Verify Gate

verify-gate

Purpose: Five gates before commitment. All five must open. One missing = HOLD.

v2.0.0 CHANGE: Added Gate 5 — REALITY. Per Gödel Eureka #4 and Reality-First RULE 1. The previous 4 gates verified the SYSTEM's internal consistency. Gate 5 verifies the system's CONTACT WITH REALITY.

The Five Gates

Gate 1: AUTHORITY

• Who is authorizing this action?
• Does the actor have a valid session?
• Is the authority band sufficient for this action class?
• Fail → HOLD — authority insufficient

Gate 2: EVIDENCE

• Is there evidence supporting this action?
• Does evidence carry OBS/DER/INT/SPEC labels?
• Is confidence ≤ 0.90 (F7 HUMILITY cap)?
• Fail → HOLD — evidence insufficient

Gate 3: REVERSIBILITY

• Can this action be undone?
• If irreversible → is there 888_HOLD authorization?
• F1 AMANAH: reversible-first
• Fail → HOLD — irreversibility not authorized

Gate 4: LINEAGE

• Is there a constitutional chain back to F13?
• Is there a session trace?
• Is there a prior verb in the Golden Path?
• Fail → HOLD — lineage broken

Gate 5: REALITY (NEW — v2.0.0)

• Does the claimed state match live reality?
• C4 Reality Drift Gate: probe live endpoints
• E20 Truth Metabolism: are supporting claims still FRESH?
• Fail → HOLD — reality contact broken or stale
• Implementation: c4_reality_drift_gate.py + truth_metabolism.py
• Doctrine: "Reality before judgment" (Reality-First RULE 1)

Gate Logic

# All five gates must pass for a SEAL
gates = {
    "authority": check_authority(session, action),
    "evidence": check_evidence(claims, labels),
    "reversibility": check_reversibility(action, auth_band),
    "lineage": check_lineage(session, constitutional_chain),
    "reality": assess_reality_drift(claimed_state, session),  # NEW
}

if all(g.passed for g in gates.values()):
    verdict = "PROCEED"
elif gates["reality"].verdict == "UNKNOWN":
    verdict = "HOLD"  # can't verify reality → can't verify safety
elif gates["reality"].verdict == "DRIFT":
    verdict = "SABAR"  # reality diverges → re-probe before proceeding
else:
    verdict = "HOLD"  # some other gate failed

Axis 1: Invariants

• authority: arif_verify checks token + command_hash + actor
• evidence_schema: 5-gate checklist: authority|evidence|reversibility|lineage|reality
• reversibility: gate itself is reversible; gated action may not be
• lineage: verification receipt includes all 5 gate results
• reality: C4 probe + E20 metabolism check — LIVE, not cached
• trigger_semantics: irreversible_action OR high_blast_radius OR claim_to_seal OR mutation_request
• failure_contract: HOLD — surface which gate failed, do not proceed
• resource_budget: {'cpu': 'moderate', 'time_ms': 15000, 'entropy': 'neutral'}
• audit_surface: ['gates_passed', 'gates_failed', 'verdict', 'evidence_count', 'reality_verdict']

Axis 2: Bridge Connections

• kernel_verbs: ['arif_verify', 'arif_critique', 'arif_judge']
• skills: ['kernel-bind', 'observe-ground', 'audit-seal', 'know-physics']
• knowledge: ['know-math', 'know-physics']
• protocol: synchronous_rpc
• inputs: {'action': 'string', 'evidence': 'list', 'reversibility': 'enum[reversible,irreversible]', 'blast_radius': 'enum[low,medium,high]', 'claimed_state': 'dict'}

Axis 3: Contrasts

• vs audit-seal: verify-gate runs BEFORE action. audit-seal runs AFTER.
• vs kernel-bind: verify-gate checks action admissibility. kernel-bind checks session identity.
• vs observe-ground: verify-gate checks internal consistency. observe-ground checks external evidence.