The Unthinkable Experiment

The smell was the first thing you noticed after the hum—the scent of hot insulation, ozone, and something faintly metallic. It clung to the air in Control Room Number Four, a space lit by the ambient glow of instrument panels in the otherwise darkened hall. Outside the wide windows, the Ukrainian night was black.

Inside, by 23: 00 on April 25, 1986, the day shift was a memory. The special team of electrical engineers waited. The evening shift of reactor operators, who should have been overseeing a routine shutdown, remained at their stations. They were holding Reactor Unit Four at 1600 megawatts of thermal power, a level it had maintained for nine unplanned hours.

The deputy chief engineer of the entire Chernobyl Atomic Energy Station, Anatoly Dyatlov, was there to see the delayed procedure through. He stood observing the main control desk, a compact man with a watchful stillness. To engineers who valued competence, Dyatlov was respected, a figure who had forged a career from a Siberian village to the heart of Soviet nuclear power.

To those who bore the weight of his exacting standards, he could be inflexible. His presence on the evening shift for a test scheduled for the day was a signal. The test was to be conducted during the day-shift of 25 April 1986 as part of a scheduled reactor shutdown. The day shift had been instructed in advance on the reactor operating conditions to run the test, and a special team of electrical engineers was present to conduct the electrical measurements. Now it was night, the schedule had ruptured, and the reactor sat in a state of suspended transition.

The delicate, unstable equilibrium had to be held, but for a purpose that was already slipping out of alignment with its original, safer design. This chapter charts the deliberate, step-by-step journey into the accident’s precipitating conditions. It is a movement from the origins of a flawed institutional mindset to the specific, fateful decisions made in these final hours.

The catastrophe did not begin with a bang, but with a sanctioned probe toward the edge of a reactor’s known operating envelope—a walk toward what would become recognizable as the Border of the Impossible. Every action taken in this control room would be logical within the plant’s operational culture. Each step was an exercise in epistemic containment, an attempt to manage a ferociously complex physical system using a model that was simplified, and in critical aspects, incomplete. The written procedure was a script. The reactor’s physics were about to rewrite it.

---

The test had a bureaucratic pedigree. It was a repeat. The safety systems of an RBMK-1000 reactor required electrical power. If external power failed, diesel generators would start, but they needed forty-five to seventy seconds to reach capacity. Could the spinning inertia of the plant’s own massive steam turbines, coasting down after a shutdown, generate enough electricity to bridge that gap and keep crucial coolant pumps running? The turbine run-down energy capability still needed to be confirmed experimentally. Previous tests had ended unsuccessfully.

An initial test carried out in 1982 indicated that the excitation voltage of the turbine-generator was insufficient. A new test procedure was drafted for the next scheduled shutdown. It became a line item, a technical box to be checked. Its completion was a metric of diligence, a task. The plan stipulated a stable thermal power of 700 to 1000 megawatts—about a quarter of the reactor’s maximum—when the test commenced. At that level, the RBMK was considered manageable. The test would simulate a station blackout, triggering a shutdown. The turbines would be disconnected from the grid to freewheel, and their electrical decay would be measured. That was the theory.

Practice began diverging long before the evening shift gathered. The power reduction had started that afternoon. At 13: 05, a senior reactor control engineer began withdrawing neutron-absorbing control rods to lower output toward the test window. By 14: 00, power was around 1600 megawatts.

Then the Kiev regional grid controller called. Another power station had gone offline unexpectedly. The controller requested that Chernobyl’s further reduction be postponed; electricity was needed for the evening demand peak.

The reactor was held at 1600 megawatts. It would remain there for nine hours. This delay injected the first critical variable. An RBMK reactor at steady power produces xenon-135, a radioactive fission product that absorbs neutrons voraciously. At stable high power, the neutron flux burns off xenon as fast as it forms. But when power is reduced sharply, xenon production continues initially while the neutron flux needed to burn it off diminishes. The result is a buildup of neutron-poisoning xenon in the core—xenon poisoning. It makes the reactor sluggish, requiring operators to withdraw more control rods than usual to raise power again. The phenomenon was textbook physics, a known quirk. At Chernobyl, it was a normalized part of the operational landscape, a routine challenge.

At 23: 04, permission finally arrived from Kiev. The grid controller allowed the reactor shutdown to resume. The day shift had departed. The evening shift was fatigued, expecting to leave. Now they were ordered to execute the test that should have been completed hours earlier.

The imperative to complete the sanctioned procedure began to outweigh the original conditions for its safe execution. The order was given to resume reducing power. The reactor was already poisoned from the long hold. As output was deliberately lowered toward the 700-megawatt target, the poisoning intensified. The crew embarked on the step-by-step journey: maintaining stability while driving the reactor into a lower-power zone where it was inherently less stable, all while fighting an accumulating neutron poison. Their epistemic containment—the belief that their procedures and models could safely bound this process—was under steady, invisible stress.

Shortly after midnight, reactor power approached 500 megawatts. Here, the crew made a consequential choice. To simulate a true station blackout for the test, certain automatic safety systems had to be disabled to prevent them from aborting the experiment. One such system was the Emergency Core Cooling System. Another was the automatic shutdown signal that would be triggered by the turbine itself during the test. These systems were switched off or bypassed.

In the frame of the test, this was logical: to observe what happened during a simulated accident, you had to stop the machine from automatically preventing it. The act was written into some versions of the test program. The broader safety culture that should have questioned disabling multiple layers of protection during a risky maneuver on a poisoned reactor was absent. The immediate goal—completing the test—defined operational reality. Power continued to fall. It dipped below 500 megawatts and kept dropping. The reactor, heavy with xenon, grew more unresponsive.

