Week 5 Workshop: The scriptorium

Which copying mistakes can become useful evidence?

24 March 2027

Which copying mistakes can become useful evidence?

Learning target. Classify omissions, repetitions and substitutions, and say which are weak evidence.

Before the workshop#

The print collection holds one document you have not read: the Ordinance of the Copy-House, twelve numbered lines of copy-room rules. You will not read it today either — it stays withheld until week 6.

Last week, and this week#

Week 4 started from one output and asked whether the evidence could tell two histories apart. This week starts from five copies and no original, and asks a narrower question first: what kind of mistake produced each disagreement between them?

Classifying the mistake is not the same as ordering the copies — that is next week’s task, done from the table this week produces. Getting the type right first is what makes the ordering possible at all.

Check — 5 minutes#

Name the three broad classes of copying error this week works with: an omission (something present in the exemplar is missing from the copy), a repetition (something present once is copied twice, or a run is skipped because the eye jumped between two identical stretches), and a substitution (one reading stands in for another). If you cannot state all three from memory, say so before the workshop starts rather than during it.

Receive — 15 minutes#

Read the branching instructions (the pack folder is shared with week 6 for the reason above — nothing there is a spoiler). Each of five teams copies from the previous team’s copy rather than from a shared original, so errors accumulate down a branch instead of each team making independent mistakes from the same source. Two teams share a parent partway down the branch; you are not told which two.

If you want to see a substitution and an illegible mark called out inline, character by character, before you classify your own witnesses by hand, try the character-diff visualizer. It marks substitutions and illegible characters only — it has no notion of an omission or a repetition, since those change a string’s length and this tool only ever compares two strings of equal length. It is a supplement to this workshop, not a substitute for the classification task itself.

Investigate — 40 minutes#

Produce your witness (20 minutes). Follow your team’s position in the branching assignment and copy your assigned parent text by hand, line by line, onto the supplied line-numbered sheet. Introduce at least one deliberate error of a type your team has not used yet, and log what you changed and why in your own working notes — this log is not shared yet.

Classify differences with line IDs (10 minutes). Compare the five finished witnesses (witness-a.txt through witness-e.txt in the pack) against each other, not against any original you do not have. At each line where they disagree, record the line number, what each witness reads, and which of the three classes in the check above it looks like. collation.csv lists the same disagreement points if your team’s hand copies do not turn out to be informative enough — the instructor fallback exists so that a bad run does not cost you the exercise.

Package for another group (10 minutes). Anonymise your witness labels (swap them for a scheme the other group has not seen) and hand your table to a different team. You will reconstruct their pack, not your own, in week 6 — a team that already knows its own answer would be practising memory, not inference.

Evaluate — 30 minutes#

Write the error worksheet you are depositing: one row per disagreement, with line number, the readings, your classified type, and one sentence on how confident you are in that classification.

Then answer this in your worksheet, in your own words, before you open the box below: why might two witnesses agreeing on a spelling correction be weak evidence that one copied from the other? Write your answer down first. An answer you wrote is worth more than one you recognise after reading it — that is the same rule this course applies to every other prediction.

Deposit — 30 minutes#

Anonymised witnesses and the classified error table go into the print collection as print/ordinance-a through -e. The true parent log stays withheld until week 6’s reveal — this week’s deposit is evidence, not the answer.

Accessible route#

Editable text with line identifiers; no unfamiliar-script skill is assessed. The plain-text witnesses and collation.csv are the task, not a substitute for it — nothing here requires reading an unfamiliar hand or script.

Assessment#

For The Stemma, the error types you name today are the vocabulary its rubric grades in: a shared omission or repetition supports shared ancestry more than a shared normalisation does, and saying which kind of agreement you are relying on is part of the mark.

For The Chain, due Friday 2027-03-26 at 12:00 Canberra time, this week’s classification skill is not required — but bound your conclusion there the same way: state what kind of evidence supports it.