Glance metrics constrain design; they are not acceptance tests
Aliases: design-time glance limit · NHTSA glance criteria · acceptance versus constraint
What it is
Caps on single glances and on total eyes-off-road time are constraints for cutting tasks and changing structure while designing, not an acceptance gate that, once a sample passes after launch, means the system is safe. Duration numbers in visual-manual guidance—including the often-quoted ceiling of about two seconds for one glance—are a ruler for deciding that a flow does not yet belong in motion. A trained panel staying under the line in a lab shows only that, under that protocol, the task was made measurable and short. It does not show that every driver, on every road, is safe.
Why it happens
Glance metrics describe the shape of a task’s visual demand so options can be compared: should this list become voice, should this step wait until parked, should one confirmation be split. Once the same numbers are treated as acceptance, the incentive flips. People train the task to the line instead of asking whether the task should exist. Participants know the flow, know they are being measured, and meet scheduled events, so glances run systematically short. A closed course has no city two-wheelers and no sudden pedestrians; the same task stretches again on a real road. A pass says “it did not blow up in this protocol.” Long glances on the road come from missed taps, failed search, glare, a lead vehicle that slows without warning—none of which sit in the sampling script.
Studying it
Put the same task in three ecologies and compare: laboratory simulator, closed course, naturalistic or instrumented-vehicle driving. Watch how the glance distribution widens, rather than scoring pass/fail against one guidance document. The protocol itself is an independent variable: practice allowed or not, whether people are told to “watch the road,” whether events are predictable.
Independent variables: test setting (simulator / closed course / public road), amount of practice, whether participants know the criterion. Dependent variables: distributions of single and total off-road time, how often the line is crossed, whether that agrees with the protocol verdict.
The numbers in guidance come from a particular experimental tradition and a particular traffic assumption. Change country, speed band, or cabin geometry, and the same figure means something else. Do not promote a threshold from one edition of a visual-manual document into a physical fact and stamp every in-vehicle task with it.
Where it stops holding
Type approval and regulatory conformity have their own pass lines. Those are legal procedures, not a human-factors finding of “already safe.” Using glance metrics as an internal gate during development is sound—if the gate kills tasks, rather than certifying the ones that remain as harmless. High automation, low-speed valet parking, and off-road worksite vehicles do not match the speed and event assumptions in those protocols; the numbers do not copy over. Without eye tracking, do not pretend glances were already “tested.”
Applying it
- Spend the glance constraint as subtraction in the concept phase: delete the over-line step, move it to voice, or hold it until parked. Do not keep it for a test week of parameter tuning to scrape under the line.
- Treat guidance numbers as a comparison ruler, not a certificate. Report distributions and ecology. Do not write “meets N seconds, therefore shippable.”
- If a lab evaluation is required, admit that closed conditions run short, and treat the long-glance tail from public roads or naturalistic driving as a veto, not as an outlier to drop.
- Verify by asking what design decision the number changed. If it cut no task and only appears in a test report, it is already being used as acceptance, not as a constraint.