Then came the next link in the chain. At approximately 00: 28 on April 26, an operator attempted to stabilize the falling power. The exact circumstances that caused the power drop are unknown. Most reports attribute the power drop to Toptunov’s error, but Dyatlov reported that it was due to an unexpected reactivity transient. The result was a sudden power drop to an unintended near-shutdown state, with a power output of 30 MW thermal or less. The control room indicators showed a level so low it was off the scale of normal operation.

The reactor was now deep in what operators called the “iodine pit,” a condition of severe xenon poisoning where the chain reaction is nearly suffocated.

This was a crisis for the test, not yet for the reactor. The protocol demanded a minimum of 700 megawatts. They were at less than one-twentieth of that. The experiment was impossible under these conditions.

Anatoly Dyatlov’s response was forged by this operational imperative. Aborting the test after all the delays would represent professional failure. He gave the order to raise power. To raise power from such a poisoned state required withdrawing almost all the reactor’s 211 control rods. The operators began this process. Dozens of rods were pulled out manually.

The RBMK design held a fatal flaw: each control rod had a graphite “displacer” at its tip. When a rod was fully withdrawn, this graphite segment sat in the core channel, actually improving neutron moderation slightly. As rods were inserted, the graphite was replaced by neutron-absorbing boron. In normal operation, with many rods partly inserted, this quirk was manageable.

But with almost all rods fully withdrawn—an extreme state never envisioned in safety analyzes—the core was left with minimal neutron-absorbing material. It was full of graphite moderators and enriched uranium fuel. It was primed for a surge if conditions shifted.

Power began to climb sluggishly from the pit. It rose to 200 megawatts by 01: 00—still far below the test requirement, but movement. Concern tightened within the crew. The senior reactor control engineer, Leonid Toptunov, who had been at the controls during the plunge, was visibly uneasy. The unit shift chief, Alexander Akimov, expressed reservations about proceeding. The reactor parameters were strange.

But Dyatlov pressed on. In the plant’s culture, where senior engineer authority was absolute and reactor quirks were normalized hurdles, this was a difficult recovery to be mastered. The test could still be done.

As part of final test preparations, the crew activated two additional main circulating pumps at around 01: 05. These giant pumps pushed water through the core for cooling. With four pumps running instead of the usual two, the flow of cooling water through the reactor increased dramatically.

Some versions of the test had included this step to see how the turbines performed under high pump load.

The increased water flow had an unintended physical consequence. Cooler, denser water absorbs neutrons better than hot water or steam. The flood of cool water through the core’s 1661 pressure tubes began quenching the neutron flux further, acting as another brake on the chain reaction. Power stabilized, even dipped again. To compensate, the operators withdrew even more control rods. By 01: 19, most manual control rods were fully out. The automatic control rods were also near their withdrawal limits. The reactor was being controlled on a razor’s edge, with almost no safety margin left in its control systems.

Then came a final, decisive choice. To try to raise and stabilize power at the now-accepted target of 200 megawatts, the operators switched off the automatic control system governing the last remaining regulating rods. This system might have made corrective adjustments based on parameters like power or pressure. With it off, all control was manual.

The reactor was now flying without an automatic pilot, its stability dependent entirely on human operators reacting to instrument readings that lagged behind the actual, rapidly changing conditions inside the core. At 01: 22: 30, in an adjacent room, a computer printout generated a preliminary calculation of the reactor’s operational reactivity margin—a measure of how many control rods were effectively available to shut it down safely. The number was catastrophically low, far below the permissible limit. This data existed in that moment. It was a numerical warning that epistemic containment had fractured; the model itself was signaling danger. Whether this information reached the control desk in time, or whether it would have been heeded, was about to become irrelevant.

In the control room, focus had narrowed to starting the test itself. The electrical engineers were ready. All preparatory steps, however aberrant, were deemed complete. The reactor was holding at around 200 megawatts. It was unstable, poisoned, overcooled by excess water flow, and deprived of almost all its control and safety rods. But it was holding. At 01: 23: 04, the command was given.

Someone pushed the button to begin the turbine coast-down experiment. The turbines were disconnected from the electrical grid. As they began to spin down, feeding less energy to the main circulating pumps, the flow of cooling water through the core started to decrease.

Inside Reactor Number Four, physics took command. With less cooling water flowing, it heated and boiled. Steam voids formed in the pressure tubes. Steam is a far poorer neutron absorber than water. As water turned to steam, its neutron-absorbing ability plummeted. In a well-designed reactor, this would be compensated by automatic systems or a sufficient bank of control rods. Here, there was almost nothing left to absorb neutrons. The reduction in water absorption triggered a massive, instantaneous spike in reactivity. Power surged upward beyond any instrument’s ability to measure it in that first second. The last recorded human action was at 01: 23: 40, when someone—likely Akimov or Toptunov—slammed the button for the emergency manual shutdown, attempting to drive all control rods back into the core. It was too late.

As the rods began their slow descent, their graphite displacers entered the bottom of the steam-filled core first, displacing neutron-absorbing water at a critical moment and briefly increasing moderation before the boron sections followed. This design flaw turned a scram signal into a final trigger. The control room is locked in a moment of supreme tension, with the reactor on a knife’s edge and the test’s start command imminent. Then a deep, muffled thud echoes through the building structure. The floor jerks. Lights flicker. On the control panel, needles on meters swing hard over and then drop dead. For two heartbeats there is only noise, vibration, and blinking warning lamps. The journey into precipitating conditions is over. The long walk toward the Border of the Impossible has ended at its threshold. What lies beyond is no longer an experiment, but a reality for which there is no procedure.