Neuroplasticity Lab
Every two to three weeks I pick a skill I don't care about keeping, train it on a fixed protocol, measure it, then retire it and start something else. Right now the skill is memorising a shuffled hand of playing cards. It being disposable is the point — it isn't what I'm measuring.
What I'm measuring is the practice protocol and how accurately I can predict my own performance. The hypothesis is that the skill-specific gains don't transfer — getting good at cards won't make me good at whatever comes next — but the protocol and the self-prediction do. The test is comparing the opening sessions of each new skill against the opening sessions of the last one, which doesn't mean anything until about the third cycle. I'm still collecting.
Every session runs the same way round: predict first, with no past results on screen, then take the cold test, then see both numbers together. If I'd already looked at my own data in the half hour before, that gets logged too — as something that might have moved the number, not as a rule I broke.
- STATUS
- Running
- CYCLE
- 2–3 weeks
- CURRENT SKILL
- Card memorisation
- SESSIONS / WEEK
- 6–7
The sessions
- Daily · 10 min
- Four or five days a week, at home. Prediction, then one cold trial with no warm-up, then the reveal, then a short practice loop of trial and rest. Ten minutes, and it's the session that has to survive a bad day — abandonment is the failure mode I actually worry about, so the whole flow is under ten taps.
- Lab · 30 min
- Once a week, at home, on a guided timer that runs the blocks for me: context, prediction, a five-trial test block (only the first one counts as cold), a review of the week's numbers, fifteen minutes of practice, error tagging, planning the next week, and two minutes of non-sleep deep rest at the end. It moved out of school to home because one interruption during encoding invalidates a trial, and the lab has five of them to protect.
- Journal · 30 min
- Friday free period at school. Writing rather than practice — what I did, what the data says, what surprised me, what I'm changing. It records as a session so it shows up in the activity record, but it has no trials and is excluded from every performance metric.
- Legacy retest
- A one-off re-run of a skill I retired weeks ago: pick it, predict, one cold trial, save. It sits outside the current cycle's stats but stays on that skill's lifetime chart. It's the closest thing I have to a direct read on what actually stuck.
What gets measured
- Retention index
- Today's cold test against my best from the last comparable sitting. "Comparable" means the same skill, the same difficulty setting, and the same version of the method — change any of the three and the comparison starts over rather than quietly averaging across the break. If there's nothing to compare against, or the gap is over three weeks, it reads as a baseline instead of a number.
- Calibration error
- How far my prediction was from what actually happened, kept with its sign as well as its size. Consistently over means overconfidence, consistently under means the opposite, and the two are different problems. This is the number I expect to improve across skills even when nothing else does.
- Consistency
- How much the trials inside one sitting vary. Averages hide it: five even trials and five wild ones can land on the same mean, and only one of those is a skill I'd trust on a bad day.
- Covariates
- Logged alongside every session rather than controlled for: where I was, how many times I got interrupted, and whether I'd looked at my own performance data in the last thirty minutes. They're there so I can check whether any of them is quietly driving a result — not so I can stop them happening.
The tool
I built the tracker it runs on, because logging it on paper meant I stopped logging it. It holds the prediction gate, the session runners, the charts, and a raw CSV export of every trial. Everything is entered in the moment, never reconstructed from memory afterwards.
It lives on this site behind the same password as everything else of mine here, so the link below will ask you to sign in.
Owner-only — you'll hit the login gate.