Start where the Intermediate pages left you: you already know minds anchor, and weigh losses double. The systems insight is that these mechanisms are always firing, because every interface that presents options is making design decisions. Something is the default, something comes first, something takes three taps while something else takes one. So 'we don't nudge our customers' isn't an available position. An un-designed checkout still has defaults, still has an order, still has friction — it's just architecture by accident, nudging wherever entropy points. Thaler and Sunstein's name for the discipline is choice architecture: accepting that the presentation layer is a policy instrument, and taking responsibility for what it does.
The instruments, in rough order of power. Defaults do the most work for the least effort, because they harness three forces at once: inertia, implied endorsement ('the default must be sensible'), and loss framing (unticking feels like giving something up). Pension auto-enrolment schemes abroad moved participation by tens of percentage points without changing a single incentive. Friction is the quiet twin: each extra field, screen, or decision sheds users. That cuts both ways — it subsidises the behaviour you streamline (one-click, saved cards, autofill) and taxes the one you obstruct. Salience decides which attributes even exist at the decision moment (the anchor you display, the loss you preview — the Intermediate levers, deployed as components). Timing exploits moments of openness: the insurance offer at booking rather than after, the savings prompt at salary credit. Social norms recruit conformity, with the crucial fine print that descriptive norms advertise whatever is common — including the bad thing.
Now the humility layer, which is what makes this Advanced rather than a listicle. Nudge effects replicate unevenly: they're usually modest, context-dependent, and capable of inverting. The energy-use reports that cut consumption among heavy users nudged some light users up toward the norm (the boomerang), and defaults can fail loudly where they collide with strong preferences or distrust. The EAST checklist (Easy, Attractive, Social, Timely) is a fine design mnemonic, but the operating rule is experimental: every architecture change is a hypothesis about behaviour, tested against a control with a pre-named metric, exactly as the proposition-testing ladder taught. Behavioural science gives you better hypotheses than intuition does. It doesn't excuse you from testing them.
Finally the line, drawn precisely. A nudge, in the founders' own definition, steers while preserving freedom: transparent, cheap to opt out of, and aimed at helping choosers act on their own intentions (the auto-enrolled employee who always meant to save). Sludge is the same machinery reversed — friction and framing turned against the user's interest: subscription mazes designed to exhaust you into not cancelling, pre-ticked add-ons riding on inattention, countdown timers counting down to nothing, confirm-shaming ('No thanks, I hate saving money'). The distinction stopped being philosophical in India in 2023, when consumer-protection guidelines specified thirteen dark patterns — false urgency, basket sneaking, drip pricing, and their cousins — as unfair trade practices. The practitioner's test is Thaler's own publish test, sharpened by the regulator: if the architecture were shown to the customer with full annotation — here's the default we chose, here's why — would it read as help or as a trap? Sludge converts trust into one-time revenue. That isn't just wrong, it's a terrible trade.