Tend: a plant you cannot draw, only tend — and the prune button that never pruned

[aj-dev-smith]

What this is

tend.html is a third page and the first one you do something on. You can't draw the plant. You can only do what a plant physiologist does, with the three instruments auxin was discovered through:

year	experiment	instrument	what the engine does
1880	the Darwins	Lamp	growing tips steer to normalize(up + k·toward_lamp), the exact equilibrium of additive sine laws, k = photoGain·I^0.4
1928	Went	the glow	every living tip's auxin drawn as motes running rootward at polar-transport speed
1933	Thimann & Skoog	Shears	the stream below a cut drains; the buds it held come free, race, and the first to commit puts the rest back to sleep
1933	their control	Auxin	put back on the cut, the buds stay asleep until you take it away

Every sentence the page says afterward is measured off the plant. Example: "The drain freed 17 of the 17 buds the tip was holding. One grew out, the first 1 cm below the cut, about 2 days after it, and put 16 back to sleep." Share copies a link that regrows your exact plant from its seed, cut for cut.

The bug it started from

The main page's "cut the apex" button has always frozen the plant it cut. prune() set the tallest axis to not-alive, and a not-alive axis returned from step() before its buds were ever looked at again. Measured before anything was written:

	before	after
Cathedral Fern, pruned at step 500	1 axis, 13 organs, forever	5 axes, 96 organs
Sun Coral, pruned at step 500	(same freeze)	7 axes, 96 organs

test/tend.mjs §2 is the regression test.

The science, and what the literature changed

docs/research_9_22_26_tend.md is a literature sweep, briefed before building and flagged [D]/[I]/[OURS]/⚠. It corrected the design in five places:

• Agar, not lanolin. Thimann & Skoog used agar blocks renewed every 6 h, and their buds grew "as soon as the application ... was stopped". The page copy is fixed.
• The depletion front decides commitment, not release. The first flush after a cut is sugar and cytokinin (Mason 2014). Auxin decides which freed bud commits (Morris 2005; Balla 2016). So a bud here is held, then free, then committed, and the page says the first flush isn't modelled.
• Instant commitment forced exactly one winner. A freed bud now exports on a ramp for tauCommit, and the one nearest the cut puts the ones below it back to sleep. Nothing states the winner.
• Transport speed is a lookup: 6.6× growth rate, Kramer 2011's median of 227 measured speeds.
• Light's gain goes as I^0.4 (Bastien 2015 fitting Galland 2002). It is not the saturating curve the draft used.

It also found Snow 1931 measuring inhibition rising with distance, so dominance is booked as a dial with the only data point against it. SCIENCE.md has the accounting table:

• lookups: patRatio, commitReach, photoExp;
• a genuine parameter: photoGain;
• dials: dominance and branching;
• numerical: BUD_BAND.

Nothing is added to the numbered imposed list. Cuts, auxin and the lamp are environment.

Numbers

Shipped path unchanged. Uncut specimens are bit-identical to main: 4 species × 2 seeds grown through senescence, with every stem point hashed every 400 steps, compared against a pristine copy of main's src/. smoke 73/73. views 23/23 (Cathedral Fern) and 35/35 (garden of eight).

node test/tend.mjs, all eight bench species × two seeds:

• every cut had held buds below it, and every one ended in a commitment (16/16);
• where several came free, fewer committed than came free;
• with auxin back on the cut, not one held bud came free (16/16);
• taking it away let one commit (16/16);
• the drain reaches the nearest held bud within one step of t_cut + d/v (predicted 700.23, freed at 701) and it commits exactly tauCommit later;
• the lamp's equilibrium lands within 0.05° of atan(k) at three irradiances.

node tools/tend_replay.mjs: a live browser session is shared as a ~500-character link and regrown in a second page to the same fingerprint of every stem point at the same step. The session has a dragged lamp, a cut, a cut dressed with auxin, and the auxin taken off. Checked on Ember Creeper and Cathedral Fern.

Cost: a bushy tended Ember Creeper (25 axes, 380 organs) runs at 1.5 ms a step and 7.6 ms to build geometry.

The plant's clock, calibrated

The drain front runs at patV world units a step. A real one runs at ~1 cm/h (Morris 2005). So one step of a Cathedral Fern is about an hour of a plant's life, and one second on screen is about five days. The page reports every interval in those plant hours.

What's in the change

• src/40_plant.js:
  • Axis.streamAt, releaseBuds (held, then free, then committed), budsPending and phototropicPull;
  • Plant.cut, paste and wipe, and prune() as a real cut;
  • a spent() that waits for a stump's buds.
  • All of it is off unless something asks (budField: 'apex', plant.light = null).
• src/70_app.js: a stemGlow hook in drawSpecimen and a program hook in makeSpecimen, both undefined elsewhere. src/80_main.js: the prune tooltip now says what happens.
• tend/: build script, program, the TendApp bench, the page, template and README. tend.html is the built artifact.
• test/tend.mjs: the harness, derivation first.
• tools/tend_shot.mjs (photographs the experiments), tools/tend_replay.mjs (checks a share link bit for bit), tools/tend_film.mjs (renders each experiment frame by frame to mp4).
• Docs: SCIENCE, JOURNAL, PITFALLS (six traps), ROADMAP (two follow-ons: run the canalisation switch for bud activation; the AC model for every tropism), CLAUDE.md, README.

What still needs a human eye

• The drain at 1× is about a second across a whole plant, so the page slows the clock and leans the camera in for ~3 s after a cut, and says so. Is the slow-motion window the right length?
• The motes' density (0.42 units apart) is the one purely visual number in the stream drawing.
• The bench's reach of dominance (18 units) is a dial chosen so a cut is legible. The untouched plant branches only near its base as a result.
• Hosting. tend.html is one self-contained file like canalisation.html. The Pages workflow only publishes the three.js demo, so putting it on the site is your call.

Clips for a post are rendered but not committed (shots/ is gitignored); node tools/tend_film.mjs shots/tend rebuilds them.

🤖 Generated with Claude Code