Skip to content

Signals: Cognitive Kernel Abstraction

Overview

Signals are the fundamental abstraction used to represent normalized cognitive quantities inside the ARVIS kernel.

They replace raw scalar values (float) with semantically meaningful, typed objects.

Design Principles

  • Normalized: all signals are constrained to [0, 1]
  • Immutable: signals are frozen dataclasses
  • ZKCS-safe: no internal metadata is exposed
  • Composable: signals can be combined and transformed
  • Semantic-first: logic should depend on signal meaning, not raw thresholds

Available Signals

RiskSignal

Represents collapse risk.

RiskSignal(0.0 → 1.0)

Helpers:

  • is_low()
  • is_moderate()
  • is_high()

Note: these signal bands (0.3 / 0.7 / 0.85) are descriptive observability vocabulary and deliberately distinct from the input-risk gate's decision thresholds (0.4 / 0.8, arvis/kernel/gate/input_risk.py). is_high() describes a signal; only the gate's constants decide. - is_critical() - is_transition_zone() - is_unstable_zone()

UncertaintySignal

Represents epistemic uncertainty.

UncertaintySignal(0.0 → 1.0)

Helpers:

  • is_high()

DriftSignal

Represents system drift / instability.

DriftSignal(0.0 → 1.0)

Helpers:

  • is_high()

Signal Boundary

Controllers should never depend directly on raw floats.

Instead, they should coerce at the boundary:

from arvis.math.signals.coercion import to_float, to_risk

or use semantic helpers:

if risk.is_critical():

Migration Strategy

The system supports hybrid usage:

  • float (legacy)
  • Signal (preferred)

Controllers are progressively migrating toward full signal-native logic.

Long-term Vision

Signals will become the primary carrier of:

  • uncertainty
  • risk
  • stability
  • trust

Future evolutions may include:

  • probabilistic signals
  • interval-based signals
  • distribution-aware signals

Role in Control Loop

Signals are not passive representations.

They are actively used to:

  • drive control adaptation (epsilon, exploration)
  • influence decision gating
  • propagate stability constraints

They form a closed-loop interface between measurement and control.