Factor discovery: name the forces first
Before imagining futures, the engine identifies what actually drives resolution: typically six to fifteen factors, each with a concrete description, its current status (what is true today, not what is generically relevant), its trend relative to the outcome, and an importance grade. Factors are ranked, so the scenario stage builds on what matters most.
One rule is non-negotiable: at least one factor must represent structural friction — the inertia that keeps the status quo in place. Forecasters systematically catalogue the forces of change and forget the forces of continuity; the engine is required not to.
Scenario construction: distinct worlds, not variations on one
From the factors, the engine generates causally distinct world states — three at Quick depth, up to fifty at Ultra. Each scenario is a coherent story about how the future unfolds: a name, a narrative, a stance (bullish, mixed, or bearish with respect to YES), and a record of how each factor plays out in that world.
Distinctness is the whole point. Two scenarios that reach the same outcome through the same underlying mechanism are not two pieces of evidence — they are one piece counted twice. So every scenario declares a mechanism family (“regulatory shift”, “macro liquidity”, “supply shock”), and scenarios sharing a mechanism must share a family. That honesty has teeth: family concentration is discounted downstream, so the engine cannot inflate a forecast by telling the same bullish story through different doors.
Narrative generators — human or machine — are biased toward worlds where something happens, because those make better stories. In most real questions, “current trends simply continue” is among the most probable single futures. The engine generates it explicitly in every run, so its weight is measured rather than forgotten.
Two further rules keep the set honest. Anti-duplication across batches: at higher depths scenarios are generated in batches, and each batch sees everything already generated with instructions not to repeat causal paths. No meta-scenarios: worlds about the question itself (“the question is ambiguous”, “resolution is disputed”) are banned at generation and flagged by the judge if one slips through — they contaminate aggregation and are discounted.
A named causal driver with current status, trend, and importance — the atoms of the analysis.
A coherent, causally distinct world state with a narrative and per-factor variations.
The honest label for a scenario's underlying causal engine; concentration in one family is discounted.
The enforced scenario in which no major catalyst arrives and current trends persist.