Time-on-task diagnosis

Track where hours actually went versus where you thought, so "I studied all day" becomes data you can fix.

When to use this: When something feels wrong
Learning impact: Clear help, modest gain
Learning evidence: Credible with caveats
Where you are with the topic: Checking readiness
How often to use it: Worth repeating often

When to use it

You say “I studied all day” and scores are flat.
You want hours split into retrieval, re-reading, admin, and distraction, not vibes.
You are about to add more hours instead of fixing where hours already go.

What to ask AI

Paste a real-time log with timestamps and what you did. Ask for categories and the biggest mismatch versus your estimate.

Ask for one structural change, not a pep talk.

Refuse post-hoc memory logs dressed up as data.

How it helps

Subjective study time is inflated by being near books. Diagnosis replaces vibes with categories: deep work, shallow review, admin, distraction, looking up, re-reading.

Until you see the pie, you optimise the wrong slice, usually more total hours of the same leak.

AI helps categorise a log and spot patterns. You must capture the raw log honestly for a few days.

How to run it

Log in real time for two or three study days (timer plus short category tags).
Paste the log; ask for totals and surprises versus your estimate.
Name the largest leak.
Change one structural fix (phone out, retrieval before reading, problems before notes).
Re-measure one week later.

Example exchange

You

Here's yesterday's time log with timestamps. I thought I did five hours of solid chemistry. Can you categorise it and show me the biggest mismatch?

Rough totals: retrieval about twenty-five minutes, worked problems forty, re-reading over two hours, video fifty, admin thirty-five, distraction forty. You called a lot of re-reading 'solid.' What's one structural change you'd try tomorrow?

You

Phone out of the room, and start with problems before any notes.

Log tomorrow the same way. We only celebrate if the pie shifts, not if total hours go up.

Copyable prompt

Here is my time log: [timestamps + what I did].
Categorise into: retrieval practice, worked problems, re-reading/highlighting,
video/lecture passive, admin/org, distraction, other.
Compare to my estimate: [ ]. Show the biggest mismatch and ask me for one
structural change, not a pep talk.

The Tell

Pride in long days with flat scores usually means the log would show re-reading dressed up as work.

If you only maximise total hours, you are optimising attendance, not learning.

Principle evidence

Strength of the underlying learning idea, not a claim about AI products.

Moderate: time-on-task diagnosis is basic self-monitoring of study behaviour. Seeing category shares beats trusting “I was at my desk.”

AI delivery evidence

Whether an AI tutor delivers this method well is a separate question.

AI inference of time use from chat alone is weak. It needs your honest timestamped log. Categorisation help is convenience, not validated measurement.