The Right Review Cadence Prevents Drift

Processes mutate over time, so review needs the right frequency to keep them true.

A small manufacturer of precision metal parts has a procedure for setting up one of its older lathes between jobs. It is written down. There are eleven steps, and when the procedure was authored, six years ago, every one of them earned its place — each step was there because, at some point, skipping it had produced a scrapped batch, a damaged tool, or an injured hand.

For the first year the procedure is followed exactly, because the man who wrote it stands at the machine. Then he is promoted to run the shop floor, and the setup passes to the people he trained, and the people they train, and the procedure begins, very slowly, to live a second life that nobody intends and nobody notices.

Step four says to let the spindle run for two minutes before loading a workpiece, to bring the bearings up to temperature so the tolerances hold. But the shop is busier now, and two minutes is two minutes, and someone discovers that ninety seconds is usually fine, and then that sixty seconds is usually fine, and “usually fine” quietly becomes the rule. Step seven says to check a particular clamp with a torque wrench. The torque wrench lives in a drawer two benches away, and one day it is in use when someone needs it, so they tighten the clamp by feel, and it holds, and feel becomes the method. Step nine, a verification step, was always a little redundant — it checked something step six had mostly already caught — so it is the first to fall away, and nobody can quite remember when. None of this is laziness. Every single change is a sensible local adaptation by a competent person solving a real, immediate pressure: a queue, a missing tool, a redundancy that looked like waste. Each one, on its own, was probably right.

And the written procedure on the wall still says eleven steps. It has not changed a word. What has changed is the gap between the document and the doing — a gap that did not exist six years ago and now is wide enough to drive a fault through. The procedure people believe they are following and the procedure they are following have come apart, silently, one reasonable shortcut at a time.

Then a new quality manager, in her first month, does something almost nobody thinks to do: she stands at the lathe with the written procedure in her hand and watches an actual setup, step by step, and ticks off what really happens against what is written. She finds the drift in twenty minutes. Some of it she keeps — step nine genuinely was redundant, so she removes it from the document on purpose, with a reason recorded. Some of it she restores hard — the spindle warm-up was load-bearing, and she can show the scrap data to prove it. The procedure that comes out the other side is true again: it describes what is actually done, and what is actually done is, once more, what should be done. Then she puts a recurring date in the calendar to do it again — not constantly, but at a frequency matched to how fast this particular shop changes — because she understands that she has not fixed the drift. She has only caught up with it, and it will start again the moment she walks away.

That recovery is not luck, and it is not a story about one diligent person. It is a specific mechanism doing what it reliably does, and it is one of the quiet engines under most processes that stay trustworthy for years.

The Principle

Every living process drifts from its original intent over time — not through malice or neglect but through the steady accumulation of small, sensible local adaptations — so it stays true only if someone reviews it at a cadence matched to its actual rate of drift. The skill is not the reviewing. It is calibrating the frequency: often enough to catch drift before it sets, rarely enough not to thrash.

The comfortable assumption is that a process, once designed well and written down, stays designed. You build it carefully, you document it, you train people on it, and from then on it does what it was built to do until you deliberately decide to change it. The document is the process. Change is an event you author; absence of an authored change means absence of change.

That assumption is the mistake, and it is a quiet one because it is wrong in slow motion. A process is not the document; it is what people actually do, and what people actually do is under constant, gentle pressure to deform — toward whatever is faster, easier, less effortful, or more locally convenient in the moment. Each individual deformation is small and reasonable and invisible. But they accumulate in one direction, because the pressures that cause them are persistent and pointed the same way, and the document does not deform with them, so the gap between the written intent and the lived reality opens steadily and silently. Review is the only force that closes that gap. And review has a frequency, which means it can be wrong in two directions: too rare, and the drift compounds into a new normal before anyone looks; too frequent, and you spend the organisation’s attention re-litigating a process that has barely moved, and you exhaust the very people whose buy-in keeps it alive. The craft is reading the rate of drift and setting the cadence to match it.

Why It Is Inevitable

This is not a flaw in lazy organisations that better ones avoid. It is a consequence of how processes and the people running them actually interact, and it pushes anyone who looks closely toward the same conclusion.

