSM
→
2GraphDirected road network
→
3DisruptionCoupled stochastic worlds
→
4DecisionsGreedy + Global Min-Cost
→
5InferencePaired + bootstrap
→
6EvidenceGitHub + Zenodo DOI
METHODS
Coupled stochastic severity worlds
The same stochastic world is progressively stressed across severity.
Transparent decision baselines
Greedy and Global Minimum-Cost Assignment are compared on identical scenarios.
EXPERIMENTAL DESIGN
A staged research cycle, ending in a confirmatory freeze
Paired Monte Carlo
Repeated-world inference and corrected paired comparisons.
Phase boundary
Severity-resource grids expose a transition region.
Confirmatory boundary
Fine-grid confirmation closes the final boundary.
FROZEN V1.0.0 RESULTS
More responders move the boundary upward, but the gain is not linear
| Responders | Boundary estimate | 95% world-bootstrap interval |
|---|---|---|
| 12 | 0.142500 | 0.134000–0.148462 |
| 14 | 0.149167 | 0.142500–0.156000 |
| 16 | 0.152500 | 0.145333–0.160000 |
| 20 | 0.154500 | 0.147000–0.163214 |
| 24 | 0.158333 | 0.148462–0.168214 |
| 32 | 0.162500 | 0.154167–0.172727 |
0.1425
80/80 boundary at 12 responders
0.1625
80/80 boundary at 32 responders
102/102
Confirmatory cells significant after Holm correction
REPRODUCIBILITY
The evidence chain is preserved
The final evidence is traceable to two frozen GitHub Actions runs.
Frozen software record
v1.0.0
10.5281/zenodo.21903851
Concept DOI for all versions
10.5281/zenodo.21903850
REPOSITORY MAP
app/Interactive Streamlit research interfacesrc/turkiye_disaster_twin/Data, simulation and research softwarescripts/Reproducible benchmark entry pointstests/Automated regression and method testsresults/Frozen research evidence and checksumsdocs/Public GitHub Pages presentation.github/workflows/CI and research benchmark automationLIMITATIONS AND RESPONSIBLE USE
A research platform, not an operational warning system
The results demonstrate a reproducible computational method rather than validated operational performance.
CITATION
Kowsari, F. (2026). Türkiye Disaster Intelligence Digital Twin (v1.0.0) [Computer software]. Zenodo.
https://doi.org/10.5281/zenodo.21903851
For a project-level reference that should follow future versions, use the Concept DOI:
https://doi.org/10.5281/zenodo.21903850
AUTHOR
Faramarz Kowsari
Author · Software Engineer · AI Researcher
Faramarz Kowsari is based in İstanbul and develops open research software and technical research systems.