The Verification Venue · a reflex, and a claim that outran it

The Plant That Stopped Flinching

Brush a Mimosa pudica and it folds shut in a second — no nerves, no muscles, no brain. A real electrical spike does it. Keep pestering the same plant with a harmless jolt and it stops bothering to fold. That second fact got retold, and retold, until "a leaf that stopped flinching" became "a plant that remembers." Here is the machinery of the first, exactly — and the precise seam where the evidence for the second runs out.

Two things are true and they are not the same size. The fold is settled physics you can compute. The memory is a live scientific argument with a rebuttal and a reply still on the table — plus a flashier cousin, a claim that plants can be trained like Pavlov's dogs, that failed when someone larger tried to repeat it. This page keeps all three apart so each claim can stand on its own evidence.

01 The mechanism — no nerves, no muscles

Fire the spike, watch the leaf go soft

Touch a leaflet. A receptor potential at the contact point crosses a threshold and triggers a genuine action potential — the same kind of all-or-nothing electrical event a neuron fires, here running through ordinary plant tissue. It travels down the leaf and reaches a swollen hinge at the base of each leaflet and each leaf: the pulvinus. When the spike arrives, the hinge lets go. Press the button and run the real clock.

Pulvinus mechanism simulator

touch →pulvinus hinge
at rest — leaf open

Spike reaches the hinge after

1.60 s

= distance ÷ conduction speed

Osmotic pull the ions were holding

0.50 MPa · 4.96 bar

= the turgor the extensor cell loses

How far the spike must travel down the petiole to reach the motor hinge.

Measured at 2–3 cm/s (Fromm & Lautner 2007; Volkov 2010) — slow for a nerve, fast for a plant.

K⁺ and Cl⁻ flood out through voltage-gated channels; water follows.

A fixed measured value (~150 mV, ~5 s wide) — shown here, not tunable, because it's a fact not a knob.

1

Receptor potential

The touch depolarises the membrane at the contact point past a threshold.

2

Action potential fires

An all-or-nothing ~150 mV spike, ~5 s wide, propagates at 2–3 cm/s toward the pulvinus.

3

Voltage-gated channels open

Arrival at the hinge throws open K⁺ and Cl⁻ channels in the extensor motor cells.

4

Ions leave for the apoplast

K⁺ and Cl⁻ pour out of the cell into the wall space, dragging the cell's osmotic pressure down.

5

Water follows, turgor collapses

Water leaves osmotically; the extensor cells go slack while the upper flexor cells stay firm.

6

The leaf drops

The turgor imbalance across the hinge swings the leaf stalk down; the leaflets fold upward. No nerve, no muscle: a hydraulic collapse.

That is the whole trick and there is nothing mystical in it. The plant has no muscle to contract and no nerve in the animal sense; it moves by letting go of water pressure on one side of a hinge. The electrical spike is the trigger, potassium is the currency, and osmosis is the engine. Two numbers on this page are computed live from the sliders and nothing else:

The spike's travel time is just distance over speed. The turgor the cell loses is just the osmotic pressure of the ions that walked out the door.

The check — the two live numbers, from first principles

Travel time is t = d / v. Osmotic pressure is the van 't Hoff relation Π = i·c·R·T — the pressure a dissolved solute exerts (i=2 for KCl splitting into K⁺ and Cl⁻, R=8.314, T=298 K). For your current sliders:

t = 4.0 cm ÷ 2.5 cm/s = 1.60 s
Π = 2 · 100 mol/m³ · 8.314 · 298 = 495 514 Pa = 0.50 MPa = 4.96 bar

A motor cell's resting turgor is of order 0.5–1.0 MPa, so dumping the osmotic pull of ~100 mM KCl is enough to slacken it — the arithmetic and the collapse line up. The 150 mV / ~5 s / 2–3 cm/s spike figures are measurements, not things you can derive from a desk; the verifier range-checks them against the primary reports and recomputes everything else. Run it: node research/the-plant-that-stopped-flinching/verify-the-plant-that-stopped-flinching.mjs → 26/26 PASS.

What's solid here, and what's still open at the molecule

Solid. The macro-chain (mechanical stimulus → receptor potential → propagating action potential → voltage-gated K⁺/Cl⁻ efflux → osmotic water loss → extensor turgor collapse → fold) is the consensus mechanism, laid out in the 2020 Plants review, which states "plants do not possess the nerves and muscles that facilitate movement in most animals." The action potential is a real, membrane-based, all-or-nothing event; calling it "electrical signalling" is not a metaphor.

Idealised. The animation is a schematic of one hinge, not a finite-element model. Real pulvini have both extensor and flexor populations whose antagonism sets the resting angle; we draw the extensor collapse and hold the flexor firm, which is the direction of the effect but not a full mechanical solve. The Π figure is the osmotic pressure of the moved ions in isolation (van 't Hoff, ideal-dilute), an order-of-magnitude bookkeeping of the turgor at stake, not a measured pressure trace.