The first force is that local pressure is constant and one-directional, while the written process is static. Everyone executing a procedure is under real-time pressure — time, effort, missing tools, a queue, a redundancy that looks like waste — and the natural human response to pressure is to adapt the procedure locally to relieve it. Crucially, those adaptations nearly all push the same way: toward less effort, fewer steps, faster throughput, the path of least resistance. They are not random noise that cancels out; they are a current with a direction. So a process left alone does not stay still and it does not wander aimlessly — it migrates, steadily, away from the costly-but-important and toward the cheap-but-convenient. The document, meanwhile, sits on the wall and migrates nowhere. The gap can only widen.

The second force is that the adaptations are individually rational and invisible, which is exactly what makes them dangerous. If drift were caused by carelessness, you could catch it by catching the careless. But each shortcut is taken by a competent person for a good reason, and each one, in isolation, is usually fine — that is the whole trap. “Usually fine” is a teacher that only ever rewards the shortcut, because the cost of having dropped the step does not arrive on the day you drop it; it arrives later, rarely, and by then the link between the dropped step and the failure is invisible. So the feedback the operator gets is uniformly positive right up until it is catastrophic. Nobody is doing anything wrong on any given day. That is precisely why nobody notices, and why the drift cannot correct itself from the inside.

The third force is the one that makes this an argument for cadence rather than just for vigilance: drift has a rate, and the rate varies, so a single fixed answer is always wrong somewhere. A process in a fast-changing area — new people, new tools, new pressures every quarter — drifts quickly, and reviewing it once every few years guarantees you only ever meet a process that has already become something else. A genuinely stable process reviewed every week, by contrast, wastes everyone’s time and trains people to treat review as theatre, so that when it finally matters they have stopped paying attention. Neither “review constantly” nor “review once and trust it” survives contact with reality. The only thing that does is matching the frequency of looking to the speed of drifting — which means review is not a one-off act of design but a maintenance rhythm someone has to set, tune, and keep tuning.

How It Shows Up

• The written procedure and the lived procedure are noticeably different, and the gap opened gradually rather than at any single identifiable moment — no one can say when step nine stopped being done.
• Every individual deviation, examined on its own, turns out to have been a reasonable response to a real local pressure; there is no villain to find, only a direction everything has quietly leaned.
• The drift is uniformly toward the cheaper, faster, lower-effort version — the costly verification steps and the slow careful ones fall away first, and the convenient ones survive.
• The process is reviewed on a deliberate, recurring schedule whose frequency is set to how fast that particular process actually changes — fast-moving things are looked at often, stable things seldom, and someone chose those intervals on purpose.
• Review compares the document to the observed reality of what people do, not the document to itself or to what people say they do — someone goes and watches.
• When the review finds drift, the response is sometimes to restore the original step and sometimes to formally adopt the change into the document — drift is treated as information, not simply as error to be reversed.
• People can tell you when the process was last reviewed and roughly when it will be again, because the rhythm is a known, designed thing rather than an event that happens when something has already broken.

Why It Causes Benefit

When a process is reviewed at a cadence matched to its drift, an environment gets something that looks, from outside, like a kind of effortless reliability: its procedures keep meaning what they are supposed to mean, year after year, without anyone seeming to fight for it.

It stays reliable because the gap between intent and reality is never allowed to grow past the point where it can do damage. Drift is caught while it is still small — while it is one dropped step rather than a whole mutated procedure, while the link between the shortcut and its risk is still visible and recoverable. The important-but-costly steps, which are exactly the ones local pressure erodes first, are restored before their absence has a chance to find the rare day on which they mattered. The reliability does not depend on every operator resisting every pressure forever, which is impossible; it depends on a structural rhythm that periodically pulls the lived process back into line with the intended one. That is a far more durable safeguard than asking people not to drift, because it works with the certainty that they will.

