Persuasion and manipulation use the same techniques. The line isn't in the technique.

Mechanism: scarcity, urgency and loss signals are content-free — the brain's response to "it's running out" does not check whether it is actually running out. That is the fundamental reason manipulation is cheap: forging a signal costs far less than creating the fact behind it. So the thing that separates the two is never the technique, it's whether that fact exists.

• Hard evidence (classic experiment; scarcity changes valuation directly) · the cookies never changed, the jar did: Worchel, Lee & Adewole (1975, Journal of Personality and Social Psychology) had people taste identical cookies. Cookies from a jar holding 2 were rated better tasting and more valuable than the same cookies from a jar of 10; and when the supply was visibly cut from 10 to 2 in front of them (a manufactured shortage), ratings went higher still. Same cookie, different signal — the button does not distinguish real scarcity from fake.
• Hard evidence (neural level; wording changes the choice) · framing runs through the emotional circuit: De Martino, Kumaran, Seymour & Dolan (2006, Science) gave people mathematically equivalent choices framed as "keep £20" or "lose £30". Behaviour swung sharply with the wording; on imaging, going along with the frame was associated with stronger amygdala activity, while people with stronger anterior cingulate activity were better at resisting it. Wording is not packaging — it takes the emotional route. That's why selling outcomes rather than features works, and also why the same gun can be aimed at anyone. (Single fMRI study; reverse inference from imaging needs caution.)
• Boundary: the disclosure test can't adjudicate a benign frame. Saying "nine out of ten succeed" rather than "one in ten fails" survives disclosure, because nothing is being concealed. The test catches what depends on concealment, not all rhetoric.

Evidence strength: both are peer-reviewed classics with stable directions (effect sizes for the 1975 study vary across replications). The disclosure test itself isn't a research finding — it's an operating rule derived from the fact that manipulation depends on information asymmetry. Useful, not a theorem.

Mechanism: defaults run on three forces at once — status quo bias (changing costs effort and makes you responsible for the change), implicit recommendation (people read a default as "this is what most people pick"), and plain inertia. Friction is the same trick: you don't have to talk them out of cancelling, only make cancelling more tiring than doing it tomorrow.

• Hard evidence (cross-national policy data; enormous effect) · the default all but decides the outcome: Johnson & Goldstein (2003, Science) compared European organ donor consent. Where you must opt in: Denmark about 4.25%, Germany about 12%, the UK about 17.2%, the Netherlands about 27.5%. Where consent is the default and you must opt out: Sweden about 85.9%, France about 99.9%, Austria about 99.98%. (The cross-country comparison is observational, not randomised — but the gap is far too large for culture to explain, and the authors reproduced the same direction in controlled experiments.)
• Industry evidence (large automated census) · this is the industry default, not a few bad actors: Mathur et al. (2019, Proceedings of the ACM on Human-Computer Interaction) crawled roughly 11,000 shopping sites and identified over 1,800 dark pattern instances on about 11% of them, with a taxonomy (false urgency, misdirection, hidden costs, and so on). The use of that number: it confirms the pressure is real — if you don't do it, competitors do — which is exactly why the long-run ledger below has to be worked out properly.
• Boundary: not every default is a dark pattern. The difference is who the default serves: automatic backups on by default serves the user; auto-renewal ticked by default serves you. The test is the previous section's: "we ticked it for you, because most people won't untick it" — can you say that out loud?

To be straight: the FTC settlement and the DSA article are public enforcement and legislative facts, not research findings. They show the risk has been priced; they don't show dark patterns are net-negative over time. In the short run they do make money — which is why they are everywhere. The real ledger is the next section.

Mechanism: in almost every real market, selling is a repeated game — the same person buys again, and gets asked about you. The optimal move in a one-shot game (take what you can) is negative-return in a repeated one. Layer on one more thing: bad news carries systematically more weight in memory and in transmission than good news of equal size, so the two sides are not one-for-one.

• Hard evidence (large review, consistent across domains) · bad is stronger than good: Baumeister, Bratslavsky, Finkenauer & Vohs (2001, Review of General Psychology) pulled together evidence from impression formation, relationships, learning and emotion, concluding that at equal intensity, bad events have systematically larger effects than good ones. For you: moving a customer from "fine" to "delighted" takes many deliveries; moving them from "fine" to "I was had" takes one.
• Hard evidence (imaging plus behaviour; robust) · people pay their own money to punish unfairness: Sanfey, Rilling, Aronson, Nystrom & Cohen (2003, Science) used the ultimatum game. Facing clearly unfair splits, participants showed markedly stronger anterior insula activity (associated with disgust and negative affect), and the stronger that activity, the more likely they were to reject — and rejecting means they get nothing either. So "it's not like they have another option" is wrong at the neural level: they have one, which is to hurt you at their own expense.
• Boundary (this one has to be said) · don't treat "honesty pays" as a universal law: this arithmetic assumes buyers can compare notes. At the restaurant outside a tourist attraction, or anywhere buyers never meet, cutting corners genuinely does pay — which is why the food there is reliably bad. In markets with repeat business, a community and public reviews, holding the line pays. In a truly one-shot market it doesn't, and holding the line rests on your own standards rather than on returns. Rather than relying on willpower inside a bad market, don't pick that market.

Evidence strength: the first two are peer-reviewed, heavily cited, and directionally reliable. "Repeated game" is an economic model rather than an experimental result — before applying it to your business, check the premise: is there genuinely repeat business and a word-of-mouth channel? And this section gives no figure for "how much revenue unethical selling costs you", because I've seen no credible general estimate and won't invent one.

Mechanism: two forces. Goals under pressure compress attention onto the metric, squeezing everything else — including the other side's interests — out of view. And the slippery slope: each step is small relative to the last, so no step feels like crossing, because your reference point is always the previous step rather than the origin. Writing the list in advance nails the reference point back to the origin.

• Hard evidence (experiment, directly on goal pressure and misreporting) · people just short of the goal overstate the most: Schweitzer, Ordóñez & Douma (2004, Academy of Management Journal) had participants perform a task and self-report performance. Those given a specific goal misreported more than those told to "do your best" — and those who fell just short of the goal misreported most of all. Quarter end is that experiment in the field. This isn't a claim about your team's character; it's a claim that the metric itself manufactures the pressure.
• Hard evidence (experiment series, slippery slope) · the cost of the first time isn't paid the first time: Welsh, Ordóñez, Snyder & Christian (2015, Journal of Applied Psychology) showed that when the opportunity to be dishonest grows gradually, people end up cheating substantially more than when the same large opportunity is offered outright. The real cost of "just this once" is that it lowers the bar for next time.
• One piece of evidence I wanted to give you and can't: a famous study reported that moving an honesty declaration to the top of a form reduces cheating (Shu et al., 2012, PNAS). It later failed to replicate at scale (Kristal et al., 2020, PNAS), one of the original field datasets was found to be fabricated, and the paper was retracted. So don't add an integrity pledge to the top of a page and assume the problem is handled. That's also this issue's subject applied to itself: mislabel the strength of your evidence and the conclusion becomes harmful.

Boundary: a list stops deliberate crossings; it doesn't stop biases you can't see in yourself — such as sincerely believing the product suits everyone. So pair it with an external signal: refund rates, early churn, and "who signed the customers who left inside three months". Data won't cover for you. That's the second gate.