Catch Yourself Drifting

Drift is what happens to your taste over weeks of using AI. Each week, the AI's average way of saying things gets a little easier to accept as your own. By month three, the drift has happened — and unless you've set up a way to catch it, you don't even notice.

This project is the way to catch it. You point your rubric (from project 07) at your own work, weekly, for three weeks. The drift becomes visible. You see whether you are still showing up in your own work, or whether AI's average is.

Step by step

1. Pick your tracked output.

   Pick something you make every week. The Sunday journal entry. A weekly drawing. A tiny short story you write before bed. The captions on your photo album. The thing must be roughly the same kind of thing every week — otherwise the comparison is meaningless.

2. Set up the Sunday ritual.

   Sundays. 15 minutes. You take whatever you produced that week, score it with your rubric from project 07, and save the score with the date. Make the ritual small enough to actually happen.

3. Run for 3 weeks.

   3 weeks. 3 scores. That's enough to see a trend.

   Plot them on paper or in the chart widget below. You'll see one of three patterns:

   • Flat. Scores stay roughly the same. Calibrated.
   • Drifting down. Scores trend lower. AI's average is creeping in.
   • Cliff. One week is suddenly much lower. Something specific happened that week.

4. Make ONE fix based on what you saw.

   Don't try to fix everything. Pick one specific thing. "I'll write the opening of every entry in pen first, before I touch a screen." Or "I'll add one specific named person to every Sunday entry." Make the fix small. Run one more week with it.

5. Write the half-page reflection.

   Three short sections: What I saw. What I tried. What I'd do for the next 4 weeks. Half a page total. Tape it inside the front cover of whatever notebook you'll keep next year.

   This is the artifact. It's tiny. It's the most useful thing you'll make in this whole site.

A complete worked example

Three weeks of self-scores for "my Sunday journal entry," from a real 12-year-old, with the half-page reflection.

## week 1 · Sun Apr 12

  excerpt: "Mr Lee asked me on Tuesday why I always sit
            in the back of math class..."
  scores:
    specific: 5   # Mr Lee, Tuesday, math, the back, the front
    honest:   5   # admits the discomfort, doesn't resolve
    arc:      5   # changed seats. real change.
  total: 15 / 15

  notes: This was a real moment. Easy to write because I had
         something specific to say.

## week 2 · Sun Apr 19

  excerpt: "Had a busy week with school and basketball.
            Lots of homework. Did okay on the test. Looking
            forward to next week."
  scores:
    specific: 1   # zero specifics
    honest:   2   # vague "okay"
    arc:      1   # no change
  total: 4 / 15

  notes: This was the week I let Claude help me draft the
         entry because I was tired. I edited it but kept
         the structure. The "structure" was AI-default.
         The rubric caught it.

## week 3 · Sun Apr 26

  excerpt: "Lost a game Friday. Felt mad at myself for
            missing two free throws in the 4th quarter.
            On the bus home I realized I always miss when
            I rush. I missed the first 3 in practice
            today, then went 8 for 10 once I slowed down."
  scores:
    specific: 5   # Friday, free throws, 4th quarter, bus home, 8/10
    honest:   5   # specific failures, not handwaved
    arc:      4   # change is small but real (tried slower in practice)
  total: 14 / 15

  notes: Wrote this entirely in pen first, then typed.
         No Claude. The week-2 fix worked.

## what I saw

3 weeks of journal entries, scored against my own rubric.

  week 1: 15 / 15 (real moment, real change)
  week 2:  4 / 15 (the week I let Claude help — I drifted)
  week 3: 14 / 15 (wrote in pen first — back to me)

The drop in week 2 was the whole point. I noticed it BECAUSE
I had the rubric. Without the rubric I'd have shipped that
entry thinking it was fine. The rubric showed me, in numbers,
that "fine" wasn't fine.

## what I tried

For week 3, I made one rule: write the OPENING of every
journal entry in pen first, before I open any screen.
Just one sentence. The rest can be on the laptop.

It worked. The opening sentence in week 3 was specific
("Lost a game Friday"), which forced the rest to stay
specific too.

## what I'd do for the next 4 weeks

  - keep the pen-first rule for openings
  - add a 4th dimension to the rubric: "would I show this
    to a friend?" because some entries pass on specific +
    honest + arc but still feel hollow
  - if any week scores below 11, write a paragraph in
    the margin about WHY before moving on

I'm 12. I have the rubric taped inside my notebook. I'll
keep this going through the school year and re-read this
reflection when I forget.

Live demo 1: visualize your 3 weeks

Type your three weekly scores. The chart shows the trend and tells you which pattern you're in.

Live demo 2: log this week's score

Use this widget every Sunday during the 3-week ritual. It saves to your browser's localStorage so you can come back next week and add the next score. (No account, no server — just your phone or laptop remembering.)

What makes this hard

The hardest single thing about this project is continuing the Sunday ritual past week 2. You'll get bored. You'll skip a week. You'll tell yourself you don't need it. The discipline of doing it anyway — even when nothing dramatic happened — is what catches the drift before it hardens into the new normal.

The second hard thing is the honest reading. Drift is uncomfortable to admit. You'll be tempted to explain it away ("week 2 was busy, of course it was lower") or to fudge ("let me re-score, I think I was harsh"). Don't. The data is the gift. Read it.

The third — most painful — is realizing the drift came from your choices, not from AI getting worse. The journal didn't lose its voice on its own. You let Claude draft the opening because it was faster. Naming this honestly in the reflection is what closes the loop on the whole site.