Dive deeper

The harder parallax pairs

This is the depth layer behind the parallax page. The walk keeps to a rock, DNA, a document, a thermostat, because the move only needs one clean case to land. Here are the cases that need a little field knowledge — each one a place where some discipline already does the record-side reading, just under its own name. Read it if you want the heavier pairs; skip it and the move is unchanged.

The move is the same every time: describe one thing the normal way, then describe it as a record — the surviving form a funnel of pressure produced, still carrying what made it, still shaping what's downstream — and look at the gap. On the walk the pairs were everyday. These are the ones where the "record reading" already has a technical name, which is the interesting part: people arrived at the record side on their own terms, from inside fields that never talked to each other.

Rate–distortion

Normal reading: a result in communication theory about how few bits you can get away with sending. Record reading: the fewest bits that still rebuild a thing well enough to use — which is exactly a record-quality bound. It already answers "how good does the surviving copy have to be?", which is a record-side question wearing an engineering hat.

The information bottleneck

Normal reading: a machine-learning method for squeezing an input down. Record reading: squeeze the input until only the part that predicts the thing you care about survives — and what survives the squeeze is a record built for downstream use. The name even admits it: a bottleneck. The whole construction is compression toward the relevant record.

Requisite variety

Normal reading: a cybernetics law about controllers — to hold something steady, a controller needs enough internal states to match the disturbances it faces. Record reading: the controller has to carry enough of a record of what can go wrong to absorb it. It's a record-capacity law in disguise — the thermostat from the walk, stated as a theorem.

Peirce's index

Normal reading: a category from the study of signs — a sign physically connected to the thing it points at (smoke and fire, a footprint and a foot). Record reading: an index is a record — a mark tied to its object by a real causal trail. It names, precisely, the slot the record side has been pointing at the whole time: not a stand-in chosen by convention, but a structure the world actually laid down.

The documentary record

Normal reading: a debate in library and information science about what counts as a "document." Record reading: the field's own answer is that anything treated as evidence is a record — the classic example is an antelope brought into a zoo, which becomes a record of its species the moment it's kept as one. That's the record idea arrived at from the side of archives and evidence, with no physics in sight.

Genetic sequence

Normal reading: the molecular-biology account of DNA as a sequence carrying heritable instructions. Record reading: the durable copy of what worked, read out to rebuild a body downstream — the same pair as the walk's DNA, but here it's the field's own framing (sequence-as-information) doing the record-side work, not a metaphor laid on top.

What all of these have in common is the thing worth taking away: these aren't cases where the record idea was imported and stretched to fit. Each field built its own version, for its own reasons, and the record reading was already sitting inside it. That convergence is evidence the lens is useful — that it reads something really there — not evidence it's been proven. Convergences, not corroborations.

The receipts

The full crosswalk — every family with its source handle, its record-side status, and the false friends that only share a word — is in the raw files: