Resources

Notes from the canopy.

What we're learning while building Aegis, written for the people who will have to defend a scenario tree in a real review.

Method note

12 May 2026 · 7 min

Why similarity search fails at risk prediction

We fine-tuned a multilingual embedding model on 60,000 industrial events. It clustered well, and still could not see the accident that had not happened yet.

Read the note →

Method note

2 Apr 2026 · 9 min

MADS-MOSAR, thirty years on: what still works

The French systemic-risk method got the ontology right: source, flux, target, barrier. What it never had was a way to stay current. That part is now solvable.

Read the note →

Field note

18 Feb 2026 · 8 min

Mapping cascades in urban lifeline networks

Power, water, health, transport: four networks, one failure surface. Notes on modeling the dependencies that turn a local hazard into a territorial event.

Read the note →

Notes from the canopy.

Why similarity search fails at risk prediction

MADS-MOSAR, thirty years on: what still works

Mapping cascades in urban lifeline networks

See your territory the way risk actually moves through it.