APT4's English fertility tax exceeds its preamble tax on 2 of 4 scientific corpora (English arXiv abstracts and English Python code) and is below it on the other 2 (English LaTeX method sections and GSM8K problems).

Relation: Exceeds on a count of texts

metric

Fertility taxQuantity compared.

subject

APT4 tokenizerTokenizer whose tax is measured.

comparator

Mistral-derived tokenizerTokenizer the tax is measured against.

reference text

Constitution preambleText compared with.

corpora above

English arXiv abstracts; English Python codeCorpora where the tax exceeds the reference.

corpora below

English LaTeX method sections; GSM8K problemsCorpora where the tax does not exceed the reference.

count above

2 corporaNumber of corpora above the reference.

count total

4 corporaNumber of corpora compared.

Method and evaluation protocol

Compare each English corpus's APT4 tax with the preamble tax; all four 95% intervals exclude the reference value.

Dataset

Four English corpora: arXiv abstracts, LaTeX method sections, Python code, GSM8K problems.Version: unspecified · Access: restricted

Reported results

Above: English arXiv abstracts, English Python code. Below: English LaTeX method sections, GSM8K problems.

Uncertainty and replication

Each per-corpus interval excludes the reference value; the reference itself has no interval.

Evidence references

atlas.json · publichttps://github.com/stw2/tokenizer-science-tax/blob/719d51eb4ac7e6e0f36d69cf339c8e976266ee98/experiments/E01-fertility-atlas/results/atlas.jsonmetrics.json · publichttps://github.com/stw2/tokenizer-science-tax/blob/719d51eb4ac7e6e0f36d69cf339c8e976266ee98/experiments/E01-fertility-atlas/results/metrics.json

Limitations

Numeric tables, named in the hypothesis, were not measured; LaTeX covers method sections only.