Pausanias Analysis

Two sentence classifiers, compared across the whole corpus

Original vs. Greta-inspired

Each Greek sentence is tagged by two independent classifiers (both gpt-5.4-mini, temperature 0):

The two agree on the bucket for 82.0% of 11,302 sentences.

Corpus-wide bucket rates

BucketOriginalGreta-inspired
mythic31.8%25.5%
historical30.4%27.4%
both1.6%
other37.8%45.5%

Where they disagree (bucket confusion)

Rows = Original, columns = Greta-inspired; the shaded diagonal is agreement.

Original \ Greta-inspiredmythichistoricalbothother
mythic2,6684277811
historical822,76395494
other137288123,833

By book

BookSentencesAgree OriginalGreta-inspired
mythhistother mythhistbothother
Book 11,35482.0%34%36%31%27%32%3%38%
Book 21,25882.5%46%14%40%36%13%1%50%
Book 395980.8%32%33%35%24%28%3%44%
Book 41,07075.3%24%50%26%21%37%2%40%
Book 594481.9%30%23%47%24%23%1%52%
Book 687979.5%16%40%45%11%46%1%42%
Book 793084.5%24%42%34%18%37%2%43%
Book 81,47385.6%32%23%45%27%21%1%51%
Book 91,13982.8%42%20%37%34%19%1%46%
Book 101,29682.8%31%30%38%24%27%2%47%

Click a book to see every sentence where the two classifiers disagree.