Loop Readiness Gate — Governed Loop Engineering Addendum

Purpose

Define the minimum readiness record before a Work Slice may use a loop-shaped execution pattern. This gate reconciles existing loop authorities into one review surface: Execution Pattern Selection, Objective Gate Policy, Loop State Contract, Maker-Checker Policy and Independent Validation Hardening.

The gate is docs-first. It does not create a scheduler, runtime, policy engine, autonomous agent, queue, database, collector or new lifecycle.

Boundary

The Loop Readiness Gate answers one question: is this loop topology admissible for this slice?

It does not:

execute the loop;
schedule future runs;
authorize tool calls;
approve merge/deploy/release;
replace AHGE, AHC, objective gates, maker-checker, human review or loop state.

Minimum record

loop_readiness_gate:
  gate_id:
  work_slice_ref:
  execution_pattern_selection_ref:
  proposed_pattern: loop_until_done | scheduled_stateful_loop
  loop_purpose:
  recurrence:
    kind: single_session | recurring | scheduled
    max_iterations:
    max_runtime:
    budget_ref:
  objective_stop_condition:
    condition:
    verification_method:
    evidence_ref:
  objective_gate:
    required: true
    gate_ref:
    gate_type: test | schema | lint | typecheck | reproduction | rubric | independent_review | human_review
  state:
    required: true | false
    loop_state_ref: not_needed | pending | string
  review_capacity:
    reviewer_ref:
    cadence:
    max_unreviewed_iterations:
    escalation_owner:
  drift_controls:
    drift_signals: []
    stop_on: []
    rescope_requires_human: true
  safety_controls:
    irreversible_action: true | false
    sensitive_context: true | false
    human_review_ref: not_required | pending | string
  evidence_refs: []
  source_refs: []
  verdict: ready | ready_with_constraints | not_ready | human_led_required
  constraints: []

Normative rules

No objective gate, no autonomous loop. If an objective gate is unavailable, the verdict must be human_led_required or not_ready.
No open-ended stop condition. The stop condition must be externally verifiable. “Until it looks good” is invalid.
Recurring loops require persistent state. If recurrence.kind is recurring or scheduled, state.required must be true and loop_state_ref must not be not_needed.
Review capacity is part of readiness. A loop is not ready when no reviewer, cadence, maximum unreviewed iterations or escalation owner is declared.
Drift is a stop signal, not a normal loop output. Scope drift, repeated failed gates, rising risk, repeated unavailable evidence or reviewer overload must stop or rescope the loop.
Human review remains required for sensitive or irreversible actions. No loop may make or finalize irreversible/sensitive changes without a human review reference.
Evidence and source references are mandatory. The readiness gate must point to the records that justify its verdict. Missing telemetry or unavailable evidence is represented as unavailable, never invented.
The gate is not a scheduler. A ready verdict says the topology is admissible; it does not cause execution.

Verdict semantics

Verdict	Meaning
`ready`	All required readiness conditions are present and source-backed.
`ready_with_constraints`	The loop may proceed only under explicit constraints such as lower iteration cap or human checkpoint.
`not_ready`	Required loop preconditions are missing; do not use loop topology.
`human_led_required`	The work may continue only as human-led or human-gated execution.