Moat · science layer

Science

The deep primitives beneath AIVI/AIRS — belief accuracy, integrity, and measured causal lift. Internal instrumentation; numbers are labeled measured only when backed by real panel observations.

Internal

Belief accuracy

Grounding-Gap

Where each model's parametric belief diverges from cited reality. High score = the model already believes the truth without searching. Every gap names the stale belief and the source that fixes it.

EntityAliases (comma-separated, optional)NicheCity (optional)Key prompts — one per line (max 6)

Engines

The mock engine can't ground, so it's excluded.

Enter a target and its key prompts, then run the probe to measure belief accuracy.

Integrity

Organic · Ours · Poisoned

Every disclosed deliverable is a known cause with a timestamp. A stable series reads organic; a fast, unexplained, adverse-signature shift reads poisoned. The feature vector is emitted so this becomes fully supervised as labels accrue.

EntityCity (optional)Niche (optional)

Enter a tracked entity to classify its most recent authority shift.

Effect size

Measured lift, not guessed

Real AIVI/AIRS lift per intervention, paired at the cell level (Twin-Branch) from before/after audits in the panel. Where data is thin we fall back to the catalog's directional estimate — always labeled.

Niche

measuredreal before/after panel data (n ≥ 2), with 90% CIdirectionalcatalog planning estimate — not yet observed

Enter a niche to estimate measured lifts across every intervention.