Week 1 Workshop: The office that ate itself

How can we measure a copy becoming harder to read?

24 February 2027

How can we measure a copy becoming harder to read?

Learning target. Define a repeatable measurement, and explain why two readers of the same copy disagree without concluding that legibility is merely a matter of opinion.

Before the workshop#

Download the week 1 pack and read card.txt and counting-rule.txt. Ten minutes.

Reading. David Owen, Copies in Seconds (Simon & Schuster, 2004), chapter 1, on Chester Carlson and the years xerography spent being refused. No online copy is linkable, so the lecture supplies a summary of the specific claim used: that the technology was repeatedly judged to solve a problem nobody had.

The specimen#

Copy room test card 0S1I-B85l. Four lines, built from characters that look like one another — 0 and O, 1 and l and I, 5 and S, 8 and B, 2 and Z, 6 and G, 9 and g. That is what a legibility card is for: it concentrates the failure so it can be counted.

You are given four copies of it:

The control runs the same eight operations as the longest photocopy chain. That is the whole point of it: a control copied once would tell you nothing about a chain of eight, and the comparison this week rests on would not hold.

Each copy comes as two transcriptions, made by two different readers. Every transcription is exactly as long as the card, so you can compare position by position and never have to decide what lines up with what.

Everything in the pack is an authored teaching scenario. These are constructed transcriptions, not recordings of what a real photocopier and two real readers did, and every file says so.

If you want to watch this compounding effect happen live on ordinary text before you start counting the card, try the chain simulator. To practise marking a misreading against an illegible mark before you do it on the card itself, try the character-diff visualizer instead. Both are a supplement to this workshop, not a substitute for the pack or the worked answer.

Investigate — 40 minutes#

Read the transcriptions (10 minutes). Take one copy and both its readers. Mark every position where a transcription differs from the card. A position transcribed as · is an illegible: the reader could not resolve the character and declined to guess. That is not the same as getting it wrong, and your table keeps the two apart.

Count all four copies (15 minutes). Apply the rule in counting-rule.txt without modifying it. worksheet.csv arrives blank; fill in misreadings, illegibles and total for each copy and each reader.

Then count one transcription a second time. You should get the same number. This is a check on your repeatability, not on the rule’s: the algorithm is deterministic by construction, and what you are testing is whether you applied it the same way twice.

Compare readers, and check the control (15 minutes). Record the positions where the readers differ, not only their totals. Two readers can reach the same total by disagreeing in different places, and the totals alone would hide that.

Then look at the control. It ran eight operations. What did it come back as?

Evaluate — 30 minutes#

Write the table up with a method note. Then answer four questions in a paragraph each — these are what the week is actually assessed on, not the counting.

1. Why not average the two readers? An average would give you one number per copy, which is tidier. Say what it would destroy.
2. What does the control support? State the claim it licenses, precisely, and say what makes the comparison fair.
3. What can your evidence not establish? One card, one machine, two readers, one authored scenario. Name a conclusion someone might draw from your table that the table does not carry.
4. What would you change? One concrete improvement to this experiment, and what it would buy you.

A worked answer is in the pack. Open it after you have your own table and your own four paragraphs.

What you should find#

Inside the photocopy channel the totals rise with the number of operations. The control ran the same eight operations and came back with none.

That pair of facts is the course’s opening claim and its first qualification. This copying process damages what it copies. Copying as such does not.

The second finding is smaller and more awkward. After a single photocopy operation the two readers already disagree: one found an error, the other found none, in the same copy. Both followed the same transcription procedure; the scenario records them interpreting one damaged mark differently. There is still a fact of the matter — the card says what it says — but a reader facing an ambiguous mark has to decide, and two people decide differently.

This is where the week could go badly. It is tempting to conclude that the measurement is therefore worthless. Keep three things apart:

• the counting algorithm is deterministic — the same transcription against the same card gives the same number every time;
• your own repeatability is something you check, by counting twice;
• readers still differ, because resolving a damaged mark is an interpretation.

Legibility is a relation between a copy and a reader. Note too that reader R1 scores below R2 after four operations and above R2 after eight, so you cannot even rank the readers consistently — which is why an average would have been worse than useless here, not merely less informative.

Deposit#

Your character-error table, with the control recorded beside the photocopy generations, goes into the print collection. Week 2 reuses the discipline — a stated rule, applied the same way to every copy, with disagreement reported rather than smoothed — on a recording rather than a page.

The five-minute check#

Two copies of the same card. Cite one observed difference between them, and name one thing that difference cannot establish.

Assessment#

This week introduces The Chain, where you will run your own artefact through ten copying steps. Two things to note now.

Record your prediction before you start. Ten percent of that assessment is comparing what you expected to lose with what you actually lost. A prediction revised after the fact scores nothing.

Include a control. The one in this pack is the whole reason week 1 can make a claim rather than a gesture. No mark is given for dramatic degradation.

The first Generation Check is the one above. Best ten of twelve count, so a missed week costs nothing.