Temporal Convergence Accumulation
Idea
The principle that agreement among forecasting methods about a future date should build gradually as the date approaches: a near date (2032, at 6 years out) can legitimately show high agreement because events can already be checked, while a far date (2040, at 14 years out) legitimately shows less because most of the chain hasn't unfolded yet. High confidence about a distant date is a warning sign of a system making things up, not a sign of strength — and comparing near-date and far-date agreement as if they were equivalent is an error.
The engine's definition — word for word
Convergence on a future prediction date is a TEMPORAL PROCESS, not a snapshot metric. Framework-count convergence grows as the causal chain between now and the target date resolves. A prediction date close to the present (e.g. 2032 at 6 years out) can legitimately show high convergence because observed events can already be measured against framework predictions. A prediction date far from the present (e.g. 2040 at 14 years out) will legitimately show lower convergence because most of the causal chain required to validate or falsify intermediate framework states has not yet unfolded. **Premature high convergence on a distant date is a red flag, not evidence of strength.** cyclical-framework lean on a date 14 years out would indicate confabulation — the engine would be confidently resolving questions it cannot yet resolve from observed data. External audits that treat convergence as a snapshot metric (comparing 3/8 at 2040 against 8/8 at 2032 as if they are comparable evidence strength) commit a category error. The engine's job is to hold space for convergence to accumulate as causal chain resolves, and to maintain pre-committed falsification windows at each predicted date. Applies generally to every multi-year prediction the engine makes and to every future framework audit.
Walk this on the live map →