Still open at the molecule. The exact primary mechanoreceptor — what physically senses the touch — and the precise identity of some channels are active research (mechanosensitive-channel work, e.g. 2021 Plant Physiology). "A channel opens" is safe; "this named gene is the touch sensor" is not yet. The 2–3 cm/s and ~150 mV figures also vary by tissue (leaflet pinna vs. main pulvinus) and method; they are the primary measurements, not a universal constant.


02 The harder question — is it learning?

"It's not learning, it's just a reflex"

Now the interesting part. Drop a potted Mimosa a short, harmless distance and it folds. Drop it again. And again. After a while it stops folding — it has apparently decided this particular jolt is nothing to worry about, while still snapping shut if you genuinely threaten it. In 2014, Monica Gagliano and colleagues reported exactly this, and that the plants stayed uninterested for up to ~28 days (Oecologia 175:63–72).

The reflexive dismissal is "that's not learning, it's just a tired hinge." But here's the catch that makes this worth an hour of your attention: "it's just a reflex" is not, by itself, a refutation. What Gagliano claimed is habituation — the simplest recognised form of non-associative learning — and habituation is such a low bar that even brainless single cells clear it. The protist Stentor stops contracting to a repeated poke. "It has no brain" refutes nothing, because habituation never required one.

So the real question is narrow and technical, and it is genuinely unsettled: is the decline Gagliano saw actually habituation, or is it motor fatigue? A tired hinge and a habituated one both stop responding. Learning theory has a standard toolkit for telling them apart — the Thompson & Spencer (1966) criteria. Two of them do the decisive work. Open the scorecard.

Where the evidence actually stands

Of the two criteria that separate a learning brain-free system from a merely tired one, dishabituation — where a sudden strong, novel stimulus should snap the response back — was never demonstrated, and the control that would nail down stimulus specificity was, Biegler argues, not adequately run. Without those two, most of the 2014 data are equally consistent with a hinge that simply ran low on the water pressure to keep folding.

But not all of it. Biegler says the faster decline in low light is compatible with exhaustion. His qualified reservation concerns the 28-day light swap: plants switched in opposite directions between high and low light folded to the same extent. That may conflict with exhaustion, he says, but a ceiling effect in leaf openness could explain it. The question remains open, and Gagliano's 2018 reply, "Plants learn and remember: let's get used to it", is still a reply worth reading, not a rout.

Habituation is a low bar even a brainless cell can clear. The open question isn't whether a plant could learn this way — it's whether this decline, in this experiment, is learning or a tired hinge. The precision is the whole point.


03 The evidence: three claims kept separate

How far each claim actually reaches

There are three separate claims in this story, of three completely different strengths. Two of them are about different species. The ledger below keeps them apart. Open each to see exactly how far its evidence reaches, and where it stops.

The headline, and the separate pea claim

On 16 January 2014, Sci.News ran this headline about the Mimosa drop study:

"Mimosa Plants Have Long Term Memory, Can Learn, Biologists Say"

✓

The month-long result traces to the 2014 Mimosa habituation result. Real experiment, but contested (Biegler's rebuttal, Gagliano's reply, no independent replication) and it's retention tested to ~28 days, not "memory" in any rich sense.

✗

The separate pea claim concerns associative learning in the 2016 pea study, a different species and a much stronger claim, which failed to replicate in a larger study (Markel 2020, p = 0.387). It is not part of this headline.

✗

Learning and memory are the researchers' interpretation of the Mimosa response decrement. The dispute is whether that decline was habituation or fatigue; the headline does not settle it.

The check — every claim tagged to its live source

The three statuses (established / contested / failed replication), the two-species firewall, the nine Thompson & Spencer criteria, Biegler's three prongs, and Markel's 203 plants across four main groups; F vs L comparison n=82, p=0.387 are all enumerated and checked in research/the-plant-that-stopped-flinching/. Nothing on this page asserts a status the ledger can't source.

The honest boundary lines — read before you repeat any of this

The species firewall is load-bearing. The 2014 habituation work is Mimosa pudica. The 2016 associative-learning work is Pisum sativum — garden pea. They are different organisms, different experiments, and different strengths of claim. Any sentence that uses one to vouch for the other is wrong by construction.

Don't overcorrect into "plants can't learn." That would be its own error. Habituation is a textbook form of non-associative learning that aneural cells demonstrably show; nothing here rules out a plant habituating. The claim is precise: this particular 2014 dataset lacks the two tests that would separate its decline from motor fatigue, and the one associative-learning result that did claim Pavlovian conditioning failed independent replication.

Biegler's critique, stated straight. Three prongs: (1) no demonstrated dishabituation; (2) the experimental condition needed to establish stimulus specificity was missing; (3) motor fatigue adequately explains most — his word — of the data. He concedes the light-environment effect resists a pure-fatigue account. Caricaturing him as "it's all just fatigue" is not what he wrote.

