NROL-αΩ
? What is NROL-AO? ▶
NROL-AO is a self-hosted Bayesian epistemic engine for forecasting real-world events — built around one load-bearing principle: beliefs only move when the world moves. Every posterior change is traceable to a typed observation routed through a pre-committed likelihood. Humans and LLMs are perception: they notice, extract, and propose. The server alone is authority: it validates and commits.
How a topic works
A topic poses a question with a date-banded resolution (e.g. "when does Hormuz reopen?"). It carries mutually-exclusive hypotheses (H1, H2, …), each with a prior and a midpoint. Indicators are pre-committed, falsifiable signals — each with explicit likelihoods (LRs) saying how it moves each hypothesis if it fires. Anti-indicators are the inverse: their LRs are authored to suppress their targeted hypothesis.
When evidence matches an indicator, a typed transition (FIRE / OBSERVE / PARK / SCHEMA_GAP) is proposed. Only a validated, approved commit runs the Bayesian update — there is no freeform "set H3 to 0.72" path, even with approval.
Active vs Shadow
The active (committed) posterior is moved only by approved typed commits through indicator likelihoods. It quantizes sub-threshold evidence to LR=1.0 and has only a deadline cliff for time.
The shadow posterior is a second, independent estimate computed from a model of how long these situations tend to last — not from news. The key idea it captures: elapsed time is itself evidence. If a strait has been closed 60 days with no reopening, that fact alone makes an early reopening less likely — and the shadow reflects it. The active posterior doesn't count this; it only moves when a specific news event fires an indicator.
The shadow is a guide, not the system. It never changes the real probabilities — it just gives you a second opinion to compare against. When the two disagree, the useful question is: is the active estimate ignoring how much time has passed?