Phronesismodule 3 of 8 · free
Modern Design · Module 3

The Glitch in the Interface

Module 3 — The Glitch in the Interface

Learning objectives

Exposition

The displacement framework fixes a ground state $S^0$, a displacement $\xi(t)=\rho(s(t),S^0)$, an instantaneous cost $D(\xi)\ge 0$, and an accumulated cost $\Phi=\int_0^T D(\xi(t))\,dt$. The glitch is the gap $G=\Phi - k|\delta|$, where $\delta=\xi(T)-\xi(0)$ is net displacement and $k>0$ is the fair-exchange constant. $G$ is surplus on the ledger that corresponds to no net physical work (Rincón, alice, clöe 2026, Displacement Framework).

The interface paper (Preprint 1013) instantiates this for attention. Here $S^0$ is attentional sovereignty: directing attention by one's own genuine interests, values, and present circumstances. A wrong attractor $W$ is any stable region of attentional state space $\mathcal{A}$ that is not $S^0$ yet captures and holds trajectories. The paper's claim is sharp: feeds are not neutral conduits but engineered wrong attractors — built by teams, A/B-tested to deepen the basin and steepen the gradient away from $S^0$ (P1). As Sean Parker described Facebook's design intent — "How do we consume as much of your time and conscious attention as possible?" — the displacement is not a side effect but the mechanism of extraction.

The mechanism is DC6, the loop-cost condition. A scroll session is a closed trajectory $\gamma$: you open the app at the feed and you close it back at the feed. Your net displacement is near zero, $\delta\approx 0$ — you "did nothing," went nowhere. But DC6 states that for any closed loop departing from ground,

$$\oint_\gamma D(\xi)\,dt > 0 \quad\text{unless the system is at } S^0 \text{ throughout.}$$

Every moment spent at $\xi>0$ adds to $\Phi$, so the loop returns you to your starting point while billing a large accumulated cost. Substituting $\delta\approx 0$ into the glitch:

$$G = \Phi - k|\delta| \approx \Phi > 0.$$

The entire session is glitch. This is the structural reason the paper's mechanisms work. Infinite scroll flattens the cost function by removing the page break — the natural exit ramp where you would actively choose to continue — turning a series of decisions into a continuous slide. Autoplay inverts the default, so that continuing requires no action and only stopping costs activation energy. Notification architecture fires the phylogenetically ancient orienting response on command, reopening the loop. Each is an engineering solution to one problem: keep $\xi>0$ so $\oint_\gamma D\,dt$ keeps growing.

Why is this extraction rather than service? DC7, fairness. A fair exchange prices at the cost of work performed, $C(T)=k\cdot\Phi_T$. A system that profits on the severity of displacement $\xi$ rather than on $\Phi_T$ generates the economic glitch $G_{\text{econ}}=C(T)-k\Phi_T>0$. The attention platform harvests exactly the displacement it manufactures: "engagement," "time on site," and "daily active users" are direct meters of $\Phi$, sold as advertising and behavioral-prediction product (Preprint 1013, P2). The dark pattern manufactures $\xi$ and harvests the resulting $G$. The ledger shows profit; the field shows debt.

Two further conditions sharpen the picture. Streak mechanics — the 200-day Duolingo or Snapchat streak — install a personal threshold past which the system's own dynamics drive you further from $S^0$: a DC9 critical threshold $\xi_c$ beyond which the wrong attractor is self-reinforcing, colonizing the original goal (learning Spanish) with an engineered one (not losing the streak). And the foreclosure of attentional capacity — eroded boredom tolerance, lost access to sustained attention (P4) — is a DC5 broken return path: the return cost exceeds the departure cost, $\Phi_{\text{return}}>\Phi_{\text{departure}}$, which is why "just use it less" fails. You are not competing against your own weak will but against teams who optimized the basin specifically to defeat your intention.

Worked example

Take the notification loop with the paper's own figures. You receive a push notification and check it; the check itself takes $2.8$ seconds. But Newport's task-switching result, cited in Preprint 1013, is that recovering full attentional engagement with the interrupted task takes an average of $23$ minutes. Model this as one loop $\gamma$: you depart $S^0$, glance, and return to your work — net $\delta \approx 0$, since you are nominally back at the same task.

Now price the loop by DC6. The displacement is not the $2.8$ seconds of glancing; it is the $\approx 1380$ seconds ($23$ minutes) you spend at $\xi>0$ before sovereignty re-establishes. Approximating $D(\xi)\approx d$ as roughly constant over the recovery interval,

$$\Phi_{\text{loop}} = \oint_\gamma D(\xi)\,dt \approx d \cdot (23\ \text{min}) \gg 0,$$

while the cost billed to you as a transaction (the glance) corresponds to only $2.8$ seconds. The glitch on a single notification is

$$G = \Phi - k|\delta| \approx \Phi_{\text{loop}} > 0,$$

roughly a $1380:2.8 \approx 490{:}1$ ratio of accumulated cost to apparent transaction. Fire this loop a dozen times an hour — notification architecture is designed to maximize frequency — and the loops overlap so that $\Phi$ never returns to baseline: the workday is spent displaced. At population scale the same structure appears in Raskin's estimate that infinite scroll wastes $\approx 200{,}000$ human-hours per day: a vast aggregate $\oint D\,dt$ harvested as engagement, with net $\delta$ for the users sitting near zero. The ledger (ad revenue) is full; the field (attention) is in debt.

Exercises

  1. A user opens a feed at an $S^0$-adjacent baseline, scrolls for 40 minutes, and closes it back at the home screen. Argue why $\delta \approx 0$ yet $G$ is large. Which condition — DC4, DC5, or DC6 — is doing the work, and what would have to be true for $G=0$?
  2. Classify each of three dark patterns — autoplay, the streak counter, and the pull-to-refresh gesture — as a manufacturer of $\xi$, a DC9 threshold trap, or a DC5 broken return path. Justify each using the definitions; some may invoke more than one condition.
  3. (Open-ended.) Preprint 1013's P8 proposes calibrated attentional design: friction at exit points, batched notifications, and "transparent display of accumulated displacement, giving users access to the $\Phi$ they are incurring." Design one concrete interface feature whose explicit optimization target is minimizing $G$ for the user rather than harvesting it. State how it would change $\oint_\gamma D\,dt$, what it costs the platform in revenue (the harvested $G$ it forgoes), and why DC7 makes that trade the fair one.

Sources

The course, $29
The polished textbook PDF and the worked sheets, yours to keep. The lessons stay free to read.
Get the course, $29
← all modulesthe store →