The last lesson taught you why crossing a threshold feels sudden — nothing, nothing, nothing, then everything. This one asks the follow-up question that should be keeping you up at night: once you’ve crossed, can you cross back? Sometimes yes, cheaply and instantly. Sometimes the door only swings one way, and getting back to where you were means shoving the system far, far past the line you originally tripped over. That asymmetry has a name, and once you can see it you’ll stop assuming you can always “course-correct later.”
The word for a one-way door: hysteresis
Here’s a fact about your light switch that you’ve never noticed because it’s too obvious. Push the toggle gently and nothing happens; push a little harder and it snaps to ON — and now, sitting there at ON, that same gentle finger-pressure won’t snap it back. To turn it OFF you have to push it the other way, past its own little ridge, with real force. The switch’s current state doesn’t depend only on how hard you’re pushing right now; it depends on which way you came from.
That is hysteresis: the state of a system depends on its path and history, not just on the current conditions. Formally — this word comes from physics and materials science, where it describes magnets and metals that “remember” how they were last pushed — the level of some driving condition at which a system flips one way is not the same as the level at which it flips back. The two thresholds are offset. So to reverse a flip, you can’t just return the conditions to the tipping point; you have to overshoot it, sometimes by a huge margin, coming from the other side. The system, in a real sense, remembers which side of the line it’s on.
Tie this straight back to the engine of this whole course. A reinforcing feedback loop — the loop whose gain k crossing 1 drove the tip in the first place — doesn’t switch off once you’re over the top. It keeps running. And now it runs in the new state’s favour: the very same amplification that pushed the system across the line now defends the new side. The loop that tipped you becomes the loop that traps you.
Before you read — take a guess
A shallow lake gets murky and choked with algae once fertiliser runoff pushes nutrient levels past a tipping point. A town cuts runoff back to exactly the level the lake had *before* it turned murky. What's the most likely result?
The ball in the double valley
The cleanest picture of hysteresis is a ball rolling on a landscape with two valleys separated by a ridge — a “double-welled” landscape. The ball at rest sits in one valley. That’s a stable state: nudge it a little and it rolls back down to the bottom. To move it to the other valley you must give it enough of a kick to carry it up and over the ridge. Once it tips over the crest, it rolls down the far side and settles into the second valley — and now that’s home. It’s stable there too.
Here’s the asymmetry that is the entire point. Suppose the kick that got the ball over the ridge came from tilting the whole landscape rightward. To get the ball back, it is not enough to un-tilt the landscape to level — the ball is now sitting in the right-hand valley, and level ground leaves it happily there. You have to tilt the landscape the other way, hard enough to shove the ball back up and over the ridge from the far side. The push-out condition and the push-back condition are different, and the gap between them is exactly the height of the ridge. Level ground — the original neutral condition — is no longer a place where the system changes its mind.
Two stable states, one set of conditions
The signature of hysteresis is that the same conditions can support two different outcomes, and which one you get depends on your history. Level ground can hold the ball in the left valley or the right valley. Cool temperatures can support ice sheets or open water. A market can run on QWERTY or on a better layout. History picks the winner, and the reinforcing loop then defends it.
Worked example: ice, albedo, and the point of no return
Ice is white, and white things bounce sunlight back to space — that reflectivity is called albedo. Open water and bare dark ground are the opposite: they absorb sunlight and warm up. Now assemble the loop. Warming melts a bit of ice; the newly exposed dark water absorbs more heat; that extra heat melts more ice; which exposes more dark water. This is the ice–albedo feedback, and it’s a reinforcing loop with a threshold — a critical mass of warming past which it runs away on its own.
The hysteresis is brutal. Suppose a summer of unusual warmth melts an ice sheet past its tipping point and it collapses to open water. To grow it back, you cannot merely return temperatures to the value they had when it melted. At that temperature the newly dark ocean is now absorbing heat and staying warm — the loop is defending the ice-free state. Re-freezing a lost ice sheet requires sustained temperatures far below the original melting point, held for a very long time, to overcome all that absorbed warmth and re-establish a reflective surface. The push-out and push-back thresholds are wildly different. This is why climate scientists speak of a point of no return: not because reversal is forbidden by physics, but because it demands overshooting so far in the other direction that, on any human timescale, it may as well be.
