How much water

A pump on a relay needs two numbers nobody was giving it. water() recorded that a watering happened and said nothing about its size — fine for a person with a watering can, useless for a pump.

The species is not the number that matters

The obvious design is a table of millilitres per archetype. It is wrong by an order of magnitude, not by a little:

dose( { archetype: 'tropical', pot: { diameterCm: 12 } } )   //   80 ml
dose( { archetype: 'tropical', pot: { diameterCm: 40 } } )   // 2986 ml

Same plant, same archetype, thirty times the water. What the archetype decides is the fraction of substrate to wet and how far it may dry back; the pot decides the volume. Both are required, and no pot size is assumed:

That fraction is the part the plant decides; the volume is the part the pot decides […] a default would be wrong by that same factor.

So { pot: { diameterCm } } belongs in the plant's configuration, and the diagnosis asks for it when it is missing.

	fraction	dry back to
cactus	0.20	0.08 — almost every dead houseplant cactus was overwatered
aroid	0.30	0.45
fern	0.45	0.70 — the one group where letting it dry once is a real setback
woody	0.40	0.35 — a shallow watering wets the top and leaves the roots that matter dry

Seconds are measured, not looked up

Flow rate depends on this pump, this tube and this head height, and it drifts as the tube ages. Without a calibration the system gives the volume and refuses the duration, which is the honest half rather than a plausible whole.

And it is open-loop, which is the dangerous part

A blocked line, an empty tank and a tube that has fallen out of the pot are identical from the relay, and each records a watering that did not happen:

plant.confirmWatering( soilBefore )
// { arrived: false, fault: true,
//   why: 'The pump ran and soil moved 1.0 points, which is not a watering. […]
//         a second dose on a blocked line does nothing except record a second
//         watering that did not happen.' }