The most reliable error-reduction technique in industrial safety might also be the most visually undignified. It’s not a checklist. It’s not a new sensor. It’s not even a procedure in the modern sense. It’s pointing at things while saying out loud what you see.
That’s the entire intervention. And as far as anyone has measured, it cuts operator errors by roughly 85 percent.
What it looks like
If you’ve ridden the Tokyo Metro, you’ve seen it. The conductor leans out of the cabin window. He scans the platform. He sees a green signal. He extends his arm, points directly at it — index finger leading, the gesture sharp and theatrical — and announces “shingo yoshi” — “signal good.” He points at the platform clock. “Jikan yoshi.” He points at his speedometer. “Sokudo yoshi.” He’s not performing for tourists. He’s running a one-person crew resource management protocol that the Japanese rail industry has been refining since 1913.
The technique is called shisa kanko (指差喚呼) — “pointing and calling.” Adopted in stages across Japan Railways throughout the early twentieth century, formalized as procedure in the postwar period, and exported to a handful of other industrial contexts inside Japan: steel mills, nuclear plants, hospital surgery teams. Outside Japan, it has spread to a thin, slightly embarrassed slice of the global rail industry. New York City Transit picked up a partial version in the 1990s — conductors point at a black-and-white striped board on each platform to confirm the train has stopped at the correct mark. London and Paris have not. Most of the rest of the world has looked at it, recognized that it works, and declined to adopt it on what amounts to aesthetic grounds.
The most-cited number is from a 1996 study by Japan’s Railway Technical Research Institute, which compared operators performing simple tasks with and without the protocol. Errors dropped from roughly 2.4 per 100 operations to 0.38 per 100. That’s the 85 percent figure, achieved with no equipment, no software, no training beyond demonstrating the gesture and rehearsing the call.
I went into this expecting the 85 percent number to be apocryphal — the kind of management-consulting statistic that gets cited everywhere because nobody can actually find the original paper. The number is real. The study design is unimpressive by modern cognitive science standards (small samples, narrow tasks, no preregistration), but the effect has held up under more rigorous follow-ups. A 2011 study by Shigemori et al. on laparoscopic surgery teams using a modified pointing-and-calling protocol showed measurable reductions in both omitted steps and confirmation errors. A 2017 review of the Japanese nuclear industry documented sustained reductions in instrumentation read errors over decades of use. The intervention is not a curiosity. It is robustly effective.
Why it works
The mechanism is interesting precisely because it does not look like one. Modern ergonomics would predict that adding extra steps to a procedure would increase cognitive load and produce more errors, not fewer. Pointing-and-calling adds a motor gesture and a verbal utterance to a task that could, in principle, be performed silently and motionlessly. By the standard logic of mid-century industrial design — minimize unnecessary motion, eliminate verbal redundancy, design for the lowest-effort path — it should be a regression.
It is not a regression because the standard logic was wrong about what causes operator errors. The dominant failure mode in routine industrial tasks is not cognitive overload. It is under-engagement: the trained operator entering a near-automatic state in which the procedure is performed without genuine attention, and small deviations or anomalies are not noticed because nothing is actually being processed. Highway hypnosis. The pilot who has flown the same approach a thousand times and misses the gear-down call. The nurse who scans the wristband but doesn’t see it.
Pointing-and-calling forces three sensory and motor channels to converge on the object of attention simultaneously: the eyes track to it, the hand extends toward it, the mouth produces a verbal label of what the eyes are seeing. This is not redundancy in the engineering sense, where multiple parallel channels confirm the same signal. It is binding in the cognitive science sense — the act of constructing an integrated representation of an object across modalities. Empirically, perception that recruits multiple modalities is more durable, more verifiable, and harder to confabulate than perception that recruits only one.
The protocol also exploits a prediction-error mismatch that the verbal call introduces. If the conductor sees a red signal but his trained automaticity has him expecting green, the moment he tries to vocalize “shingo yoshi” — “signal good” — there is a microsecond conflict between what his eyes are reporting and what his mouth is preparing to say. The conflict is detectable. The error is caught at the verbal stage, before it reaches the motor stage where the train would already be moving.
This is the same architecture that successful aviation crew resource management protocols have been groping toward since the late 1970s — challenge-and-response, callouts, sterile cockpit. But aviation built its protocols around two-person verification, where one operator’s error is caught by another. Pointing-and-calling is the same architecture compressed into a single operator. The conductor verifies himself.
Why it hasn’t spread
The honest answer is that pointing and shouting at signals looks ridiculous to a Western industrial culture that has spent a century pricing dignity into job design. In an American or European workplace, the engineer who proposed that locomotive operators loudly announce “speed: good” while pointing at the speedometer would be told that the operator is a professional, not a child, and that the implication of unprofessionalism would damage morale. The protocol is incompatible with a workplace ethos that treats explicit verification as a reflection of inadequate trust in the worker.
Japan resolved this aesthetic problem differently. The protocol entered the rail industry early, before the postwar emphasis on individual professional dignity calcified, and was institutionalized as a marker of seriousness rather than infantilization. The conductor who points and calls is signaling discipline, not deficiency. By the time the rest of the world considered adoption, the cultural framing had ossified in the opposite direction, and importing the procedure would require importing the framing — which, empirically, no one has been willing to do.
There is also a problem of measurement. The 85 percent error reduction is a real number, but it is real only when the protocol is performed in the trained way: sharp gestures, definitive verbalizations, deliberate engagement of attention. A bored or resentful operator going through the motions of pointing while muttering the call achieves none of the cognitive binding that makes the protocol work. Adoption attempts in Western industries have repeatedly produced the bored version of the protocol, observed that it produces no measurable benefit, and concluded — correctly, given the data they collected — that pointing-and-calling does not work for them. What they have not measured is that the protocol they adopted was not, in any cognitively meaningful sense, the same protocol.
The general lesson
Most error-reduction interventions assume that errors come from missing information, and respond by adding information. More sensors. More alarms. More dashboards. More checklists. The pointing-and-calling result suggests that a non-trivial fraction of routine errors comes not from missing information but from unbound information — perceptions that arrive at the operator’s eyes but are never integrated into a representation that the operator can verify or act on.
The intervention that addresses unbound information is not technical. It is embodied. It involves engaging the operator’s body in the act of perception in a way that forces the perception to become explicit, integrated, and verifiable. There is no software equivalent. You cannot point-and-call through a heads-up display. The thing the protocol does — closing the loop between perception, motor act, and verbal commitment — is something that happens in a body or it doesn’t happen at all.
This is uncomfortable for a technological culture that prefers solutions which can be procured rather than performed. The buyable interventions in the error-reduction market — sensor fusion systems, AI-assisted alerting, voice-activated checklists — sit on shelves. The non-buyable intervention, which empirically beats most of them on most routine tasks, is standing on the platform of every Japanese train station, in plain sight, costing nothing.
I don’t think pointing-and-calling generalizes everywhere. It works best for tasks that involve verifying discrete states (signal, speed, time, position) at known checkpoints in a procedure. It works worst for tasks that involve continuous monitoring or open-ended judgment. But within its operating envelope — which covers most of routine industrial control — it is, by the available evidence, the most effective single intervention anyone has documented.
What’s interesting isn’t that it works. It’s that we have known it works for thirty years and chosen not to use it, because to use it correctly would require admitting that the body knows things the procedure manual cannot encode — and that admission, more than the gesture, is what most engineering cultures cannot abide.