theluckystrike

Reading estimate worked examples without guessing

This is a reference extract from token-estimate-worked-examples.csv, a 21-row dataset with the columns sample_id, profile, raw_chars, effective_chars, non_ascii_chars, token_estimate. The figures below are copied directly from the file; nothing here is estimated or interpolated.

Provenance recorded in the dataset: Token-count estimation: worked examples for a whitespace-collapsing character model. DERIVED, and re-checkable by recomputation. Every token_estimate value below was computed 2026-08-04 by reproducing the model INDEPENDENTLY in Python, not by copying

sample_id: plain_english_sentence | profile: Prose | raw_chars: 64 | effective_chars: 64

sample_id: plain_english_sentence | profile: Code | raw_chars: 64 | effective_chars: 64

sample_id: plain_english_sentence | profile: Mixed | raw_chars: 64 | effective_chars: 64

sample_id: english_paragraph | profile: Prose | raw_chars: 186 | effective_chars: 186

sample_id: english_paragraph | profile: Code | raw_chars: 186 | effective_chars: 186

sample_id: english_paragraph | profile: Mixed | raw_chars: 186 | effective_chars: 186

The full table, with the remaining 15 rows and an interactive calculator built on it, is at the claudhq reference table