The poor image#
For three weeks you have known what was done to the copy. This hour is about the ordinary case, where you do not.
The image that travels#
Steyerl’s argument: the low-resolution copy is not a degraded version of something better, but the form in which images actually circulate.
From Hito Steyerl, “In Defense of the Poor Image”, e-flux journal #10 (2009), free at e-flux.com.*
The narrower claim, and it’s ours#
A circulating image may arrive without the metadata, logs or intermediate files needed to reconstruct its history.
Sometimes it arrives with them. That weaker situation — evidence thinner than the question — is what this week works in.
Output 7#
Five observations: 1200×800, 148 KB, 4:2:0 subsampling, no EXIF.
Route A: resized once, saved once.
Route B: saved, resized, saved again.
Both fit#
Apply the tolerance — both land inside it on every measurement.
The file has exactly one history. These five observations underdetermine it.
The sentence to get right#
“These two routes are indistinguishable” is incomplete, and almost certainly false as stated.
What is true: they are indistinguishable on these measurements, at this tolerance.
A claim about the evidence, not the objects#
Change the tolerance — it may dissolve.
Add a measurement — it may dissolve.
Say which measurements and which tolerance, every time.
Ask for evidence before you get it#
Name an observation that would discriminate — before obtaining it.
Look first and decide what it means after, and whatever you saw will support whichever route you preferred.
Three records, one distinction#
| Record | Separates the routes? |
|---|---|
| Working-directory listing | No — co-existence is not descent |
| Handover note | No — same gap, human direction |
| Transformation record (input hash matches) | Yes |
Ruling out is not establishing#
Suppose the transformation record excludes route A.
You have not shown route B happened — only that of the two histories proposed, one is excluded. Others fit the same record shape.
The claim you can defend: not route A, and consistent with route B.
Before Wednesday#
Read the observations and the tolerance. Do not open the reveal, and do not decide anything yet.
Workshop: The poor image
Text walkthrough#
For a reader who cannot use the deck view: Output 7 arrives with five observations — dimensions, file size, chroma subsampling, and the absence of EXIF — and two proposed processing histories both fit them within the stated tolerance, so the initial evidence underdetermines which one happened. The key sentence to get right is that “these routes are indistinguishable” is incomplete on its own; what is actually true is that they are indistinguishable on these particular measurements at this particular tolerance, and changing either could separate them. Of three further evidence items available, two — a working-directory listing showing a full-size file existed, and a handover note confirming one existed somewhere — fail to separate the routes because co-existence is not descent: an interim file could be unrelated. A transformation record that names the full-size file as the actual input to the resize, with a matching hash, does separate them, because it records what read what rather than merely what existed. Even having excluded one route this way, the evidence has not established the other — any history of the same shape would produce the same record — so the defensible claim is “not route A, and consistent with route B,” which is weaker than “it was route B.”