Worked example: the fishery that can’t recover
A fish stock has a happy secret: at healthy population sizes it grows faster per fish, because there are plenty of mates, safety in numbers, and enough adults to defend the young. But that same feature bites in reverse. Below a certain critical breeding population, the survivors are so sparse that they struggle to find each other at all, the per-fish growth rate collapses, and the stock can spiral toward zero. Ecologists call this the Allee effect — a reinforcing loop where being few makes you fewer.
So imagine overfishing drives a cod stock below that critical mass and it crashes. The intuitive fix is “stop fishing and let it bounce back.” But the pre-tip world and the post-tip world need very different conditions. Before the crash, a modest catch limit was plenty to keep the stock healthy. After it, even a total fishing ban — zero catch — may not be enough, because the problem is no longer fishing pressure; it’s that the population is now too sparse to reproduce its way out of the hole. The Grand Banks cod fishery off Newfoundland famously collapsed in 1992 and, despite a moratorium, has taken decades to show meaningful recovery. Prevention would have cost a small quota. Cure cost an entire industry and a generation of waiting.
Worked example: standards lock-in and moats
Look down at your keyboard. The top letter row spells QWERTY, a layout designed in the 1870s partly to slow typists down so mechanical typewriter arms wouldn’t jam. The jamming problem vanished a century ago. Demonstrably faster layouts exist. And yet here we all are, still on QWERTY — because the value of a keyboard layout isn’t in the layout, it’s in the fact that everyone else uses it too. That’s a network effect (lesson 3), and it’s the reinforcing loop that creates hysteresis in standards.
Going in, the network effect builds the standard: each new QWERTY user makes QWERTY more valuable to the next. But once it dominates, the very same loop turns and defends it. A challenger — even a genuinely better one — can’t get a foothold, because being better isn’t enough; it also has to be worth abandoning everyone you’re compatible with, retraining every typist, and rewriting every manual. The network effect that a challenger needs is currently working entirely for the incumbent. This is path dependence: where you end up depends on the accidents of how you got started. It’s the story of VHS beating the (arguably superior) Betamax, of dominant platforms that outlast better rivals, and it’s precisely where durable moats come from — a moat is hysteresis working in the incumbent’s favour.
Worked example: the habit you let slide
The same physics runs at the scale of a single human. A daily routine held above its critical mass — the gym three mornings a week, the language practice, the tidy inbox — runs largely on autopilot. The loop is reinforcing: doing it makes it feel normal, and feeling normal makes it easier to do again. Momentum defends momentum. Maintenance is nearly free.
Now let it fall below the threshold. Skip a week, then two. The autopilot disengages, “normal” resets to not doing it, and restarting no longer costs the trivial energy of maintenance — it costs the large, deliberate energy of pushing the ball back up over the ridge from a standing start. Anyone who has rebuilt a fitness habit after a lapse knows this in their bones: keeping a good loop running is far cheaper than restarting it. The gap between those two costs is your personal hysteresis loop.
Reversible vs. locked-in
Not every tipping point is a one-way door. The crucial skill of this lesson is telling apart the tips you can walk back from the ones you can’t. The difference is whether a reinforcing loop clicks on to defend the new state after you cross.
| Reversible — tips back easily | Locked-in — hysteresis, hard to reverse |
|---|---|
| A small campfire you can still stamp out | A forest fully ablaze, spreading tree to tree |
| A rumour caught and corrected within the hour | A rumour that’s become “common knowledge” |
| A pot of water just below boiling — cool it and it’s fine | An ice sheet melted past the albedo tip |
| A fish stock lightly overfished for one season | A stock crashed below its breeding critical mass |
| A new app you can still uninstall before it hooks you | An entrenched platform standard with full network effects |
| A habit skipped once | A habit lapsed for months, its autopilot gone |
| A ball resting on a gentle slope — let go, it returns | A ball kicked over the ridge into the second valley |
Read the pattern down the columns. The left side hasn’t yet switched on the defending loop — the fire is still small enough to lose, the rumour hasn’t self-propagated, the ball is on a single slope with one bottom. The right side has crossed into a second stable state that its own feedback now protects. Same underlying model; the question is always: is there a loop that will fight to keep the new state?
Which feature best distinguishes a hysteretic (locked-in) tipping point from a merely reversible one?
