Spec · CC-BY 4.0

Adjudication and Auto-Modification Council.

Five frontier Voters × five Lenses adjudicate each Inference — advisory at 3-of-5, binding at 4-of-5, Council self-modification at 5-of-5. Builds on the identity primitives defined in ATS v1.0.

[02]Voters × Lenses5 frontier Models · 1 verdict per Inference

Five Voters, five Lenses.

Direct-SDK inference to each frontier Model — no L2-router competitors in the path. Each Voter brings a different Lens to the same Inference.

Risk lens
Claude Opus 4.7
claude-opus-4-7
Statistical lens
GPT-5.5
gpt-5-5
Anti-Gaming lens
Gemini 3.1 Pro
gemini-3-1-pro
Regulatory lens
Grok 4
grok-4
Behavioral lens
Mistral Large 3
mistral-large-3

[03]Quorums3-of-5 · 4-of-5 · 5-of-5

Three quorums, three consequences.

3-of-5 · advisory

A majority flags the Inference. The call proceeds, but the flag travels with the AuditEvent so an Auditor can see what the Council saw.

4-of-5 · binding

Strong consensus blocks the Inference at L3 before any debit. The Wallet stays whole; the AuditEvent records a refusal with the per-Voter reasoning.

5-of-5 · Council self-modification

Unanimous consensus blocks the Inference and tightens the originating Spend Policy. The Council writes its own auto-modification into the AuditChain.

[04]RolloutT1 launch · T2 week 1 · T3 Series A

T1 · Pre-Seed launch
Advisory mode, 3-of-5 threshold, post-call analysis. The Council observes every Inference and writes its verdicts into the AuditChain.
T2 · Week 1
Varda performs manual review on each flagged Inference. Disagreements between the Council and Varda land in the Ontology for spec refinement.
T3 · Series A
Full binding mode and Council self-modification. The Council enforces; Tenants opt in per Agent or per Spend Policy.

AAMC v1.0 is published as a spec — Ainfera implements it, and welcomes feedback from other agent-infrastructure builders before a v1.1. The full spec lives in github.com/ainfera-ai/specs.