On the morning of January 28, 1986, the temperature at Kennedy Space Center was thirty-six degrees Fahrenheit — fifteen degrees below the lowest temperature at which the Space Shuttle had previously launched. The night before, engineers at Morton Thiokol had argued in a teleconference with NASA managers that the launch should be postponed. Their concern was specific: the O-rings that sealed the joints between segments of the solid rocket boosters lost resilience at low temperatures. The data showed erosion at temperatures above thirty-six degrees. The data did not extend to thirty-six degrees. Roger Boisjoly, the engineer who had been studying the O-ring problem for months, could feel — in the engineering sense — that thirty-six degrees was wrong. NASA managers asked for quantitative proof. The engineers could not provide it, because the phenomenon had not been tested at the relevant temperature. The absence of quantitative proof was interpreted as the absence of risk. The launch proceeded. Seventy-three seconds after liftoff, the O-ring in the right solid rocket booster failed. Challenger broke apart. Seven astronauts died.
Petroski treated the Challenger as the clearest illustration of engineering judgment operating against institutional demand for quantitative proof. Boisjoly's argument was not based on a formula but on accumulated understanding — the sensitivity of a career spent working with materials and seals to the conditions under which materials behave in ways their specifications do not predict. The argument was correct. It was rejected because the institution could not weigh calibrated judgment against the request for proof that judgment could not, by its nature, supply.
The deeper institutional failure, documented extensively by Diane Vaughan in The Challenger Launch Decision (1996), was what Vaughan called the normalization of deviance — the process by which O-ring erosion, initially unexpected, had become routine at NASA through repeated launches where erosion occurred without catastrophic failure. Each successful launch confirmed, to the institution, that erosion was acceptable. The acceptance narrowed the margin. By January 1986, the margin was zero. When Boisjoly's judgment said the margin was zero and the data did not yet demonstrate that it was zero, the institution chose the absence of demonstration.
The Presidential Commission that investigated the disaster concluded the decision-making process was flawed — that the institution had treated the absence of proof as proof of absence. The recommendations included better communication channels between engineers and decision-makers, clearer protocols for evaluating risk under uncertainty, and greater weight given to engineering judgment. These reforms addressed the institutional problem. Petroski's framework makes visible a deeper issue they did not address: the developmental problem of how to produce engineers whose judgment is worth trusting, and institutions whose practices can recognize judgment when it is offered.
The Challenger's relevance to the AI era is precise. AI systems possess extraordinary capacity for calculation but no capacity for the felt, extrapolative judgment Boisjoly brought to the January 28 teleconference. When an AI system reports that the evidence does not demonstrate a specific failure threshold, its report is technically accurate and structurally identical to the NASA managers' reasoning on the night before the launch. The difference between the AI's accuracy and Boisjoly's judgment is not a matter of calculation quality. It is a matter of what each can do when the evidence is incomplete. The AI can report the incompleteness. Only the engineer can feel what the incompleteness portends.
The disaster occurred on January 28, 1986. The Rogers Commission, established by President Reagan, issued its report in June 1986, identifying the O-ring failure and the flawed decision-making process. Diane Vaughan's The Challenger Launch Decision (1996) provided the most thorough sociological analysis, establishing the normalization-of-deviance framework that has shaped subsequent understanding of institutional failure. Petroski drew on both the Commission's findings and Vaughan's analysis in his treatment of the case, particularly in Design Paradigms (1994) and subsequent work.
The engineers' judgment was correct. Boisjoly and his Thiokol colleagues had, through decades of materials experience, developed the judgment that correctly identified the threshold of catastrophic failure. Their judgment was rejected not because it was wrong but because it could not be expressed in the quantitative form the institution required.
Absence of proof is not proof of absence. The institutional habit of treating the absence of quantitative demonstration as equivalent to the absence of risk is the specific failure mode the Challenger revealed. Under conditions that have not been tested, the absence of failure data is not evidence of safety — it is evidence that the relevant test has not been performed.
The normalization of deviance is a structural process. Margins are consumed incrementally through routine acceptance of small deviations. Each acceptance confirms the next. The process is invisible from inside the institution because each step appears continuous with what preceded it.
AI reports incompleteness; only engineers can feel it. The difference between the AI's accurate report of insufficient data and the engineer's judgment that insufficient data means danger is the difference between processing the map and having walked the territory.
Some subsequent analyses have argued that the NASA decision was not unreasonable given the information the decision-makers had at the time — that calling the judgment "correct" after the fact reflects hindsight bias. The argument has technical merit: the quantitative case for postponement was weaker than Boisjoly's felt case, and institutional decision-making cannot always privilege felt judgment over quantitative analysis. Petroski's response, consistent with his broader framework, was that institutions must develop the capacity to weigh calibrated judgment appropriately, even when it cannot be reduced to quantitative proof. The alternative — demanding quantitative proof for every engineering concern — guarantees that the concerns which only judgment can detect will be systematically dismissed, with consequences measured in lives.