MIT Open Source Offline Capable

Predicting violence
before it satisfies.
Evaluating what prevents it.

Open-source AI for programme evaluation and conflict early warning. 12 years of direct field experience, encoded into tools that work offline.

0
Conflict Events Analyzed
0
Countries Tracked
0
Evaluation Indicators

Featured Research

No single event type predicts violence. Co-occurrence does.

PRAXIS applies epidemiological surveillance logic to ACLED event data. Analysis of over 100,000 events shows that when individual conflict types occur in isolation, escalation drops to baseline. When two or more specific types co-occur in the same district-week, lethal violence follows within 7 days.

The composite co-occurrence model was validated across 13 countries over 15 years of weekly admin2-level panel data. All results survive Bonferroni correction at p < 0.001.

Read the methodology
Evidence Summary
8.8x
Where to look
KAFD+IED districts at 8.8x clean baseline, excluding triggering types. N=292.
1.2x
When to act
Signal week marks 18% further escalation in non-triggering violence above pre-signal corridor.
13
Where it works
Central Sahel, Lake Chad Basin, Coastal West Africa. All p < 0.001.
Full interactive visualization

Evaluation Infrastructure

Built for the realities of the field.

Six live tools covering 298 indicators. Every tool runs client-side with zero data transmission, ensuring you can calculate sample sizes or build evaluation matrices even when the internet drops.

The Lab

Built from field reality

PRAXIS is grounded in 12 years of programme evaluation across conflict-affected settings including Burkina Faso, Mali, Niger, South Sudan, and Burundi. The methodology draws on epidemiology, econometrics, and applied M&E practice.

Emmanuel Nene Odjidja, Founder and Director. Section Editor, Journal of MultiDisciplinary Evaluation. 29 peer-reviewed publications.

12+
Years in the field
29
Publications
6
Countries
MIT
Open Source

Open Source

Where PRAXIS is going

The methodology transfers wherever ACLED operates. Roughly 230 countries. The evaluation tools are MIT licensed and built to be forked, adapted, and extended. This is an open invitation.