Methodology
How the system reads the world.
Paqshi is not a newsletter with opinions. It is an instrumented model of structural stress with published predicates, horizons, and falsification criteria. This page describes how it works and how its calls are scored.
The ontology
100+ chokepoints, seven categories.
The world is modeled as a set of structural chokepoints — the specific nodes where a disruption propagates disproportionately into geopolitics, energy, finance, or the physical network. Each is instrumented with its own stress reading on a 0–1 scale, updated continuously as new signal ingests.
The forecasting approach
Probabilistic disruption models — explicit horizons, explicit triggers.
Each forecast is a probabilistic disruption model over a named chokepoint. It carries three non-negotiable components, fixed at issuance:
- PredicateA specific, checkable claim about what happens — not a vibe. 'The MoU won't hold 60 days,' not 'tensions may rise.'
- HorizonThe window inside which the predicate must resolve. A call with no clock is not a call.
- Falsification triggerThe observable condition that would prove the forecast wrong. Published before the outcome is known.
Probabilities update continuously as signal arrives. A forecast is a living object until its horizon closes or its trigger fires.
The reasoning layer
Beyond scoring — simulation, adversarial play, conditional projection.
Cascade simulation
Disruptions are propagated through the chokepoint network to find second- and third-order effects a single-node reading would miss.
Multi-agent theatre
Key actors are modeled as agents with incentives and constraints, then played forward to surface which moves are actually available.
Conditional foresight
Projections are issued as branches — if X, then Y within Z — so a forecast stays legible even as the world updates.
How calls get scored
Confirmed means reality matched the prediction, inside horizon. Failures are named.
Every forecast has explicit falsification criteria published at issuance. A call is confirmed when reality matched the prediction within its horizon — not when a headline vaguely rhymes with it after the fact.
A call is failed when its trigger did not fire inside horizon, or when the falsification condition was met. Failed calls are named openly on the record — the track record is only worth reading because the misses are on it too.
No retroactive edits. No quiet deletions. The timestamp is the point.