flowers.html: the flowers, given a Three.js stage of their own

[aj-dev-smith]

What this is

A second single-file piece, flowers.html, built by flowers/build.js exactly the way canalisation.html is built — one file, no server, no CDN (Three r147 UMD is vendored in vendor/), parsed before written, duplicate-scan gated. It grows a real specimen with the shipped makeSpecimen and points everything at the flower, which on the shipped page is a minor character wearing the foliage code.

It adds no growth code and no organ geometry. flDrawSpecimen is the shipped organ loop (minus occlusion cull and microscope) calling the shipped emitters; a Buffers subclass records per-organ [start,end) ranges per stream — organ identity the shipped buffers never carried — and captures veins the way the Blender bridge does (a, b, w0, w1, colour, emis, side dropped), expanded camera-facing per frame in the vertex shader. That is ROADMAP 11's "a ribbon as twelve floats", incidentally working.

The numbers (review currency)

• flowers/parity.test.mjs: 20 checks, 0 failures across Ember Creeper + Nightglass Parasol @ 1400 steps — captured streams reconcile with shipped drawSpecimen float for float (tri: shipped 152220 == captured 139500 + petal 12720; exact ribbon-count parity 8608 == 8608 for the mirrored vein pass), organ ranges monotone and covering, petal q recorded on every petal.
• node build.js && node test/smoke.mjs: 73 checks pass, canalisation.html byte-identical (this PR touches nothing in src/).
• Real browser (Metal): ~115-130 fps growing an Ember Creeper live, all streams dynamic every stepped frame.

The shading is transliterated, not improvised

First boot lit the plant with a guessed lighting model and got a silhouette — the palettes only mean what 60_render.js makes them mean. The scene now carries that pipeline line for line: hemisphere ambient, key at 0.9, back-transmission at 0.55, rim at 0.7, emissive x3, fog measured from the subject's near face, veins additive at colour x emissive with depth writes off, ACES → vignette → grain → gamma. ("Match the shipped look first" — it is now literal.)

The petal mechanisms, each from a published result, each reading a computed channel

The shipped petal is a flat card with a 5% droop, coloured by a gradient. The petal stream (16 floats/vertex) carries dd (distance-to-vein), q (floral identity — computed since the whorl work, never drawn), u, v, dev, lib:

• Anthesis is a bifurcation (Liang & Mahadevan 2011 [D]): a petal is a curved shell; edge growth strain past beta* unbends the longitudinal curvature and rolls the lateral one. A bud is a cup and a bloom is a recurved petal because the equation says so — no animation curve. Driving the load with the organ's own dev is the one [OURS] coupling; high q lagging the bloom gives SCIENCE.md's unbuilt "enclosing growth at high q" a mechanism.
• Edge ripples (Cerda & Mahadevan 2003 [D], experimentally verified): wavelength and amplitude in closed form from a 150 µm petal (published range 75-419 µm) at the world's own 0.0625 m/unit, with the lily's measured base-to-tip strain rise. Drawn at the millimetres the formula returns. Veins are mapped through the same displacement (first build had them floating off the curved surface).
• Translucency (van der Kooi 2016 [D]: petal transmittance usually exceeds reflectance): back-transmission boosted where tissue is far from a vein (dd as thinness is [OURS]), squared-albedo transmission tint.
• Conical-cell velvet (Ren 2017, Gorton & Vogelmann 1996, Noda 1994 [D]): ring-NDF sheen tilting 22°→52° with cell maturity; saturation deepens at constant pigment because the cone steers light into the vacuole.
• Bullseye (Todesco 2022 [D], n=1589): one threshold on u per specimen, drawn from the published trimodal distribution.
• Spots (Ding/Yuan 2020 [D]): their Gierer-Meinhardt system, complete published parameter set verbatim, run on the petal's own lamina lattice — zero-flux free on the cut lattice, arrangement inherits the real tissue, never twice the same. Baked once per library petal (~50ms, cellTable category).

Not built, deliberately: the diffraction blue halo (RGB can only fake it), nyctinasty (needs a temperature cycle — a genuinely new global, ROADMAP 0z1's argument, left for a session that wants it). flowers/README.md carries the full provenance table including every stated constant.