On authority. The 2014 and 2016 studies are weighed here on evidence — Biegler and Markel are the load-bearing counters — not by pointing at the author's more mystical popular writing. Ad hominem is not an argument, and neither is treating any single researcher as the last word.

Corrections

Corrected 2026-10-14: The schematic description and chain said the leaflet swung down. The leaf stalk drops; the leaflets fold upward; found by the assay line, 2026-10-14. (Hagihara and Toyota 2020, section 2).

Corrected 2026-10-14: This page quoted the review as saying "no nerves or muscles". Its words are "plants do not possess the nerves and muscles that facilitate movement in most animals"; found by the assay line, 2026-10-14. (Hagihara and Toyota 2020, abstract).

Corrected 2026-10-14: The body and scorecard said Biegler treated faster low-light decline as evidence against fatigue and argued an energy-starved plant should tire more slowly. He treats that decline as compatible with exhaustion. His qualified reservation concerns the 28-day light-swap result, for which he suggests a possible ceiling effect; found by the assay line, 2026-10-14. (Biegler 2018, p. 35).

Corrected 2026-10-14: The page presented a composite with a Pavlov clause as a real headline and described it as coverage fusing two studies. The quoted Sci.News headline is "Mimosa Plants Have Long Term Memory, Can Learn, Biologists Say". Its 16 January 2014 article covers the Mimosa study and contains no Pavlov clause or pea result; found by the assay line, 2026-10-14. (Sci.News, 16 January 2014).

Corrected 2026-10-14: The scorecard attributed "Suggestive, not decisive on its own" to Biegler on potentiation. He instead lists characteristic 3, potentiation, among the tests the study omitted; found by the assay line, 2026-10-14. (Biegler 2018, p. 33).

Corrected 2026-10-14: The ledger gave about 203 plants for Markel's p=0.387 comparison. That F vs L comparison used 42+40=82 plants; 203 is the total across the four main groups (61+60+42+40); found by the assay line, 2026-10-14. (Markel 2020, Results and discussion).

Corrected 2026-10-14: The related-page note said both this page and the expanding-universe page paired a rebuttal with a reply. The linked universe page compares discovery evidence from multiple sources; the paired rebuttal and reply are on this Mimosa page; found by the assay line, 2026-10-14. (The Redshift Before the Law).

Corrected 2026-10-14: The scorecard marked stimulus intensity as claimed met using the light treatment. The study used a fixed 15-cm drop; changing light changed the energetic environment, not drop intensity; found by the assay line, 2026-10-14. (Gagliano et al. 2014, p. 65).

Corrected 2026-10-14: The scorecard marked frequency as not tested. The study compared 5- and 10-second intervals and reported an interval effect in low light (P<0.0001), but not high light (P=0.5). This was a limited comparison, not proof of the full criterion; found by the assay line, 2026-10-14. (Gagliano et al. 2014, pp. 65, 67).

Corrected 2026-10-14: The scorecard put "Not demonstrated in the 2014 study" in Gagliano's column. Gagliano and colleagues reported restored closure to shaking as dishabituation; Biegler argued that this did not demonstrate dishabituation under the proper definition; found by the assay line, 2026-10-14. (Gagliano et al. 2014, p. 66; Biegler 2018, pp. 33–34).

The claims record 212 claims re-read against their sources, 14 October 2026

Written 2026-07-03. Claims re-read against their sources on 2026-10-14: 212 checked, NaN confirmed, 33 wrong, 0 unverifiable. By the assay line (research/assay-line/cited-1014b/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex and a Claude agent.

This was the directed 2026-10-14 remediation pass on the 42-item triage in `research/assay-line/cited-1014b/the-plant-that-stopped-flinching/triage.json`, against layer files verified identical to read revision `d1d7401ad4` before editing. The fixer reread the evidence for every triage item, fixed 33 WRONG items spanning ten errors within this layer, retained seven listed MINOR items, and rejected two findings with evidence. Each triage item has its own claim line below. The remaining CONFIRMED entries carry the earlier checkers' source reads from N1-sources.json, N2-codex.md and N2-claude.md, once per source-record entry; repaired copies are explicitly restated after correction rather than carrying forward false wording. Ten dated correction lines appear on the page and at the end of the markdown body. V1 records one additional MINOR verifier repair required by the instruction to correct false assertions; it adds no page line and does not resolve or edit K4's listed page sentence. MINOR lines count with confirmed for the gate, without changing their verdicts. The 2026-10-14 repair review confirmed three remaining problems in copies outside this layer. Eight WRONG findings (X9, X10, X11, X18, X19, X22, X23 and K7) now carry open board notes, while retaining their completed local corrections. Claim counts are unchanged: 171 CONFIRMED + 8 MINOR + 33 WRONG = 212; findings comprise 26 fixed (25 WRONG and V1), 7 declined MINOR and 8 open WRONG.

This record is long (212 claim lines and 41 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.

The assay office: what a record is, and every layer re-read so far.