Abhishek S.
Shipping in public. Listening in private.

Abhishek

I lead women’s Indo-Western & Premium at Max Fashion. I also wrote the AI that runs the buying floor.

Rare profile. Category operator who ships production code.

Senior Buying Leader · Max Fashion Women’s Indo-Western & Premium · 530+ India stores NIFT ’12 · Twelve years on the floor

abhishek@bengaluru ~ %
>role: senior buying lead
>dept: women’s indo-western + premium
>floor: 530+ stores india

Poka-Yoke

A 3.5-inch floppy disk is a tiny argument against blaming humans: it only enters the drive one way because the object refuses the wrong action. Shigeo Shingo named this idea poka-yoke in Toyota production work in the 1960s, after rejecting the harsher phrase baka-yoke, or fool-proofing. The point is not that people become careful. The point is that the system catches the slip before it becomes a defect.

How it works

Poka-yoke treats error as a design input. If a part can be installed backwards, shape the mating part so backwards is impossible. If a worker can skip a step, make the next step physically or electronically depend on the skipped one. If a machine can run without material aligned, add a sensor that stops it before scrap accumulates.

Shingo separated defects from errors. An error is the human or machine action: wrong orientation, missing part, missed torque. A defect is the bad unit that reaches the next process. Poka-yoke tries to intercept the first before it hardens into the second.

The sharp line is this: training asks memory to carry quality; poka-yoke moves quality into the fixture, connector, sequence, or constraint.

Where it shows up

Object or process Error absorbed Design move
3.5-inch floppy disk Wrong insertion direction Asymmetric shell and shutter
USB-C connector Upside-down plug insertion Reversible connector geometry
Microwave door switch Running while open Interlock breaks the circuit
Fuel nozzle diameter Diesel in petrol car, or reverse Different nozzle and filler sizes by market
Factory pick-to-light rack Wrong part picked Light points to the bin; scan confirms pick

The same idea appears in software. A destructive command asks for a repository name, not just "yes." A form refuses an impossible date. A database uses a foreign key instead of trusting every caller to remember the relationship. The best version is boring: the user never sees the mistake because the path never opens.

What's contested

The live argument is not whether mistake-proofing works. It does. The argument is where constraint becomes brittleness. A hard interlock can prevent a bad state, but it can also block rare recovery work unless the override path is designed with equal care.

There is also a human-factors trap. If every obvious slip is absorbed by design, the remaining failures are rarer, stranger, and harder to rehearse. James Reason's 1990 Swiss cheese model is useful here: barriers reduce risk, but barriers also create new interfaces where hidden assumptions collect.

Why this touches other realms

Poka-yoke is engineering's version of epistemic humility. It starts from the same place as concept bayes theorem: do not pretend the observer is perfect. Build the update, the prior, or the correction into the system.

That is why it belongs near concept shannon information theory too. A keyed connector is a physical encoding scheme. It removes invalid messages from the channel before anyone has to interpret them. The 3.5-inch floppy disk is not just storage history; it is a little theorem about reducing possible states.

Abhishek's take

What grabs me about poka-yoke is how anti-heroic it is. The good solution does not ask for better discipline, a sharper memo, or another review meeting. It changes the object so the wrong move has nowhere to land.

Where I've used this

I use this pattern on the buying floor when a workflow can quietly accept a bad input. A blocked state, a forced comparison, or a required scan beats a postmortem after the error has already traveled through three people.

Tags: #mistake-proofing #toyota-production-system #design #quality #human-error

Key sources

Further reading

See Also