And it stays efficient, which the naïve answer — review everything constantly — does not. By matching frequency to the actual rate of change, the organisation spends its scarce attention where drift is fast and spends none where drift is slow. It neither lets the volatile process rot nor thrashes the stable one. There is a quieter benefit too: because review treats drift as information rather than as misbehaviour, it learns from it. Half the shortcuts turn out to be improvements the front line discovered and the document should adopt; the other half are erosions that should be reversed. A well-cadenced review harvests the first and corrects the second, so the process does not merely hold its line — it gets genuinely better, incorporating the real intelligence of the people running it while filtering out the decay. The procedure on the wall becomes, over years, not a fossil of one person’s original intent but a living, accurate record of the best current way to do the thing. That is the compounding payoff: a process that stays true and keeps improving, because someone built a rhythm to keep checking it against the world at the right speed.

How To Cultivate It

• Review the process against observed reality, not against itself. Go and watch the thing actually being done, with the document in hand, and tick off what happens against what is written. Asking people whether they follow the procedure will get you the procedure they believe they follow, which is precisely the one that has drifted out from under them without their knowing.
• Set the cadence to the rate of drift, not to a calendar convenience. Estimate how fast this particular process changes — how much new staff, new tooling, new pressure it absorbs — and review fast-moving ones often and stable ones seldom. Resist both the false economy of “we’ll review everything annually” and the thrash of reviewing the unchanging. The interval is a design choice; make it deliberately and write down why.
• Tune the cadence over time using what you find. If a review turns up almost no drift, you are looking too often — lengthen the interval. If it turns up a great deal, you looked too late — shorten it. The right frequency is discovered by observing how much has moved between looks, so let the findings calibrate the rhythm rather than fixing it once and forgetting it.
• Treat drift as information, not just as error. When a review finds a deviation, ask which kind it is before reflexively reversing it. Some shortcuts are erosions of load-bearing steps and must be restored, with the reason made visible. Others are genuine improvements the front line found, and the right move is to adopt them into the document on purpose. A review that only ever restores throws away the intelligence of the people doing the work; one that only ever adopts ratifies the decay. Do both, deliberately.
• Make the original reason for each step survive alongside the step. Drift accelerates when nobody remembers why a step is there, because a step with no visible reason looks exactly like waste and gets cut first. Recording the intent — the scrapped batch, the injury, the failure it prevents — gives the next reviewer what they need to tell a load-bearing step from a redundant one, and protects the costly-but-important from being quietly optimised away.
• Build the rhythm as a standing maintenance task with an owner, not a project that ends. The single most common failure is that someone heroically corrects the drift once, declares the process fixed, and walks away — at which point the same pressures begin the same migration again. The correction is not the fix; the recurring review is the fix. Design the cadence, give it a name and an owner, and protect it the way you would protect any other piece of maintenance.

What Good Looks Like

An environment whose written procedures still describe what people actually do, years after they were authored — not because nothing ever changed, but because someone deliberately keeps closing the gap that change opens. Where every process that matters has a review rhythm whose frequency is matched to how fast that process drifts: the volatile ones looked at often, the stable ones seldom, and the intervals set on purpose and tuned by what the looking reveals. Where review means going and watching the real work against the real document, so the silent gap between the believed process and the lived one gets found while it is still small. Where the drift that turns up is read as information — half of it improvements to adopt, half of it erosions to restore — so the process does not just hold its line but quietly gets better, absorbing what the front line learned and shedding what it let slip. Where the original reason for each step is remembered, so the costly-but-important survives the constant pressure toward the cheap-but-convenient. And where nobody mistakes the one-off correction for the cure, because the cure was always the rhythm. From outside it looks like luck, or like an unusually disciplined workforce: how do their procedures stay so accurate when everyone else’s quietly rot into fiction? They stay accurate because someone reads the rate of drift, sets the cadence to match it, and keeps the appointment.

A Reflective Question

Pick a process in your environment that hasn’t been reviewed in a while — and ask honestly: how confident are you that what is written down is still what is actually done? If you went and watched it tomorrow, document in hand, do you think you would find the two had quietly come apart? And if you are not confident either way, what does that uncertainty tell you about whether your review cadence is matched to your rate of drift — or whether you have a review cadence at all?