← TERMS
decision reversibilityreversibility classreversibility classesreversibility awareness
OrganisationCommitment & authority

Decision Reversibility Classes

How costly it is to undo a committed choice — sets rigor, runway length, and supersession speed

MARKO ŽIVKOVIĆ In review

Decision reversibility classes classify how costly it is to undo a committed choice from the moment of bind forward — not how scary the decision felt in the room. Decision Making Principle 5: Reversibility Awareness requires this classification before debate closes so process rigor, quality mechanisms, execution runway length, and mandatory supersession speed match stakes.

Reversibility is orthogonal to outcome severity (how bad being wrong is) and to task nature (deterministic vs probabilistic). A decision can be reversible but catastrophic if wrong (rollback possible, harm severe during the wrong path) or irreversible but minor (rare, but cannot be unwound). The (A)DR records both reversibility class and severity at bind.

The three classes

ClassMeaningTypical examplesProcess implication
ReversibleCan undo or replace without lasting structural harm; rollback, feature flag, policy toggleFeature-flag library choice; marketing experiment; draft architecture behind a flagLighter pre-bind process; short execution runway; supersede quickly when triggers fire — delay turns cheap fixes expensive
Costly to reverseCan undo only with significant migration, reputational damage, write-off, or coordinated reworkAdopted core data model; major vendor contract; public product direction with shipped dependenciesStronger quality mechanisms; shorter runway than feels comfortable; tighter falsification triggers; mandatory review on fire
IrreversibleCannot be unwound — only superseded forward; past consequences remainSWIFT payment sent; drug administered; person terminated; immutable L0 facts consumed downstreamStrongest pre-bind rigor; human at commitment boundary when severity warrants; fewer commits at this class; still must revise on catastrophic contradicting evidence — stronger bar, not infinite hold

Classification is forward-looking from bind: if we are wrong, what does correction cost from here? not how much have we already spent? (sunk-cost commitment uses the wrong question).

De facto irreversibility

A decision classified reversible at bind can become de facto irreversible if revision waits too long — dependencies harden, narrative locks in, customers adapt, data migrates. Refutation propagation and prompt supersession exist partly to prevent that drift. Reversibility class sets how fast “usually” becomes “mandatory” once a falsification trigger fires (We Already Decided).

In the (A)DR

At bind, record:

  • Reversibility class — reversible | costly to reverse | irreversible
  • Severity if wrong — catastrophic | significant | minor | negligible (outcome severity)
  • Decision class — matrix cell when applicable
  • Runway and triggers scaled to class — reversible → short runway, fast supersession; irreversible → strongest evidence before bind, catastrophic override paths still armed

Misclassifying downward to reduce process friction is a governance failure — same family as understating severity in article 19.

→ Decision Strategy Framework · (A)DR Decision Log