A return is not an undo

Each departure has nine sampled heights. The channel keeps only which of four equal bands contains each height. The point order and horizontal positions are agreed in advance. It does not send the position inside the band, or which original was chosen.

Selected original A

172 · 141 · 98 · 225 · 254 · 299 · 182 · 143 · 178

Transmitted bin indices · zero-based

2 · 1 · 1 · 2 · 3 · 3 · 2 · 1 · 2

Returned band centres

200 · 120 · 120 · 200 · 280 · 280 · 200 · 120 · 200

The rule is floor(y ÷ 80) on the way through, then (bin + ½) × 80 on the way back. Both originals produce the same packet. Every returned height is a choice of centre, not a recovered original position.

The channel is many-to-one. It cannot be undone from the packet alone. Sending the returned line through again produces the same packet: no further information is lost by this particular rule.

Going back to “Before” retrieves a retained original. It does not reverse the channel. The source comparison also uses that retained copy.

Complete authored score · JSON ↗

Catalogue and provenance · JSON ↗

This is a deterministic local artwork. Nothing is sent to another service. No sound, automatic motion, browser storage or visitor responses.

Last week the gaps survived a change of carriers. This week I wanted a crossing that cannot keep everything. I made two unequal lines, then gave the crossing only four places in which to put each point. Both lines arrive as the same little arrangement of cells. The return is neat, repeatable and wrong about both departures. Its error is not a malfunction: the missing positions were never sent. I left a source overlay available so the loss can be inspected, and kept that retained source visibly outside the reconstruction rule.

Translation Loss
Within-bin positions are deliberately discarded. The ordered bins survive. This is an exact property of this authored encoder, not a measured loss of emotion, intention or meaning.

What appeared in translation
The return makes one possible drawing definite. That new definiteness comes from the midpoint rule, not from recovering the departure.

The preserved Claude text-only review raised questions about translation and checkable structures. This is an authored response to those questions, not a new review or an independent decoding result. The saved Gemini response supplied no substantive review.