The handwriting clock inside the Dead Sea Scrolls

A curved stroke and a scraped fragment of parchment contain very different kinds of time. One records a writing habit; the other carries a chemical signal from once-living material. A study published in June 2025 brought the two together in Enoch, a system designed to estimate the ages of Dead Sea Scrolls manuscripts from handwriting.[1]

The Groningen-led project addresses a historical difficulty: many surviving fragments do not carry an unambiguous date. Scholars can compare letter forms and manuscript traditions, but the evidence becomes harder to order when dated reference material is scarce. The researchers sought a connection to physical dating that could help calibrate those judgements.[2]

First, make sure the clock is reading the parchment

Radiocarbon work on the scrolls has a particularly tangible complication. Researchers had used castor oil to make some writing more legible. A 2001 study investigated whether that treatment could contaminate later dating: the younger carbon in the oil could make a sample appear younger than the underlying material. Its experiments found that the tested conventional cleaning procedure did not reliably remove all the oil.[3]

This is a good illustration of why the training label matters. An image model can learn a relationship very efficiently while inheriting a mistake in the dates it is asked to predict. In this case, the question begins before image analysis: what material does the laboratory measurement represent?

The 2025 study used a set of 24 radiocarbon-dated manuscripts to train the handwriting-to-date model. Its image pipeline separated ink from the background, then quantified features including the curvature of strokes and forms of characters. A Bayesian regression model connected those features to date distributions.[1]

What the reported results actually mean

The authors report mean absolute errors of about 28–31 years, depending on the analytical approach. Separately, experts judged 79% of predictions for 135 previously undated manuscript samples realistic. The latter is agreement with expert assessment, not confirmation against 135 independently known calendar dates.[1]

In the university’s account, the results also put some examples of handwriting earlier than conventional palaeographical expectations and suggest that styles could overlap more in time. That is historically interesting because a sequence of styles need not behave like a sequence of exclusive fashions: the arrival of one does not instantly end another.[2]

A helpful analogy is dating an old family photograph from clothes after checking the dates of photographs whose histories are known. The comparison can narrow a period. It does not turn every coat into a calendar, and an older style may remain in use. Here the physical dates provide anchors while the letter shapes provide additional information.

There are also distinct historical questions: when the writing material originated, when a particular copy was written, and when the text was first composed. Estimating one is not automatically an answer to all three. Enoch’s contribution is a way of connecting independent kinds of evidence about the manuscripts—not a shortcut around their history.[1][2]

Report an error or suggest a correction