Sort each tipping point by how hard it is to walk back. Ask: once it flips, does a reinforcing loop switch on to defend the new state?
- A rumour caught and corrected within the hour
- An ice sheet melted past the ice–albedo tipping point
- QWERTY entrenched by full network effects
- A months-long habit lapse that killed the autopilot
- A fish stock crashed below its breeding critical mass
- A habit you skipped exactly once this week
- A campfire still small enough to stamp out
- A new app you can still uninstall before it hooks you
The practical asymmetry: prevention beats cure
Everything above collapses into one rule for acting near a hysteretic threshold, and it is worth more than any clever recovery plan.
Near a one-way tip, 'undo' is not sold at the price of 'don't'
When a threshold shows hysteresis, the cost of reversing a flip is not the cost of avoiding it — it’s dramatically higher, because you must overshoot all the way back past the tipping point against a loop now working against you. Sometimes reversal is impossible on any timescale you care about. So the decisive move happens before you cross: respect the point of no return while you’re still on the safe side of it. Prevention isn’t merely cheaper than cure here — it may be the only option that exists. A small quota beats a collapsed fishery; a maintained habit beats a rebuilt one; a degree of warming avoided beats an ice sheet that won’t come back.
This is why “we can always fix it later” is such a dangerous sentence near a hysteretic tip. “Later” is on the other side of a ridge you’ll have to climb from the wrong direction, and the toll is set by the height of that ridge — which you don’t get to negotiate.
The hopeful flip side: hysteresis working for you
Now the good news, because this coin has two faces and the physics doesn’t care which way you point it. Everything that makes hysteresis a trap when you fall into a bad state makes it a gift once you’ve pushed something good past its critical mass. Get a habit over the top and its autopilot now defends you — the reinforcing loop keeps it running cheaply, and skipping feels wrong instead of doing feels hard. Get a network past critical mass and the network effect that fought you now guards your moat against every challenger. Get a saved species back above its breeding threshold and its own growth loop takes over the work of keeping it alive.
The lesson isn’t “hysteresis is bad.” It’s that hysteresis is a ratchet — it locks in whatever state you last pushed the system into, good or bad. The whole art is choosing the direction before you spend the effort. Pay the one-time cost to shove the ball over the ridge into the valley you actually want, and the landscape will hold it there for you. Same physics; you just picked which valley.
Pitfall: the two mirror mistakes
There are two ways to misread a tipping point, and they’re mirror images.
- Assuming every tip is reversible — the “we can always course-correct later” error. You cross a line expecting a symmetric door and discover the reinforcing loop has bolted it behind you. This is how ecosystems get pushed past a point of no return by people who genuinely believed they could ease off in time. The fix: before crossing, ask whether a loop will defend the far side. If yes, treat the threshold as one-way and let prevention govern.
- Assuming every tip is permanent — the mirror error, treating every setback as an irreversible catastrophe when some tips really do tip back. Not every fire is unstoppable; not every habit slip is the end; not every murky lake is beyond restoring. Panicking that a reversible flip is permanent wastes the cheap early window when you could still just stamp out the campfire.
The skill that threads between them is the one question this whole lesson has been circling: is there a reinforcing loop that defends the new state? If there is, the tip is hysteretic — respect the point of no return and prevent. If there isn’t, it’s reversible — act early and calmly, because you can still walk it back. Getting that diagnosis right, case by case, is the whole game.
Link each idea from this lesson to its precise definition.
Check yourself: one-way doors
What single feature turns an ordinary reversible threshold into a hysteretic, one-way one?
Check your answer to continue.
Where this goes next
You now hold the sharp end of the critical-mass model: thresholds don’t just flip suddenly, some of them flip permanently, and the tell is whether a reinforcing loop defends the far side. Prevention beats cure near a one-way door — and the same ratchet becomes a gift the moment you aim it at a state you want.
Next up, lesson 6 — Where the Model Lies, the last teaching lesson before the exam. Every model has an edge where it stops being true, and this one has three: the trap of extrapolating a curve straight toward a threshold that isn’t there, the twin errors of mistaking sub-critical for dead (and dead for merely sub-critical), and the seductive habit of expecting everything to tip when most trends just quietly die below k = 1. Learn where the model lies and you’ll know when to trust it — which is the difference between a tool and a superstition.