A Three-Layer Framework for Measuring Names and Its Census Application on a Token Launchpad (arXiv (Cornell University))

Summary

• The paper builds a census-level dataset of all 513,647 naming attempts on the four.meme token launchpad on BNB Chain between February and June 2026.p.1
• The form layer extends processing-fluency measurement to Chinese names through 38 deterministic features adapted to the Chinese writing system.p.1
• Within the research window, 237,937 unique names correspond to 513,647 naming attempts, and 71,585 names were used more than once.p.7
• The maximum reuse count for a single name reached 1,130, and 58,816 tokens (11.4%) never produced a single trade.p.7
• Cumulative unique names follow Heaps' law, V = 1.24 · N^0.921, and the share of new names declines from 52.0% in February to 37.8% in June.p.18
• The median interval between same-name copies is 0.7 minutes for same-creator copies and 5.6 minutes for cross-creator copies.p.21
• In the six hours after the Spring Festival Gala broadcast began, 4,128 new tokens were created, roughly 2.8 times the count of the six hours before, and 12.3% used festival-related names.p.22

Key insights

1. The three layer subindices are nearly orthogonal, with pairwise Spearman correlations no greater than 0.11.p.16
2. Authority borrowing is far more common in mixed Chinese-English names (45.60%) than in purely Chinese (23.30%) or purely English (19.63%) names.p.16
3. Only 6.65% of Chinese names differ in tone entropy between whole-name and character-by-character reading, while 30.6% of Chinese characters are heteronyms.p.10
4. The Chinese share of names falls from 44.5% in February to 36.8% in March and then climbs to 50.3% in June.p.18
5. Name reuse is concentrated: the largest 1% of lodes contain 17.4% of tokens, and the name "test" accounts for 1,130 tokens from 414 creators.p.20
6. Creators number 35,677 addresses, of which 32.1% issued two or more tokens, and the most prolific address issued 9,750 tokens in five months.p.9
7. Of all names, 6.4% have a precursor anchor, defined by an edit similarity of at least 0.85 to an earlier name.p.15
8. Market snapshot coverage is 87.31%, and missing snapshots concentrate on tokens that never traded.p.8