Appearance
Distinguish scribal hands when the boundaries between scribes answer a question you actually have — dating the campaigns of work, attributing authorship, mapping how a workshop split a job, or explaining why the text changes character partway through. If you only need a readable transcription, skip fine-grained attribution; it is slow, error-prone, and adds nothing to your output. The decision is a cost-benefit one, and this article lays out the signals on both sides.
When does distinguishing hands pay off?
It pays off whenever the who and where of writing is part of your evidence. Codicological projects use hand changes to detect added quires; editors use them to weight variant readings; book historians use them to reconstruct scriptorium practice. In each case the hand boundary is data. Conversely, for a single-scribe ledger you are simply reading, the exercise is pure overhead.
When should you NOT bother?
Be honest about diminishing returns. Avoid the rabbit hole when: the text is short and homogeneous; image quality cannot support fine stroke analysis; or you are tempted to posit a new hand for every page of natural variation. The cost is not only your time — every attribution you publish becomes a load-bearing claim others may build on.
| Distinguish hands when… | Don't bother when… |
|---|---|
| Dating distinct work campaigns | Single homogeneous text |
| Attributing authorship or scribe | Goal is just readable transcription |
| Studying workshop division of labour | Image too poor for stroke analysis |
| Explaining textual variants | You'd posit a hand per page |
| Building a corpus for writer ID | Variation explained by fatigue/exemplar |
What features reliably tell hands apart?
Weight habitual, low-conscious traits the scribe could not easily standardise. The strongest are the ductus (the direction and order of strokes), the form of g, d, a and the et abbreviation, ligature preferences, pen angle, and line-end word-breaking habits. Treat module (size), ink colour and even general "neatness" as weak signals, because one scribe changes all three across a day's work.
text
Strong (habitual, hard to fake):
ductus / stroke order · g and d forms · et-sign · ligatures · pen angle
Weak (change for one scribe):
module/size · ink colour · line spacing · overall tidinessHow do I record an attribution defensibly?
Make the reasoning reproducible. Tabulate the features per section so a reviewer can check your logic rather than trust your eye, and state a confidence level. A simple per-witness feature matrix works well:
csv
section,folios,g_form,et_sign,pen_angle,verdict,confidence
A,1r-12v,figure-8,tironian,30deg,Hand 1,high
B,13r-15v,single-bowl,ampersand,45deg,Hand 2,medium
C,16r-20v,figure-8,tironian,30deg,Hand 1,highIf you use a structured tool, Archetype (the successor to DigiPal) lets you annotate individual letter instances on the image and query them, which turns "it looks different" into evidence you can point to.
Can the same scribe masquerade as two hands?
Yes, and this is the central trap. A single scribe varies with fatigue, a fresh pen, a new ink batch, copying from a different exemplar, or shifting between a display script and a working hand. Because any one feature can move for innocent reasons, never split on a single signal. Require several independent habitual features to agree before you posit a second hand, and prefer the simplest hypothesis that fits the page.
What does getting it wrong cost?
A false split mis-dates sections and inflates the number of contributors; a false merge hides a real collaborator. Either error propagates: a stemma, a database of scribes, or a statistical writer-identification study all inherit it silently. That asymmetry is the argument for explicit, hedged attributions — "probably Hand 2, on g-form and et-sign, low confidence" is more useful and more honest than a bare assertion.
Key Takeaways
- Distinguish hands only when scribal boundaries answer a real question (dating, attribution, workshop study).
- Skip it for short, homogeneous texts where you just need a readable transcription.
- Weight habitual traits — ductus,
g/dforms,et-sign, ligatures, pen angle — over module and ink. - Never split on one signal; require several independent features to agree.
- The same scribe can look like two hands through fatigue, pen change or a different exemplar.
- Record attributions as a reproducible feature matrix with an explicit confidence level.
- Errors propagate into stemmata and statistics, so hedge attributions rather than overstate them.
Frequently Asked Questions
When is it worth distinguishing scribal hands?
When the boundaries between scribes carry meaning for your question — dating the parts, attributing authorship, reconstructing how a workshop divided labour, or explaining textual variants. If your goal is just a readable transcription, fine-grained hand attribution rarely pays off.
What features actually distinguish one hand from another?
Look at habitual, hard-to-fake traits: the ductus and stroke order, the shape of g and d and the abbreviation for et, ligature choices, the angle of the pen, and where the scribe breaks words at line ends. Ink and module can change for one scribe, so they are weak evidence alone.
How many hands is too many to claim confidently?
Be cautious past three or four in a short text. The more hands you posit, the higher the chance you are splitting one scribe's natural variation across a tired afternoon or a different exemplar. Prefer the simplest hypothesis that fits.
Can the same scribe look like two different hands?
Yes. A single scribe varies with fatigue, a change of pen or ink, copying from a different exemplar, or writing a display script versus a working hand. This is why you weigh several independent features rather than one.
Can digital tools attribute hands automatically?
Tools like Archetype (formerly DigiPal) support structured human annotation of letterforms, and machine-learning writer-identification systems exist, but both are aids. A confident attribution still rests on documented, reproducible feature comparison, not a black-box score.
What is the cost of getting hand attribution wrong?
A false split or merge can mis-date sections, mis-assign authorship, and distort any stemma or statistical analysis built on top. Because downstream work inherits the error, the honest move is to state attributions with explicit confidence and reasons.