Different situations can require different responses. If every available reading looks identical, those readings alone cannot tell us which response fits. This makes distinguishability important: useful observations preserve differences that matter for a task. Here the card's Being stays fixed. A new answer changes our evidence; Becoming would mean a change in the card's state. Watch the test menu and the received answers do different jobs.

The observer from episode one receives zero. Both u and v fit that lamp reading. We will compare what a chosen set of observations can distinguish, using the known model rather than the audience-only cutaway. To make the grouping visible, first switch to the site’s four-card example.

The cards are zero-zero, zero-one, one-zero and one-one. A names colour, the first digit: zero is teal and one is gold. B names shape, the second digit: zero is circle and one is square. These are new labels for a different example; they are not the hidden configuration names u, v, w and z. The card stays fixed while exact tests are read.

Make only A available. Zero-zero and zero-one give the same A answer, so this menu leaves them together. One-zero and one-one form the other group. Now make both A and B available. A complete pair of answers identifies each card separately. The signature is simply the ordered list of those answers. Changing the menu changes the distinctions this observer can make; it does not change a card’s colour or shape.

Compare two operations from the same four possibilities. In one branch, enable B but receive no new answer yet. In the other, keep the existing menu and receive A equals zero. That answer restricts the current candidates to zero-zero and zero-one. Enabling another test and learning a fact have different effects. A menu says which questions may be asked; a record says what answers have arrived.

Reading A again produces the same answer on a fixed card. Recording A twice gives a longer record, but it does not separate cards that A already left together. This claim uses a deterministic exact test. Independent noisy readings are a different situation, introduced later; we are not treating every repeated measurement as useless.

Our examples use complete readings: the reader sends an explicit zero or one. Receiving zero is different from receiving no report. Here, zero says that the tested property is false. An optional course uses a different reporting rule; its convention is explained in the supporting notes. The core route keeps complete zero-or-one records.

Pause and classify this change: we enable B, but have not read it. Have we learned that B is one?

[SILENT HOLD:5SECONDS;NOT SPOKEN]

No. We changed availability, not the received record. If we next receive B equals one, then the compatible cards are zero-one and one-one. That later restriction comes from the answer.

Return to our machine. Its core reader reports only the lamp, grouping u with v and w with z. The A and B card sensors have not been added to it. We now know what those present readings can distinguish. The next episode asks a new question: will that description also support predictions after a button press?