Our breeding hero, on the taste axis. Three accessions, one species: green and red (the clean anthocyanin colour-genetics cross) and hojiso (bicolour + serrated form, into the gene pool).

Germinate: surface-sow — it needs light — evenly moist at 20–25 °C; up in 1–3 weeks. Don't bury.

Breeding levers:

• Colour — red is anthocyanin (malonylshisonin), light-inducible. The Red×Green F2 segregates 1:2:1 (green : intermediate : self-red — incomplete dominance, one major locus, candidate PfMYB113b), so the heterozygote is a visible middle class the eye scores poorly and our CIELAB reads cleanly. (Loop 0 calibrates how hard our light pushes it; map + citations in ../docs/literature.md.)
• Photoperiod — a textbook short-day plant: flowers below ~14–16 h (red crispa ~14 h 20 min, but cultivar-dependent — measure per accession). The flower-on-demand superpower — long nights to flower one individual, a night-break to hold its siblings — is real and commercial, but needs multiple consecutive long nights (not one) and a 2–4 h night-break (the "30 min" was optimistic; our accessions' minimum is unmeasured).
• Smell — perilla aroma is chemotype-controlled with known dominant/recessive genetics (six types: PA perillaldehyde — our culinary one; PK perillaketone — a lung toxin, cull; plus PL, PT, C, EK). Ours is almost certainly PA. The culinary PA type needs dominant alleles at both a monoterpene-cyclase locus (G1>G2>g — "make limonene") and an oxidation locus H ("oxidise it to perillaldehyde"); PK runs on a separate, independent branch, so fixing dominant G + H drives flux toward PA and away from the toxin. Designing a smell here isn't a metaphor.

Scent vocabulary (for the nose): perillaldehyde reads herbal-green, faintly cumin/coriander, a cool minty lift. Crush a leaf and we log it together — your qualia, my chemistry.