Try it

node flowers/build.js && open flowers.html          # Ember Creeper from seed, live
open 'flowers.html?species=Nightglass%20Parasol&ff=850&focus=flower'   # fenestrated petals
node flowers/parity.test.mjs

?species= ?seed= ?speed= ?ff= ?focus=flower ?hold=none. A local shot is at shots/flowers_ember_portrait.png after a capture run.

Process notes

Two Opus 5 subagents did the groundwork: an engine map (which found q thrown away on petals, dd computed and used only as a tint, and pedicel written-never-read) and a petal-morphogenesis literature sweep with [D]/[I]/[OURS]/⚠ flags whose implementation shortlist is what the mechanism layer follows. Framing bugs were fixed by measuring (flower framed from its drawn bounds after two hand-derived versions put the camera inside the corolla).

🤖 Generated with Claude Code

[aj-dev-smith]

Round 2, from AJ's look: "more magnificent, and the flashy things near the petals are glitchy."

Three commits. The numbers-as-review-currency version:

The glitch ran four layers deep, each isolated in the live shader before fixing — inner organs fed through shell physics whose curvature scales 1/L (a 0.4-unit stamen-analog rolls into a crumpled tube; petals only now); the Cerda–Mahadevan ripple landing 2–3 mesh samples per wavelength and aliasing (Nyquist fade in flPetalForm — the vein-LOD law applied to a ripple); the pigment stack at full strength turning a smooth pale bake into maroon blotches (raw-albedo isolation showed the bake was beautiful alone; the stack whispers now); and a one-sided lambert reading as meat on cupped petals (thin tissue is lit from both sides now — van der Kooi applied to the model, not bolted on).

The post chain is 60_render.js whole now — bright + 3 widening blurs, half-res defocus with depth-in-alpha, COMP_FS verbatim. UnrealBloomPass is gone (vendor 53 kb lighter) and the piece finally has the shipped depth of field, racking on the shipped director's focus law.

The structure ask became the C-class mutant. Four engine knobs, every default the literal that was hardwired (smoke 73/73, herbs bit-identical): apexRenew (WUS shutoff failing in consumeApex — Lohmann/Lenhard 2001), floralElong/floralStretch/floralNode (a flower is a compressed shoot), plus petalGrade (a petaloid stamen is smaller). Measured: 6 petals/flower wild → 14 at renewal 0.6 → 23 at 0.8; corolla spread 8.4–13.0 units loose → 0.26–0.40 packed. Three traps each silently un-double the flower back to 9 petals (founding gate, floralGrace, and the two elongation terms not moving together) — all measured, all in the commit messages.

?form=double is the default now; ?form=wild is the shipped configuration. Emergent bonus nobody designed: the Nightglass Parasol's compressed corolla wraps its terminal fruit — a glowing moon in a ring of fenestrated petals.

Record shots in shots/: flowers_ember_double.png, flowers_parasol_double.png, flowers_fern_double.png (the Fern shot is also the known gap — flower focus has no sight-line cull yet, README carries it).

Also fixed: bestFlower() truthiness dropped axis 0 (the terminal flower) to the wide shot; pollen grains fade in/out instead of popping at the anther.

🤖 Generated with Claude Code

[aj-dev-smith]

Round 3 — the flower is four organs now

AJ asked for significantly more complex structure. The answer was already in the engine: floral identity q was computed per organ and then collapsed to one binary (petal / not). This round reads it properly.

The full ABC model is the default form (?form=abc). whorlBands cuts q into sepal / petal / stamen / carpel — and everything follows from the band with almost no new machinery: a sepal takes a leaf from the ordinary library (the non-petal request already handed one out — A-class alone IS a leaf-like organ), a stamen is a small anther on an elongated stalk (org.stalkX, read by petioleOf, so the bending physics and the stem's load see the same filament the renderer draws — and the pollen stream now sheds from named anthers), and the carpel is a pale green style at the crown, standing where the meristem consumed itself to nothing.

