Case

Normalisation of deviance

Normalisation of deviance is the process by which deviations from the expected state come to be treated as normal variation. The term is Diane Vaughan's, from her 1996 study of the Challenger launch decision.

In Kettei, normalisation of deviance is a failure mode of the Persist phase, in which signals are filtered out as acceptable noise. Each deviation is given a rationale and accepted through the prescribed review, so the tolerance itself grows. The people involved are following the rules as the rules have come to stand.

The case

The Space Shuttle Challenger broke apart 73 seconds after lift-off on 28 January 1986, killing the 7 crew. The Rogers Commission found the cause in a failed pressure seal on the right booster's aft field joint. Its chapter on the joint's past is titled "An Accident Rooted in History".

Neither NASA nor the booster contractor expected hot gas to touch the rubber O-rings sealing the joints. A primary O-ring came back from a flight in November 1981 burned to a depth of 0.053 inches. After erosion on 2 joints of a flight in February 1984, the senior review approved the next launch "accepting the possibility of some O-ring erosion".

The Commission found that later reviews passed recurring erosion as "acceptable" or "allowable" because it fell "within the data base" of earlier flights. As flights confirmed damage to the seals, it wrote, the reaction of NASA and the contractor "was to increase the amount of damage considered 'acceptable.'"

Diane Vaughan, a sociologist, named the process in her 1996 study. In her reading the managers on the eve of the launch were applying rules that about 5 years of accepted anomalies had already set. Vaughan herself concludes that some risk in complex technical systems is incalculable.

What Kettei asks

The remedy belongs to the validity boundary, the conditions under which a decision remains valid. The principal assumptions are written down at the moment of deciding, and each trigger names a signal, a threshold, an owner, a response window and a response rule. A threshold fixed before the evidence comes in gives later reviews a reference point no single flight can move.

For the booster joint, the assumptions on file would have said that the O-rings take no erosion and that the secondary seals if the primary fails. The erosion of 1981 would have breached the first, forcing a dated revalidation by a named owner against the design expectation. Changing a threshold then becomes a recorded decision with an owner and a date.

A revalidation could still have found the joint fit to fly, and that finding would then have stood on the record with a name against it. No record can show that a boundary would have stopped this launch.

The question to ask

A boundary keeps the reference still while the evidence moves, and the test of one is a question about the reference. Against which standard is this anomaly being judged, the original or the last one accepted?

Read the original

The Acceptable Risk was published on Kettei on Substack on 26 October 2026, and is gathered in Kettei Decision Digest 4: Failures of Stopping, Part 1. The Columbia inquiry of 2003, which noted the same problems 17 years on, is read in Learned Without Consequence.

Sources

  1. Presidential Commission on the Space Shuttle Challenger Accident (Rogers, W. P., Chairman) (1986) Report of the Presidential Commission on the Space Shuttle Challenger Accident, Volume 1. Washington DC. Establishes the cause, the flight history of the joint and the findings quoted above.
  2. Vaughan, D. (1996) The Challenger Launch Decision: Risky Technology, Culture, and Deviance at NASA. Chicago: University of Chicago Press. The source of the term.

From the essay The Acceptable Risk on Kettei on Substack, 26 October 2026.

The cover of Kettei: The Discipline of Governed Intent by Gareth A Holebrook

Kettei

The Discipline of Governed Intent