Methods · Experiments · Evidence
A map of the research, not just a generator.
Two source geometries, distinct ranking metrics, multiple construction hypotheses and published negative results. Explore what was actually tested, what was built, and what remains unverified.
JP · KanjiVG · 2,770CN · GB2312 / MMah · 6,763No controlled native-reader validation
A geometric match is a candidate, not proof of readability. A concept diagram is not the output of an experimental constructor. Draft-branch experiments are linked but not silently treated as released generators.
Chinese / PRC-source
GB2312 coarse self and distinct-pair screening, centre and extremity optimization, thickness heuristics, and R-ODD/harmonic/fusion drafts. Do not mix its ranks with KanjiVG.
Japanese · Searches and metrics
These seven historical algorithms search or rank candidate forms. They are not seven separate visual generators. Different experiments have different source fonts, cohorts and geometric scales.
Japanese · Visual constructions and natural candidates
The generator draws real KanjiVG paths. Natural candidates use source forms; synthetic methods manipulate or combine stroke paths. C₂ is self-rotation, not a distinct-character generator.
Chinese · Coarse screening
37×37 source-ink IoU over a GB2312 repertoire: screening only, not a human-reading metric.
Chinese · Constructions, operators, null results
Methods here have different evidence statuses. Some have source code and tests but no audited redistributable gallery image; draft operators remain draft. A null result is still a result.
Follow the evidence
The simpler Reader Gallery is exploratory feedback, not a validated benchmark. The full experimental inventory lists which methods have authentic, publishable visual outputs and which still require provenance and licensing review.
Full audit and availability matrixReader GalleryLimits and biasesIDS algebraFamily rulesExistence criterion