Skip to content

Policy Recalibration

Recalibration procedure — instantiates Moving-Target Tracking

The deliberate procedure for revising an operating policy when the moving objective makes the prior rule unfit — escalating when no policy can meet the target, and rolling back a recalibration that misfires.

Some of what a tracking system relies on is not a gain or a model but a policy — a human-owned operating rule that says how the organization responds. When the target moves far enough, the old policy stops fitting: the thresholds are set for a demand that no longer exists, the response rule optimizes for last year's conditions. Policy Recalibration is the deliberate procedure for revising that rule. Its two defining moves are what happen at the edges of the revision: it carries an escalation rule for when the target has moved somewhere no feasible policy can reach — declaring the objective untrackable and routing it up rather than pretending a tweak will do — and a rollback rule for when a recalibration itself proves worse than what it replaced, reverting cleanly. It works at the policy layer, by deliberate human process — which is what distinguishes it from a controller that re-tunes its own gains automatically, and from the machinery that governs the target's own version history.

Example

A transit agency runs its bus network on a service policy — headways, span of service, route priorities — set for a pre-disruption ridership pattern. After a sustained shift in where and when people travel, the policy no longer fits: some corridors are overserved and empty, others are crushloaded and under-resourced. Policy Recalibration is the structured revision: the planning team re-derives headways and route priorities against the current demand pattern, publishes the change, and watches its effect.

Two edge rules make the procedure more than a plan edit. First, when the analysis shows the demand now exceeds what the fleet and budget can serve at any feasible headway policy — the target has moved outside the envelope — the escalation rule fires: rather than quietly degrade service and call it a recalibration, the gap is escalated as a resourcing or scope decision for someone with the authority to change the constraints. Second, when a recalibrated schedule ships and on-time performance actually worsens against a guardrail, the rollback rule reverts to the prior policy while the team diagnoses — so a bad recalibration is bounded, not endured. The result is a policy that tracks the moved objective and fails safely when it cannot.

How it works

What distinguishes recalibration from an ordinary policy edit is that it is governed at its boundaries and triggered by unfitness, not habit:

  • It is triggered by a diagnosed misfit — a drift or break signal showing the current policy no longer serves the moved objective — not by a fixed calendar.
  • It re-derives the operating rule against current conditions through a deliberate, owned human process, with the authority to change the rule made explicit.
  • It carries an infeasibility-and-escalation rule: when no policy within the current constraints can reach the target, it stops recalibrating and escalates the gap to whoever can change scope, budget, or the target itself.
  • It carries a rollback rule: a recalibration that breaches a guardrail after deployment is reverted to the last good policy, cleanly and by prior agreement, rather than defended.

Tuning parameters

  • Recalibration threshold — how much policy misfit is tolerated before a revision is triggered. Low keeps the policy sharply fitted but invites constant churn and whiplash; high is stable but lets unfitness accumulate.
  • Escalation boundary — how far the target must move beyond feasibility before the gap is escalated rather than absorbed. Tight escalates honestly and early; loose keeps quietly degrading service under the name of recalibration.
  • Rollback guardrail — which post-deployment metrics, breached by how much, force a revert, and how fast. Sensitive bounds a bad change tightly but can revert a slow-improving one prematurely.
  • Revision scope — whether a recalibration may adjust thresholds only or rewrite the rule's structure; broader recovers more fit but is riskier and slower to validate.
  • Change authority — who may approve a recalibration versus an escalation; centralizing protects coherence, distributing speeds response.

When it helps, and when it misleads

Its strength is keeping a human-owned operating rule fitted to a moving objective while failing safely — the escalation rule refuses to let a recalibration paper over a target that has become genuinely infeasible, and the rollback rule bounds the damage of a revision that turns out wrong. Together they make policy revision a controlled maneuver rather than a gamble.

Its failure modes are the politics of a rule with owners. Recalibrate too readily and you get policy whiplash — constant revision that erodes predictability and the credibility of the policy itself. Its deepest trap is specific to policies over people: revising the rule changes the behavior the rule was tracking, so the target moves in response to the recalibration — the Lucas critique — and a recalibration that ignores this chases a target it is itself displacing.[1] And the classic misuse is running it backwards — recalibrating to justify a decision already taken, or endlessly re-tuning a policy to avoid the harder escalation that infeasibility actually demands. The discipline that guards against this is to trigger on diagnosed misfit rather than preference, to honor the escalation boundary instead of absorbing infeasibility, and to keep the rollback guardrail pre-agreed so a bad recalibration cannot be quietly defended.

How it implements the components

Policy Recalibration fills the archetype's governed-revision slot — the parts that revise a human policy and bound that revision at its edges:

  • infeasibility_and_escalation_rule — when no policy within current constraints can reach the moved target, it declares the objective untrackable and escalates rather than degrading in place.
  • rollback_rule — a deployed recalibration that breaches a guardrail is reverted to the last good policy, cleanly and by prior agreement.

It does not re-tune a controller's gains automatically (Adaptive Control Method), re-fit a predictive model (Model Retuning), or govern the target's own version and authority history (that objective-side governance is Objective Versioning and Change Log). It revises the operating policy and manages its failure edges.

Notes

Policy Recalibration is the policy-layer, human-owned cousin of Adaptive Control Method: both revise the response when the world moves, but one is a deliberate governed procedure over a rule with owners and accountability, and the other an automatic, continuous re-tuning of a controller. The escalation and rollback rules are what make it more than an editable plan — they are the admission, built into the procedure, that a moving target can move past where any policy can follow, and that a revision can be wrong. Owning those two edges is the whole point.

References

[1] The Lucas critique (Robert Lucas) — the observation that the behavioral relationships a policy is calibrated against will themselves shift once the policy changes, because the people governed by it re-optimize in response. A recalibration that treats the target as fixed while it revises the rule is chasing an objective its own change is displacing. ↩