Cal Newport's new proposal for cognitive fitness has something most advice about attention lacks: a set of things you can actually score.

In an August 24 episode of Deep Questions, Newport suggests three key performance indicators. Reading endurance is the number of minutes you can stay with demanding but understandable nonfiction without switching context. Contemplation control is how long you can keep making progress on a real problem in your head while walking, repeatedly bringing attention back when it wanders. Written persuasiveness asks you to argue for a contested position and judge the quality of the resulting argument.

His reading scale labels five minutes or less “brain-rotted,” about 15 minutes “distractible,” 30 minutes a “pre-2012 average,” and 75 minutes “impressive.” For contemplation, the proposed anchors are less than a minute at the low end, one to three minutes as “distractible,” 10 minutes as the “pre-2012 average,” and 30 minutes as “impressive.”

Those labels make the tests vivid. They should not be mistaken for population norms. Newport presents the historical averages as estimates, not values drawn from a validated pre-2012 dataset.

The third KPI has a different measurement problem. There is no stopwatch for persuasiveness. Newport's scale moves from disorganized, emotional argument through reasonable bullet points, then to good-faith prose argument and finally to arguments built powerfully from first principles. A usable version would need a stable rubric and some evidence that different raters score the same writing similarly. If you score your own argument, there is another problem: the person being measured is also judging whether the performance was good, which makes the score vulnerable to the same metacognitive errors the test is supposed to reveal.

Cognitive performance really can be measured

Psychologists measure attention, memory, processing speed, reasoning, and many other cognitive performances. The useful question is not whether cognition can be measured at all, but whether a particular score measures the capability we think it does.

A 2025 randomized crossover experiment offers a more standardized example. Four hundred sixty-seven U.S. and Canadian participants agreed to try blocking mobile internet on their iPhones for two weeks while leaving calls, texts, and desktop internet available. Of those, 266 set up the blocker and 119 met the preregistered compliance criterion. The primary analyses nevertheless followed an intention-to-treat approach, comparing people according to assignment rather than only the most compliant subgroup.

The researchers measured sustained attention with the gradual-onset continuous performance task, or gradCPT. Sustained attention improved modestly during the intervention, with reported within-person effects around 0.24 to 0.28 standard deviations in the two intervention sequences. Mental health and subjective well-being also improved.

The study does not validate Newport's reading test, and the recruitment and compliance pattern matters when deciding how broadly to generalize it. It does show what a more serious measurement claim looks like: a defined task, a randomized intervention, preregistered criteria, and enough detail to distinguish assignment from actual adherence.

What happens when you train the score?

For an individual trying to recover the ability to read for long stretches, a stopwatch may be perfectly adequate for a narrower purpose. If the question is “Can I now read this kind of book for longer without reaching for something else?”, the measure is close to the capability itself.

The comparison still has to be honest. Twelve minutes with a dense philosophy book in a noisy airport and 35 minutes with an accessible history book at home are not clean evidence of improvement. Use material of similar difficulty, similar conditions, and the same rule for what counts as breaking concentration.

Newport goes further when he argues that pushing these indicators upward should improve cognitive health more generally. That claim runs into the transfer problem.

A 2019 second-order meta-analysis examined evidence from working-memory training, video games, music, chess, and other cognitive-training programs. Near transfer was common: people improved on trained tasks or close relatives. Far transfer to substantially different cognitive abilities was small or absent, and in the authors' bias-adjusted analyses it fell to essentially zero.

That literature does not test Newport's exact exercises, so it cannot tell us that sustained reading has no broader benefits. It does tell us not to treat the physical-fitness analogy as evidence by itself. The same distinction appears in AI tutoring research, where retaining a taught method and transferring it to a new problem can separate.

If the target is reading endurance, keep the material and conditions reasonably comparable and track your own baseline. A rise from 12 to 35 minutes answers a useful narrow question.

Sources

Ulix app

Track Analysis

Track Analysis is a freeform Android logger for recording timestamped observations in plain English and exporting clean CSV files for later analysis.