Engine knobs added (40_plant.js / 39_fall.js), every default the shipped literal — smoke 73/73, parity 20/20, herbs bit-identical:

• whorlBands — the ABC banding, null everywhere shipped
• org.stalkX — filament/style elongation, 1 everywhere shipped
• floralDome — cap the converted dome at a characteristic size in founder-patch radii. Measured: Ember's working axillaries convert at exactly 3.0 organR; an uncapped terminal founds ten organs at q≈0 and reads as one whorl. A multiplier was tried first and provably cannot serve both — it fixed the terminal and collapsed the axillaries to 1-3 organs.
• zygomorphy — CYC/DICH (?zygo=0.85). Adaxial = the horizontal negation of the axis's first segment (the direction the bud grew out of its parent), so nothing is stated — and a terminal flower has no reference and stays radial, which is real peloria arriving free. Dorsal petals enlarge and stand, ventral reflex, the dorsal stamen aborts.

Two floral programs assigned per species from a measured 8-species sweep — Cathedral Fern is S8 P1 A0 C1 under program A and S3 P5 A4 C1 under B (smaller dome, more renewal, bands shifted to its measured q clusters). Spiral Ossuary founds 3 floral organs even wild; its flowers were always inconspicuous and stay that way honestly.

Nectar guides: vDD (distance-to-vein, already in the petal stream for translucency) read as pigment — Venosa is an MYB active only in cells overlying veins [D], so the guide pattern IS the vein network, converging on the throat.

The flower shot is a photographer now: it eases to the flower's own facing blended away from the trunk, scores down corollas pressed against the trunk (one was framed with the trunk through its face — 0.36 units off a 2.13-radius corolla), and ports the shipped sight-line cull (the round-2 known gap) with one extension the shipped director never needed: the window reaches one organ-length past the subject, because face-on sits inside the canopy. The cull is view-dependent, so capture re-runs on camera motion — at speed=0 it used to be computed once from wherever the camera began.

Record shots in shots/: flowers_{ember,fern,parasol}_abc.png, flowers_ember_zygo.png. The Fern one is the picture: a pink fruit ringed by five blue petals with white guides, anthers between, the canopy defocused behind.

Fastest look: open 'flowers.html?species=Cathedral%20Fern&seed=5&ff=5200&focus=flower'

🤖 Generated with Claude Code

[aj-dev-smith]

Round 4: the capitulum — a daisy is a hundred flowers

?form=daisy grows a composite flower head, and it is the round-3 machinery read at a different scale rather than new identity code. The same four whorl bands land on Asteraceae anatomy exactly: the sepal band is the involucre's phyllaries, the petal band the ray florets, the stamen band the disc florets — which is why the disc glows and sheds pollen into the wind: those organs were already anthers — and the centre ripens a fruit.

Two engine additions (both default-0, shipped bit-for-bit; smoke 73/73, parity 20/20)

• receptacle un-collapses the floral dome. q records the radius each organ was founded at (1 − prim.r/floralR0); placing every floret on the axis line threw that away. Under the knob a floret's base is offset by receptacle·(1−q) — rim founded first, centre last, the head becomes the disc the meristem actually was. Measured: 52-53 florets per axillary head (S~17 P5-13 A16-29, against a real daisy's 13-21 rays).
• A receptacle does not elongate. Under the same knob every floret rides the tip in elongate(), so the disc stays one station — 2.9 axis-units of smear collapse to 0.00 — while floral elongation below it becomes the peduncle: the daisy bolts a 9-10 unit scape, exactly what a real one does. tropism 0.002 (vs the herb's 0.02) holds the head ~3 units clear of the trunk; at 0.02 a 12-unit scape climbs the trunk and ends 0.4 units from it (both measured).

