Week 3: What Makes a Measurement Trustworthy?

A number is not evidence by itself — this week is about the difference between a measurement someone else could reproduce and one they'd have to take your word for.

8 March 2027

Question#

You’ve now measured height, displacement and airtime for two weeks. Suppose you hand your week-2 table to a classmate with nothing else attached, and ask them to get the same numbers. Could they? If your honest answer is “only if they already knew which preset I used, which mode, and roughly how long I held the button” — that information was doing real work, and it wasn’t written down.

Prediction (skip if you’d rather just look)#

Of the following, which do you expect varies most between two people independently running “the same” trial: the preset, the mode, or the hold duration? Which matters least once the other two are fixed?

Experiment#

Open the Jump Lab and pick any preset. Run five immediate-mode jumps aiming for what you’d call “a full hold” each time, without watching the clock. Look at your five airtime values. They won’t be identical — your thumb doesn’t release at exactly the same instant twice — but they should cluster, because the preset (gravity, launch speed, air control) hasn’t changed at all between trials. The spread you’re seeing is measurement noise from a human input, not noise in the simulation: the physics engine is deterministic, so the same recorded input replayed through it produces the same output every time, which is exactly what the Jump Lab’s A/B tool relies on later this course.

Explanation#

A measurement is reproducible when a second person, given your method and nothing else, lands inside the range your numbers already show. That method is three things: which configuration (preset name, or the exact parameter values if you changed any), which mode, and how many trials. Leave any one out and your table stops being a measurement of the jump and starts being a measurement of whatever the reader has to guess.

This is also why the Jump Autopsy brief, due this week, asks for a method section before it asks for an explanation of the difference. An explanation built on numbers nobody else can regenerate isn’t wrong because the physics is wrong — it’s wrong because there’s no way to check it.

Task#

Take one preset from the Jump Lab and run three trials in immediate mode, recording height, displacement and airtime for each, plus their averages. Write a two-line method statement above the table (preset, mode, trial count) short enough that a classmate could reproduce it without asking you anything else. This table and method statement are the exact shape the Jump Autopsy brief scales up to two jumps — if this one is reproducible, that one will be too.