What each profile addresses
Three runtime profiles.
Phionyx is the runtime evidence layer for AI decisions. It does not make the model correct — it makes what the model did, on what basis, from what input, and how the outcome closed undeniable, and forces calibrated abstention when the system is out of its depth. Every profile emits its record in AIREP.
Evidence Runtime
Turns each AI decision into a signed, hash-chained, replayable record. Tamper at any link is detected — deterministically, at the exact record. The record, not the model's word, is the account of what happened. It is the load-bearing asset, and it deploys on its own — no physics, no full pipeline required.
The artifact
A signed, replayable per-decision record (emitted in AIREP) — verifiable offline by anyone.
What it does not claim
It does not make the model correct. It makes what the model did provable.
The two pillars that complete it
The second pillar · the boundary
Abstention & Boundary
When the system is outside what it can evidence, it hedges, asks, defers, or refuses — with a calibrated confidence, not a guess. The calibration is the value, which is why it is its own runtime, not a single block.
Explore Abstention & Boundary →The gate
Safety Gate
Fail-closed gates between model output and action, with escalation to a human. The model proposes; the runtime decides what may proceed — and records the decision.
Explore the Safety Gate →The profiles compose. A high-stakes application can run the Safety Gate over Abstention & Boundary over Evidence Runtime — fail-closed where it must, abstaining when out of depth, and signing the whole decision for replay. See it as a workflow: high-stakes decision support.
The EchoState physics and the causal mind-loop are the research substrate behind the runtime — not the governance pitch, and not required to deploy the evidence layer.