Skip to content

Remediation execution authority and ledger

An approved remediation plan is advisory until it is both explicitly authorised and actually executed. RigorFoundry separates those two acts. The remediation_authority module grants bounded, independent authority to execute an exact approved plan; the remediation_executor module records an attested execution ledger against that authority. Neither module runs a procedure, spawns a process, or mutates a repository — execution happens elsewhere, under the authority, and its attested outcome is only validated and content-addressed here.

Execution authority

An ExecutionAuthority binds an independently approved RemediationPlan to an ExecutionBudget and an explicit grant:

  • Approved plan only. Authority requires a plan in the approved state with an independent approval; an advisory plan cannot be authorised.
  • Author excluded. The grantor is never the plan's author, so authoring a plan never authorises its own execution.
  • Bounded validity. The grant has a UTC window (granted_at before expires_at) and a mode: observe authorises a non-mutating dry run only, while execute authorises attested mutation.
  • Aggregate budget. ExecutionBudget caps cumulative wall and CPU seconds, peak memory, and the number of executed steps. Every plan step's declared per-step budget must fit within the aggregate ceiling.
from rigor_foundry.remediation_authority import ExecutionAuthority, ExecutionBudget

budget = ExecutionBudget.build(
    wall_seconds=3600, cpu_seconds=3600, memory_mb=1024, max_steps=10
)
authority = ExecutionAuthority.build(
    approved_plan,
    budget,
    authority_id="authority-1",
    repository_id="repo-a",
    granted_by="release-owner",
    granted_at="2026-07-15T12:10:00Z",
    expires_at="2026-07-15T18:10:00Z",
    mode="execute",
)

authority.authorises(plan) is true only for the exact approved plan the grant binds.

Attested execution ledger

ExecutionLedger.admit records an attested execution against a plan, an authority, and covering isolation claims. It proves the execution invariants and never performs the work:

  • Authority and order. The authority must bind the exact approved plan, the window must not run backwards, and the lane executions must follow the plan's dependency- and conflict-safe batches exactly.
  • Isolation. Each lane names one in-repository ExecutionClaim whose write set covers the lane's write set; claims are unique.
  • Step fidelity. Each StepOutcome matches its plan step, carries the plan step's idempotency key, fits the authority budget, and — when executed — captures evidence covering the step's declared outputs. A rolled-back step must correspond to a step that declared a rollback.
  • Budget, idempotency, mode. Cumulative consumption stays within the aggregate budget, an idempotency key never attests two divergent outcomes, and observe authority admits no executed step.
  • Failure propagation. A lane is skipped exactly when a prerequisite lane failed; a lane whose prerequisite failed may not have run.

The ledger resolves to succeeded, failed, or rolled-back and is content-addressed.

from rigor_foundry.remediation_executor import ExecutionLedger

ledger = ExecutionLedger.admit(
    plan=approved_plan,
    authority=authority,
    lane_executions=(base_execution, dependent_execution),
    claims=(base_claim, dependent_claim),
    started_at="2026-07-15T12:15:00Z",
    finished_at="2026-07-15T12:20:00Z",
)
print(ledger.resolution)

ExecutionLedger.from_dict re-verifies structural integrity, the resolution, and the self-contained mode, budget, and idempotency invariants; the plan-bound checks are the admission gate in admit. Both surfaces are reachable via submodule import (rigor_foundry.remediation_authority and rigor_foundry.remediation_executor).