A limit, a threshold, a time is set to something nobody can explain. It was changed once for a reason, and the reason left with whoever changed it.

A setting is at an odd value. Not wrong exactly — just not what anyone would choose.

Somebody changed it, once, for a reason. Nobody knows who, when, or why.

Where these live

Time limits and thresholds. How long a booking holds, when something expires, at what point a warning appears.

Permissions. Who can do what — ⚠️ frequently widened once for a specific situation and never narrowed again.

Automatic behaviour. What happens on a failure, how many retries, what gets suspended.

Templates and standard text. Edited for one occasion and left as the default.

Why they drift

Changing is easy; recording is not. ⚠️ The change takes a click. Writing down why takes a decision about where to write it, and there usually is no where.

It was urgent. Changed to solve something at 22:00, and the situation passed.

The person left. The most common single explanation.

Nobody reviews settings. They are set up once, at the start, and then they are furniture.

What it costs

Behaviour nobody expects. Something happens on a rule nobody remembers setting, and it is investigated as a fault.

Changing it feels risky. ⚠️ Because nobody can say what depends on it — so the odd value stays, indefinitely, defended by uncertainty.

Two venues drift apart. In a multi-site operation, the same setting ends up different in each, and the numbers stop comparing.

The annual read-through

Once a year, an hour.

List the settings that affect behaviour. ⚠️ Not every option — the ones with consequences: limits, timings, permissions, automatic actions.

For each, ask: is this what we would choose today? Not "why is it this" — nobody knows. The useful question is forward-looking.

Change the ones that are clearly wrong. Note the ones you are unsure about, and ask before the next review.

Recording it going forward

Two fields at the moment of changing: who, and why. ⚠️ The why is the one that matters and the one always skipped.

"Changed the hold time to 30 minutes because of the Saturday queue" is a sentence somebody can evaluate in two years. "30" is not.

Where there is nowhere to record it, that is the fix — a shared note beside the setting is enough, and it does not need to be sophisticated.

The permissions case

⚠️ The one worth doing first.

Access widened for a situation is almost never narrowed afterwards. Over years, everyone accumulates everything, and the permission structure stops describing anything.

Review who can do what, once a year. It takes twenty minutes and it always finds something.

A setting is a decision with no author

A written rule  has a date, a reason, and someone who wrote it
A setting    has a value, and nothing else

⚠️ The value is doing the same work as the rule, with none of the context — which is why it survives long after the reason for it stops being true.

The question that has no answer

⇒ ⚠️ "Why is it fourteen days?" Nobody knows. It has been fourteen days since before anyone currently here arrived, and it is now treated as a property of the world rather than a choice someone made.

The three ways they drift

Changed for one night  and never changed back
Changed by an update  a default moved, and nobody was told
Never changed at all  the original default was never a decision

⚠️ The third is the largest category by far. Most settings have never been chosen — they arrived as defaults and were simply never touched.

Which makes the audit easier than it sounds

⇒ ⚠️ You are not looking for who changed things. You are asking, of each value, whether anyone would choose it today. Most of them nobody would.

What a wrong value costs

A cut-off that is too short  work disappears before anyone finishes it
A cut-off that is too long  stale entries look current
A threshold set for old volumes fires constantly, and gets ignored

⚠️ The third cost is the worst, because it teaches people that alerts mean nothing — and then a real one arrives.

The read-through, once a year

List them  most systems will show you every setting on one screen
For each one Would we choose this today? Yes / No / Nobody knows

⚠️ "Nobody knows" is the useful answer. It marks the ones where the reason has been lost, and those are exactly the ones to decide again.

Two hours, and bring the person who uses it

⇒ ⚠️ Do not do this alone from the admin screen. The person working with the consequences of a value knows things the screen does not show.

Leaving a trail for the next person

Changed the value  and wrote nothing  you are back here next year
Changed the value  and wrote one line  the next person can undo it safely

⚠️ One line: what, from what to what, why, and the date. 　 ⇒ It takes thirty seconds and it is the entire difference between a 　　 setting and a decision.

Permissions deserve their own pass

Everyone can change it  it will drift
Two people can change it and it is written down  it will not

⇒ ⚠️ Restrict the settings that are expensive to get wrong, and leave the rest open. Locking everything produces workarounds, which is worse than drift.

Three to hold

Settings you cannot explain. ⚠️ Ask the longest-serving person to walk the list with you — they will remember some, and the ones neither of you can explain are the finding.

Changes made in the last year with a recorded reason. Usually near zero, and that is the habit to change rather than the settings themselves.

Differences between sites, if you have more than one. ⚠️ Each one is either deliberate or drift, and nobody has ever separated them.

Where the record has to sit

A setting and its reason have to live together, or the reason evaporates within a year. It does not need a system — it needs somewhere the next person will look when they wonder why the number is what it is.

tasteck keeps operational notes on the staff-facing screens, so a reason has somewhere to sit beside the work it governs.

Nothing is misconfigured. Every value was chosen deliberately, once, and the deliberation is what got lost.

The Setting Nobody Remembers Changing