The numbers

One row per ku-ro total. Entries added are the numbers in the lines the total covers, back to the heading or to the total before it. Their sum adds them up. Stated total is the number written after ku-ro. Difference puts the sum against the stated total, in words: none, so many over, or so many short. Whole numbers only. Fractions stay out of the sum.

Result is the class from our audit of 13 August 2026, the same one each document page gives. "Text missing" means damage in the lines the total covers, so the sum can't be trusted. "Entries after the total" means the list carries on past it. "As a running total" means the total counts the total before it as one of its entries. Two totals have no number after them. They're listed and not tested.

A balance means the arithmetic works, and that's all it means. A miss doesn't mean the scribe slipped, either. Clay breaks.

Each fraction sign as SigLA draws it, with its letter from GORILA's list. Two counts per layer: how many times the sign turns up, and on how many documents. The sign-by-sign text has more of them, because SigLA hasn't traced every document.

The value column only shows a value that What is known already states, from Corazza and colleagues (2021). That's five signs. The others say "not given", which means this site doesn't give one.

Every site with at least one number in its sign-by-sign text, so 18 of 55 sites in the corpus. Documents with numbers is out of that site's documents with a sign-by-sign text. The two ku-ro columns come from the audit, so a site with no ku-ro total has nothing to test.

These are counted on the SigLA tracings (the sign array, in the project's own jargon). SigLA tags every traced sign with a role. A sign that doubles as a syllable, like AB30, which is also ni, only counts where SigLA tags it as a logogram. Traced as a logogram is out of every tracing of that sign. Documents is out of every document SigLA traced.

The tracings have no numerals, so the numbers come from the sign-by-sign text. That only works where SigLA's crops and the text agree sign for sign, on 345 of 786 documents in the SigLA tracings. Crops and text agree counts those documents for each sign. A whole number follows counts the logograms there with a whole number right after them. A number on a ku-ro line is left out, so a total doesn't count twice. Sum of those numbers adds them up.

It's a small slice of the data, on purpose. Better a small sum you can check than a big one you can't. Nobody knows the units either, so treat a sum as a count of marks. The list under the table shows every document and every number behind each sum.

Every table above, as CSV for a spreadsheet and as JSON for code. The JSON carries the denominators and the licence too. The build writes these files from the same data as the page, so they match. They hold data derived from SigLA, so they're CC BY-NC-SA 4.0, like the source.