Three Mile Island's Warning Was Written Four Times, Then Filed
Everyone who tells the Three Mile Island story tells the same second half: an inspector warned two federal regulators days before the accident, and it was too late. That's true — and, per the government's own investigation into itself, almost beside the point. The document that actually predicted the failure, fourteen months earlier, never got near a regulator. It died somewhere far more ordinary.
Chapters
- 0:00 9:34 p.m., September 24th, 1977
- 0:36 The story everyone already knows
- 1:02 The procedure that said throttle back
- 1:21 A second procedure, at the same plant
- 1:41 "Very confusing"
- 2:00 Twenty-two minutes
- 2:22 "I had never seen it before"
- 2:41 The wrong lesson
- 3:01 An engineer who wasn't even on the case
- 3:30 Passed by hand, read closely by no one
- 3:49 The sentence that predicted TMI
- 4:13 One letter among thousands
- 4:41 Fifteen months for an answer
- 5:08 A second warning, dying separately
- 5:46 The part everyone actually tells
- 6:02 One line, not a warning
- 6:23 Five people, three offices
- 6:47 Seven days, not six
- 7:06 22 minutes against 142
- 7:23 What changed
- 7:52 Eleven cases, one new shape
- 8:09 Next: a memo that got the ending right
Transcript
9:34 p.m., September 24th, 1977 0:00
Nine thirty-four at night, September twenty-fourth, nineteen seventy-seven. Davis-Besse Nuclear Power Station, Ohio, running at just nine percent power. A feedwater trip starts a chain reaction of alarms. A relief valve at the top of the pressurizer opens, exactly as designed. A missing relay in the control circuit makes it cycle open and shut, again and again, before it finally sticks open. This is the same failure that would bring down Three Mile Island — eighteen months before Three Mile Island existed.
The story everyone already knows 0:36
Everyone who tells this story tells the same second half. An NRC inspector warned two federal commissioners days before the accident, and it was too late to act. That part is true. According to the government's own investigation into itself, it's also almost beside the point. The document that actually predicted the failure never got anywhere near a commissioner.
The procedure that said throttle back 1:02
For the first few minutes at Davis-Besse, everything worked as built. Emergency pumps kicked in automatically. Then the pressurizer level climbed back toward normal. A written emergency procedure told operators exactly what to do next: throttle back that emergency injection.
A second procedure, at the same plant 1:21
A different procedure, at the same plant, said the opposite. Keep system pressure above the point where steam could form, no matter what the pressurizer level said. The operators followed the right procedure for the wrong signal. Nobody had reconciled the two documents with each other.
"Very confusing" 1:41
Nine minutes in, pressure settled near nine hundred pounds per square inch. Pressurizer level pegged at the very top of its scale. The plant's own report calls this combination, in its words, very confusing. At first, the crew blamed overcooling.
Twenty-two minutes 2:00
At twenty-two minutes, a high-pressure alarm from the containment building forced the real answer. Someone finally recognized the relief valve was still open, and shut the backup block valve by hand. The shift foreman in that control room wasn't struck by sudden insight. His own deposition uses one word: confused.
"I had never seen it before" 2:22
Weeks later, in a followup meeting, one of the operators gave the plainest explanation anyone would offer. His training had never covered a leak where the pressurizer level went up instead of down. He had simply never seen it before. No action came out of that meeting.
The wrong lesson 2:41
The NRC's lead inspector for Davis-Besse was away at a training session that week. A different inspector filled in. His regional report reached a conclusion: no industry-wide lesson here. The valve that failed came from a different manufacturer than the ones used at other plants of this design.
An engineer who wasn't even on the case 3:01
One engineer, with no connection to that investigation, had already worked out why that conclusion missed the point. Carl Michelson, an engineer at the Tennessee Valley Authority, had been studying this exact kind of small leak. He'd started before Davis-Besse even happened. He handed his draft to a personal friend of twenty years, a member of the NRC's advisory committee. It was never formally filed anywhere.
Passed by hand, read closely by no one 3:30
That friend passed it on, informally, to two supervisors inside the NRC. Neither one read it closely. One skimmed it and decided there was nothing new. The agency's own later investigation put it plainly: neither man ever reviewed the paper in detail.
The sentence that predicted TMI 3:49
The following January, Michelson rewrote the key paragraph to be impossible to misread. A full pressurizer, he wrote, may convince the operator to trip the emergency pumps. And a full pressurizer is not a reliable signal for that decision. That is precisely what would happen at Three Mile Island. Fourteen months before it happened.
One letter among thousands 4:13
That sentence never reached the NRC at all. Michelson sent the rewritten report straight to the valve's manufacturer instead, asking for comment. The company had received about ten thousand letters from Michelson's employer since nineteen seventy. Roughly two thousand of them touched on plant safety at all. An engineer there scanned this one, called it too detailed to review right away, and set it aside for four months.
Fifteen months for an answer 4:41
The formal, final answer took fifteen months to arrive — two months before Three Mile Island. In writing, the manufacturer agreed the pressurizer reading wasn't reliable. It concluded that other instruments, read together, would be guidance enough. Michelson asked for clarification, with a deadline in mid March, nineteen seventy-nine. No one ever answered.
A second warning, dying separately 5:08
By then, a second warning was already dying inside that same company, completely independent of Michelson. One of its own engineers had written, back in November nineteen seventy-seven, that the Davis-Besse operators shut off emergency cooling. They never watched pressure at all. The only reply called their action correct. A colleague escalated it again, three months later, with a sharper warning: at full power, this would have uncovered the core. That memo passed through four more people in four months. The last one doesn't remember ever seeing it.
