a11oy — Brain

Anatomy role: Brand Orchestration / gates · Doctrine v10 · gate vocabulary + composition rules + live playground + unified LLM router

46policy gates

44anchor formula gates

0.90Λ floor

5LLM tiers

749Lean declarations

163tracked sorries

1 · Gate composition rules

AND-compose (conjunctive safety): a decision passes only if every fired gate passes.
Λ-floor: geometric-mean Λ across the 13 axes must be ≥ 0.90 (soundnessAxiom).
Severity-indexed witnesses: higher severity ⇒ more attested witnesses (thresholdPolicySeverity).
Monotone composition: adding a gate can only lower or keep Λ, never raise it (TH8 GLR-consistent).

Λ = geometric mean of axis scores — matches Lutar.Λ k (A2 IsHomogeneous, A4 IsBounded). Λ uniqueness is a Conjecture, not a closed theorem (open CAUCHY_ND sorry at Uniqueness.lean:120 + missing symmetry axiom).

2 · Live composition playground

Gate names (comma-separated). Leave blank for the first 3 gates.

13-axis scores (comma-separated, each in [0,1]).

// result appears here

3 · Unified LLM router — 5 founder-locked tiers

The router picks a tier from the 13-axis Λ trust vector: high-trust (Λ ≥ 0.90) → cheap fast tier; low-trust / adversarial (Λ < 0.75) → premium tier + extra gates. Same endpoint, identical logic, on every Space.

rank	model	intended use	why

// router result appears here

HONEST: no model API key is wired into this HF Space, so response is an honest stub describing what would be sent. The tier selection + Λ-receipt are real, deterministic math. The receipt signature is a PLACEHOLDER — Sigstore CI signing is not yet wired (Doctrine v10).

Source of truth: unified router platform/packages/llm-router · canonical numbers lean_numbers.json @ c7c0ba17 (749 / 14 unique / 163). LLM tiers per ROSIE_FULL_CAPABILITY_BRIEF §2. ADDITIVE surface — all 40+ GREEN routes preserved. ZERO BANDAID.