Module kanji 

Kanji table — JP kanji with on/kun readings + relative frequency.

Generated by tools/nihongo/build_kanji_rs.py from tools/scoring/data/supplemental/jp_kanji_v1.tsv. DO NOT edit by hand — re-run the script. The source TSV was produced by an LLM batch call; corrections go to the TSV first, then regenerate.

Schema: each kanji carries 1+ Hepburn romaji readings (on-yomi + kun-yomi stems where the bare-kanji is the natural surface form) and a 0–100 freq score. Composite-layer scoring uses freq to lift high-frequency JP kanji above rare Chinese chars / 生僻词组 per the user’s design rule (2026-05-23: ‘日语高频要高于中文难检字和生僻词组’).

Codepoint-identity-with-SC constraint REMOVED (was the v1.1 design). Shinjitai-only kanji (経/漢/楽/愛/東/見/書/…) are first-class entries now — Inputx’s JP plugin delivers professional-grade JP, not just the SC-overlap subset.

Structs

KanjiEntry

Constants

KANJI_TABLE

Functions

lookup_by_reading

Lookup: given a Hepburn romaji string, return all (kanji, freq) tuples whose readings include exactly that string. Linear scan over the table is fine for a few thousand entries — modern keystroke-rate cost is dominated by the FST traversal in the wubi/pinyin engines, not this.