The part everyone actually tells 5:46
Now, the part everyone actually tells. The NRC inspector who brought his concerns to two commissioners. His eighteen-month campaign was mostly about a different problem — pressurizer level reading too low, not too high.
One line, not a warning 6:02
The September event that actually predicted Three Mile Island came up in his presentation exactly once. As one line of evidence that the utility was badly run. Not flagged as an industry-wide risk. The agency's own report says so directly: that wasn't the focus of his presentation.
Five people, three offices 6:23
Even Creswell's own formal warning moved slowly, by design. His request to notify licensing boards passed through five different people, in three separate NRC offices, before it arrived. That trip alone took almost three months. Nobody along the way tried to stop it. Nobody thought it urgent enough to hurry it, either.
Seven days, not six 6:47
He met with the commissioners in person on March twenty-first, nineteen seventy-nine, and recommended the plant be shut down. Seven days later, not six, at four in the morning, the same sequence began at Three Mile Island. This time at ninety-seven percent power, not nine.
22 minutes against 142 7:06
At Davis-Besse, the block valve closed in twenty-two minutes. At Three Mile Island, it took a hundred and forty-two. Same failure, same reactor design. This time, roughly half the core melted.
What changed 7:23
The fix that followed wasn't only about instruments. Within nine months, the industry itself built a new organization, the Institute of Nuclear Power Operations. Its one job: making sure a near miss at one plant reaches every operator of that design within days. Not eighteen months. Every American nuclear utility belongs to it. Sharing operating problems stopped being optional.
Eleven cases, one new shape 7:52
Eleven cases into this channel, this one has a different shape. It wasn't silence, and it wasn't one person's failure. It was several competent people, across three organizations, each one deciding, alone, that it could wait a little longer.
Next: a memo that got the ending right 8:09
In June of nineteen seventy-two, a cargo door nearly tore a new airliner apart in midair. The same door design had already failed once before, in a ground test, two years earlier. An engineer wrote a memo predicting exactly how the next one would end. The mechanism, and the death toll. His own management decided, in writing, that warning the plane's maker wasn't their responsibility.
Description and sources
On September 24th, 1977, at Davis-Besse Nuclear Power Station in Ohio, a relief valve stuck open — the same failure, in the same reactor design, that would bring down Three Mile Island eighteen months later. Pressurizer level rose while pressure fell; operators, following a written emergency procedure to the letter, throttled back the emergency cooling keeping the core covered. A second procedure at the same plant said the opposite. It took twenty-two minutes and a containment alarm to catch the mistake. The regional inspector who filled in that week found no industry-wide lesson: the valve came from a different manufacturer.
He was wrong, and an engineer with no connection to the case had already worked out why. Carl Michelson, at the Tennessee Valley Authority, had been studying this exact kind of small leak before Davis-Besse even happened. His revised report, finished in January 1978, contains a sentence that predicts Three Mile Island outright: a full pressurizer may convince an operator to shut off emergency cooling, and it isn't a reliable signal for that decision.
That sentence never reached the NRC. Michelson sent it straight to the valve's manufacturer instead, asking for comment. It arrived as one letter among thousands, set aside as too detailed to review right away, and didn't get a final answer for fifteen months — two months before the accident. A separate, independent warning was dying inside that same manufacturer at the same time: one engineer's memo about the Davis-Besse operators, escalated by a colleague with an explicit warning about full-power core damage, passed through four more people in four months. The last one doesn't remember seeing it.
And the inspector everyone remembers, the one who went to the commissioners? His eighteen-month campaign was mostly about a different, real problem. The event that actually predicted TMI came up in his presentation once, as a line about a badly run utility, not a generic warning. He met the commissioners on March 21st, 1979. Seven days later, not six, the same failure happened again — at 97% power, not 9%. Davis-Besse closed its valve in 22 minutes. Three Mile Island took 142. Roughly half the core melted.
This wasn't a cover-up, and it wasn't one person's failure. It was several competent people, in three organizations, deciding alone that a warning could wait.
PRINT-READY, FROM THIS CHANNEL
The Failure Atlas, Vol. 01 — Tacoma Narrows · Citicorp Center · Millennium Bridge · Apollo 13 · the 2003 blackout · Hyatt Regency
https://therepository.gumroad.com/l/failure-atlas
PRIMARY SOURCES
- Three Mile Island: A Report to the Commissioners and to the Public ("Rogovin Report"), NUREG/CR-1250, Vol. II, Part 1, NRC Special Inquiry Group, Jan. 1980 — sections I.C.8-I.C.13: https://www.osti.gov/servlets/purl/6613275
- Report of the President's Commission on the Accident at Three Mile Island ("Kemeny Report"), Oct. 1979 — source for the 142-minute figure, established in this channel's previous episode: https://www.tmia.com/sites/tmia.com/files/media/Presidents-Commission.pdf
The 24 technical plates in this video are illustrations generated for the channel by a diffusion image model, styled to match its cyanotype identity. They are diagrams of the system, not photographs of the hardware, and no person is depicted in any of them.
Root Cause investigates why engineered systems fail, using official investigation reports and primary technical literature. Sources for this episode are linked above.
The technical drawings in this video are cyanotype-style illustrations produced for the channel — diagrams, not photographs of the real hardware, control room, or documents. Charts and dimensioned comparisons are drawn from the sources listed above.