A reasonable trust level varies with task and situation; there is no globally correct trust

What it is

The same spellchecker can be almost fully trusted in mail to a colleague, and cannot be in a filing for court. A reasonable level follows the task. There is no globally correct answer to “how many percent should one trust this product.” There is no globally correct trust: calibration is local; change the task, the stakes, the time, and the right tick changes with them.

Making the product a single “trust us” switch pretends that global value exists.

Why it happens

Reliability itself is not one number. Spelling is stable on common words and loose on terms and names; same engine, two frequencies. Lee and See’s calibration is trust against capability, and capability is conditional. Change the condition and the matching target changes. People, though, like one stable attitude — “I trust it” or “I don’t” — and carry it into the next situation. Attitude is stickier than task, which is how the same person is both over and under on the same system: over in mail, under in the court filing, or the reverse.

If Muir’s update from experience happens on only one task, what forms is trust for that task, not a portable overall score. If the product speaks in an overall score (“user trust 4.6”), local calibration is washed into a global myth.

Studying it

Same system, two tasks (low-stakes daily / high-stakes formal). Measure handover and reliability on that task slice, and see whether people use the same trust level. Independent variables: whether the task switch is made explicit, whether slice base rates are given separately. Dependent variables: calibration error on the two tasks, attitude transfer (carrying trust from task A onto B).

Transfer magnitude is this entry’s core quantity. Reporting one overall trust score hides the transfer.

Where it stops holding

If the system only ever does one task, global and local coincide and this entry weakens. When users have a clear habit of isolating tasks (work account / personal account), attitude stickiness falls. One severe error emptying trust on all tasks is collapse, not the fact that there is no globally correct value. A success streak pushing trust on a single task above that task’s reliability is the next cut.

Applying it

• Do not make a product-level “trust us” badge. Give handover advice by task: “daily drafts can be checked less; check names one by one on outbound documents.”
• In settings, split automation level by task, not one master switch.
• Report calibration by task slice; forbid rolling it into one trust number for external talk.
• Check: ask how much they will check when writing to a colleague versus submitting to a regulator. If the two answers are the same and your two tasks’ reliabilities differ by a stretch, a global attitude has already impersonated local calibration.

Related

• Same group: L5.09.2 A streak of successes pushes trust above the system's actual reliability · L5.09.3 Undertrust shows up as repeated manual checking, whose cost is often ignored · L5.09.4 Showing typical failures can suppress overtrust, at the cost of short-term adoption · L5.09.5 Trust is formed from personal use, not from statements and documentation
• Nearby: L5.03 Trust Calibration · L5.04 Collapse of Trust · L5.10 Asymmetric Effect of First Failures on Trust
• Search terms: no globally correct trust · task-relative trust · trust calibration