The systems a society runs on are under persistent, low-intensity, unattributed pressure: supermarket refrigeration, blood banks, hoax bomb emails, rail, drones, all in 2026. HYDRA (Hybrid Destabilization Readiness Assessment) studies the terrain those hits land on, the weave of who depends on whom, because the terrain decides what each hit becomes. This page shows the simulation under three ways of weighting the input: declared priors, and the public record of 2026 by count and by reach.
Loading the simulation results.
HYDRA does not predict who, when or how: the attacker is noise. It predicts the aftermath and the next receiver, from the terrain and from what has already happened, and writes it down with a date and a deadline before it happens. The text of a prediction is never edited; the outcome is added when the check decides it, right or wrong.
Reach priors are the balanced model. The two register groundings weight each class by what the public record of 2026 contains: by number of incidents, or by how far each one reached (blast radius 1 to 5). The register is kinetic-heavy, so the physical classes carry more under grounding.
Mean cascade = average number of terrain nodes reached per random disturbance; lower is better. Vigilance = train the person who receives the signal. Convergence = the departments share one cross-sector picture (the HYDRA intervention), modelled by breaking the adjacency and learning-loop mechanisms. The share convergence removes is the same under every grounding; the size of the cascade is not.
Share of runs in which the node is reached, sorted by the blast-radius grounding. The nodes most often reached are the ones that depend on others, not the ones first hit. Grounding on the register brings logistics and transport forward; weighting by blast radius brings retail back to the top.
Mean cascade per class is a property of the graph, so it barely moves across groundings; what the grounding changes is how often each class is dropped. Shared control surfaces and single shared dependencies reach furthest.
Register groundings only (the priors carry no duration). Bar length on a square-root scale so the small classes stay visible; the number is the mean node-days. The classes that fill the headlines are over in hours or days; the financial and vendor-coupling classes hold their victims down for weeks or for good. The forged-authority figure rests on two rows at opposite ends and is an artefact until the class has more rows; "permanent" is carried as 120 days, a lower bound.
No single doom-shape dominates in any grounding: the terrain produces a family of overlapping medium cascades, which argues for reading exposure continuously rather than preparing one scenario.
The attacker is modelled as noise: each run drops one abstract class of disturbance (eight classes, each grounded in events already public, never an attack recipe) onto the nodes it first touches, and the effect spreads along the passage rules of the terrain with distance decay. Three scenarios differ only in the defence. The classes are abstract by decision: the object of study is the terrain and its defence, not the method.