Every organisation runs two governance systems: written policy, and what is enforced by incentive. Under pressure, the enforced one wins.
On the platform, an announcement apologised for the delay into Osaka. The platform display showed eight minutes late. Then fifteen. Then twenty-two.
When the train finally arrived, the rollsign on the front read Osaka. Trains from this station did not terminate at Osaka. Ever.
Standing on that platform, I had sensed a signal. I did not yet understand what the signal was. I didn't know what had happened. I wouldn't until that evening. Earlier that morning at 9.19 am, 107 people died near Amagasaki Station, fifteen kilometres from where I stood. A rapid commuter train had entered a 70 km/h curve at 116 km/h, and derailed into an apartment building, wrapping cars around the structure. It remains one of the worst post-war rail disasters in Japan.
The cascading failures that were ignored
The investigation took two years. The driver, Ryujiro Takami, running 80 seconds late after overshooting Itami Station earlier in the run, had been distracted in the final minutes by an internal call to his conductor. In Japan, an 80-second deficit is the difference between catching a connecting train and being late to work. It also meant a razor margin for error: running closer to speed limits, taking station stops shorter than designed.
Takami was pleading with the conductor to under-report the size of the overrun to transport command to avoid nikkin kyōiku (日勤教育). This was the company’s programme of ritual humiliation for drivers who made errors. Drivers subject to it copied the rulebook by hand, wrote reflection essays that were rejected and rewritten, performed manual labour around the depot, and were placed in visible positions where colleagues could see them. Programmes ran for days. Drivers feared nikkin kyōiku more than rule bending, because the psychological and career damage was clear and visible. The consequences of rule-bending were probabilistic. Nikkin kyōiku was not.
The official report described it as a form of mental-attitude education (精神論的な教育, seishinron-teki na kyōiku), a phrase that carries pre-war militarist resonance. Drivers who erred and reported accurately faced nikkin kyōiku. Drivers who under-reported and were caught faced worse. The bet, for most, was that they would not be caught. Takami was desperate to avoid it. His attention was on the call. He failed to brake.
Takami was 23. He had been subjected to nikkin kyōiku before. The decisions he made in those final minutes were locally rational, given the system around him. Locally rational does not mean exonerated. It means the object of examination is the system that produced the rationality, not the driver who acted inside it.
Every organisation runs two governance systems: the one written in policy, and the one enforced by incentive and consequence. Under pressure, when they diverge, the enforced one wins.
On paper, JR West had a coherent governance system. In practice, the disciplinary regime ran the operation. The proclaimed policy was "safety first", but the enforced policy was the timetable, backed by nikkin kyōiku. Safety said to slow down. Practice rewarded speeding up.
Signals existed but were ignored. Investigators found timetable pressures built up over the years, across the network. Near misses happened. They were logged. Nothing was done. After the crash, drivers came forward with incidents they had never disclosed – including one driver who had previously taken the exact same curve at 85 km/h, against a 70 km/h limit. He had never reported it. The reason he gave was the certainty of nikkin kyōiku. The commission's own conclusion was sharper: the company's method of grasping incidents – requiring drivers to report them, then punishing the reporters – risked, in its words, inducing the very accidents the system was meant to prevent.
Drivers kept crossing the limits of what was acceptable. What was unacceptable a year earlier became normal a year later – what Diane Vaughan called the normalisation of deviance. Reading the signals required a mechanism JR West did not have, and that few commercial operators do.
A different industry, the same architecture
During the same period, in a different industry on the other side of the world, the Mid-Staffordshire NHS Foundation Trust was running a failure of the same kind. The Department of Health had set aggressive bed-management targets. Over years, the trust's management enforced them with career-ending consequences for senior staff who missed them. The proclaimed priority was patient care. The enforced priority was throughput.
The targets had teeth. Failure to meet them meant loss of Foundation Trust status, board reorganisation, and dismissal of senior staff. The trust was being measured on throughput and rewarded on throughput. The mechanism was similar to JR West's: a metric that mattered, enforced through career consequence, applied without regard to the operational reality it was distorting.
The 2013 Francis Inquiry documented standards of basic care that had deteriorated: patients dehydrated, calls for help unanswered, people left in their own waste. Some died in conditions that were visible to the clinical staff around them.
Staff who raised concerns were not just ignored. Helene Donnelly, a nurse in Accident & Emergency (A&E), reported falsified records and was warned to “watch her back” in the car park. Doctors who raised concerns about staffing were treated as the problem. The trust had formal governance – board reporting, clinical audit, regulatory oversight. Somehow, those governing did not see what frontline staff did.
The mechanisms differ from JR West. Theirs was a designed disciplinary instrument, while Mid-Staffs’ was target pressure operating through career incentives. The instruments differ, but the blindness is the same.
What governance cannot see
What both organisations lacked was the ability to read weak signals – patterns which are below the threshold of action, but too persistent to be dismissed as noise. Investigation is a different function.
Reading weak signals is different again. The information arrives ambiguous, partial, often contradictory, through both official and unofficial channels. Hypotheses only emerge when the information is interrogated into a pattern. These do not always pay off, but when they do, they matter disproportionately. The output is not a finding, but a question worth pursuing.
This is not a discipline audit, as it is typically constituted, is designed to perform. The skill is different. The evidentiary standard is different. The temporal rhythm is different. Audit reports on what has happened, while weak-signal reading attends to what is happening. The reward structure is different: audit succeeds when findings are closed; weak-signal reading succeeds when patterns are detected before they require closure.
Karl Weick and Kathleen Sutcliffe’s research on high-reliability organisations, though debated in its predictive claims, describes how nuclear power plants, aircraft carriers, and air traffic control operate in environments where failure would be catastrophic, and yet maintain exceptionally low failure rates. They call this property mindful organising: the practice of treating small anomalies as information rather than noise. Five principles characterise it: preoccupation with failure, reluctance to simplify, sensitivity to operations, commitment to resilience, and deference to expertise.
JR West exhibited the inverse of each. Near misses were treated as driver failings requiring re-education, not as system signals requiring investigation. The single explanation for lateness was bad drivers, and the single intervention was punitive retraining. Senior management had little direct contact with driver decision-making under time pressure. The system was designed to prevent lateness, not to recover from it. Drivers' operational judgement was systematically overridden by timetable pressure.
The signals were legible. The function to read them did not exist. Not because no one was paying attention, but because the organisation had no role whose mandate was to notice patterns, no authority to escalate them, no measure of success that rewarded detection over closure, and no protection for the people who escalated and were wrong. Equipment on JR West's own trains had been logging emergency-brake activations from a configuration error, roughly three times a day, for years. The commission found the company “could easily have known.” The data had been captured. No one had been tasked with reading it.
One hundred and seven people died at a curve the system had been signalling for years. The signals had been legible to anyone close enough to the operation, but none of them governed it.
Most governance functions lack what JR West lacked: a role whose mandate is to read patterns, with authority to escalate them, and protection for being mistaken. Most have not had the failure that reveals the absence. When it comes – and at some point it will – the question is whether the function has been built, and whether the people who escalate face career consequences when their concerns prove unfounded.
The second question is the harder one. It requires distinguishing structurally, between honest pattern-reporting that turns out to be wrong, and reckless behaviour that warrants accountability. The first must be protected. The second must not. How to design that distinction is the subject of a companion piece.
Comments