The sterile corner (a shipped bug older than this piece, fixed here)

The daisy's centre would not fill, and the diagnosis ran five layers deep: every fruit on the specimen was barren. Control first — wild Ember seed 21, zero overrides, is barren too, in flowers.html and the shipped app, deterministically. Isolated to makeSpecimen's per-seed fruit chemistry: ovule patterning on the 642-cell shell fails outright when T/D < ~5.3 — mapped on a 6×5 T-D grid, 6 fruit seeds per cell, the boundary is sharp (12-20 ovules one side, zero the other) — and the shipped draw ranges include that corner, so 5 of 40 specimen seeds were sterile for life: no ripening, ever, a seed head with nothing in it. One clamp after the draws (D ≤ T/6, ~15% margin over the measured cliff) fixes it with fertile specimens bit-identical and the PRNG call order untouched. Every daisy head now ripens a 14-16-ovule fruit at the disc's centre, measured to step 9500.

Also in this round: the flower shot reads the tip tangent rather than the base-to-tip chord (a bolting scape curves), and the away-from-trunk blend tapers with the subject's own measured clearance — a head already held out keeps its facing.

The fastest look

open 'flowers.html?form=daisy&species=Ember%20Creeper&seed=21&ff=7000&focus=flower'

Record shots in the worktree: shots/flowers_ember_daisy.png (ripe head face-on — spotted crimson rays, pale disc florets, the fruit glowing at the centre) and shots/flowers_parasol_daisy.png (the Nightglass nocturne — indigo rays under the moon-fruit). Ember seed 21 is the demo seed; daisy quality is seed-dependent (seed 5 grows one weak terminal head).

🤖 Generated with Claude Code

[aj-dev-smith]

Round 5: the columbine — `?form=columbine`. AJ asked whether we could reach Aquilegia-level complexity; the answer decomposed onto the existing machinery plus one genuinely new mechanism, and the new one comes with hard numbers.

The nectar spur (`flowers/12_form.js` + `15_petal.js`). Puzey, Gerbode, Hodges, Kramer & Mahadevan (Proc. R. Soc. B 279:1640, 2012 `[D]`) — the same Mahadevan as the pitchfork bloom and the edge ripples, so the whole corolla now runs on one lab's physics — showed spur-length diversity across the whole genus is achieved solely through cell-shape anisotropy. Built as a domain map with the flat sheet as its limit:

• the petal's proximal sheet domain rolls closed (`th = piwtN`; margins meet at a seam; identity at w=0, so the mouth is a funnel, not a join)
• the tube elongates backward by the anisotropy factor on the same dev coupling the bloom uses — aniso 6 grows a tube 1.8x the blade
• slenderness is the anisotropy too: lateral scale `1/sqrt(a)`, between the paper's width-conserving and area-conserving bounds. Both ends measured: at 1 the spur is a sausage (lateral span 0.40 vs 0.425 at the mouth — the margin holds its width until the last 2% of the sheet); at 1/a it is a wire the rasteriser reduces to its own veins
• the veins map through the same displacement, so each spur arrives with its own vasculature ending at a softly glowing nectary
• spur angle `pi + petalTilt` maps every petal's tube anti-parallel to the flower axis — a parallel descending ring, the columbine silhouette. Measured at three angles: +0.35 converges under the fruit, −0.85 bundles over the crown.

Petaloid sepals (`20_draw.js`): `sepalPetaloid` is the double's homeotic lever pointed the other way — B-class into whorl 1. Sepals go through the petal stream wearing the species' full petal colour while true blades pale toward cream (held at 0.35/0.48: past ~0.5 the transmission lobe `alb²·1.5` exceeds 1 and the blades blow white through the bloom chain).

The program (measured, col_probe): renew .70 + dome 3 + cap 32 founds S7 P8 A5-9 per flower — the q-zero founding pile IS the sepal whorl. A small `receptacle` (0.25) reuses round 4's mechanics as the pedicel: trunk clearance went 0.24 → 1.25 and the director can finally photograph an axillary. And the photographer learned that a spurred flower is shot in three-quarter — face-on, a columbine reads as a mallow (measured).

Zero engine edits this round: parity 20/20, smoke 73/73 untouched.

Fastest look:
```
open 'flowers.html?form=columbine&species=Ember%20Creeper&seed=21&ff=3400&focus=flower'
```
Records: `shots/flowers_ember_columbine.png` (the spurred crown on its curving pedicel), `shots/flowers_parasol_columbine.png` (the nocturne — moon-white pistil, indigo lace sepals).

🤖 Generated with Claude Code