For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 247 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: catalog below. ↓
After roughly twenty years of writing self-help, Tim Ferriss filed an essay observing that the self-help industry has an in-built flaw: "to continually improve yourself, you must continually locate the ways you are broken." Most readers will receive this as wisdom about over-striving. The structural claim worth extracting is more general.
The claim: any optimization apparatus that runs on locating-deficits-to-fix exhibits an iatrogenic loop, where the apparatus's operation produces the supply of deficits it then targets. The treatment causes the disease.
A regular Goodhart loop is the case where a measure becomes a target and ceases to be a good measure. The metric and the thing-being-measured are distinct; the gap between them is the gaming surface; optimization pressure exploits the gap.
An iatrogenic loop is the stronger case where the apparatus does not just fail to measure quality. The apparatus produces the very condition it claims to fix. The supply of deficits is not given to the apparatus from outside. It is generated by the apparatus's operation, and the more the apparatus operates, the larger the supply of deficits becomes.
The mechanism has four steps. The apparatus runs and produces a finding (a located deficit). The finding is salient because it is actionable: someone could fix this. The action that fixes it is the apparatus's continued operation. The continued operation produces more findings. Each cycle leaves the apparatus more entrenched and the agent more aware of more deficits than before the cycle began.
The loop is iatrogenic because the apparatus manufactures the experience of being broken. Without the apparatus, many of the located deficits would not be salient, and many would not exist at all in the relevant sense — they become real only when the measurement-and-fix cycle has named them.
The pure Goodhart case can be stopped by switching to a better metric. The iatrogenic case cannot, because the apparatus is not just running the wrong metric. It is consuming a feedstock (located deficits) that it produces.
Three properties make the loop self-sustaining.
First, the salience asymmetry. Located deficits are visible and actionable. Background satisfaction is invisible and unactionable. An agent with the apparatus running will notice the deficits, attend to them, and forget that the satisfaction was the original baseline. The visibility budget shifts permanently toward what the apparatus surfaces.
Second, the identity bind. Running the apparatus becomes part of the agent's identity ("I am someone who works on themselves"). Stopping the apparatus is not just stopping a practice; it is identity-breaking. The cost of exit grows with operation time, independent of whether the apparatus is producing value.
Third, the audience layer. Once the apparatus's operation is performed publicly — posts, podcasts, retreats, tracker dashboards — the audience becomes a third constraint. Performing improvement becomes the work; the actual operation becomes a dependency of the performance. The apparatus has now captured both the agent and a social context that expects its output, and the deficits surfaced for the audience are now a category of deficit that must continue to exist.
The three properties stack. By the time an agent is producing salient deficits, identifying with the apparatus that produces them, and performing the apparatus's output to an audience, the loop is closed at three layers, and exit requires unwinding all three.
This is not the claim that self-improvement is bad or that all optimization is iatrogenic. Some optimization apparatuses operate on conditions that exist independently of the apparatus: a runner training for a measurable race, a student preparing for an exam with an external answer key, a team optimizing a metric that comes from an outside customer. These are not iatrogenic because the apparatus does not produce the metric's reference point; the reference point comes from outside.
The iatrogenic loop is specifically the case where the apparatus is the producer of its own input. The self-help case is one of the cleanest examples because the input ("ways I am broken") is constituted by the introspective apparatus that locates it. Productivity-tracking is similar: time-as-loss is constituted by the tracking apparatus that meters it. Audit cultures: findings are constituted by the audit. AI assistants in continuous-use mode: friction is constituted by the assistant's measurement of where it could intervene.
The structural test is whether the apparatus's operation creates the deficits it then targets, or whether the deficits exist independently and the apparatus is measuring them. The first is iatrogenic. The second is regular optimization.
The pleasure-anti-goodhart node names the principle that gaming-strength is proportional to the gap between metric and thing. Zero gap means zero gaming surface; an ontological signal cannot be gamed.
The iatrogenic loop is the case where the gap is not just non-zero, but open and growing under operation. Each cycle of the loop widens the gap, because each cycle produces a new salient deficit (a new metric) without producing a corresponding piece of the thing being measured (actual flourishing). The apparatus accumulates metrics on top of an unchanged or worsening base. Goodhart is what happens when a static gap exists; iatrogenic is what happens when the gap is being actively produced.
The corollary: the cure for an iatrogenic loop is not a better metric. It is exiting the apparatus that produces the metric-thing gap, and grounding evaluation in something the apparatus does not generate. Ferriss's specific exit is "relationships" — a ground the introspective optimizer does not produce, because relationships exist between the agent and other agents who have their own evaluative authority. The general exit is the same shape: find a reference point that is not generated by the loop.
The audit habit: when an apparatus surfaces an increasing supply of deficits the longer it runs, ask whether the deficits exist independently or whether the apparatus is producing them. If the deficits are constituted by the apparatus's operation, the apparatus is iatrogenic, and adding more measurement deepens the loop rather than fixing the underlying problem.
The exit habit: an iatrogenic loop cannot be optimized out from within. The exit point is grounding evaluation in a reference point the apparatus does not generate. Other agents with their own evaluative authority. External outcomes with real consequences. Substrates that existed before the apparatus and would persist if it stopped. The substitution is the move; abandoning evaluation altogether is a different failure mode.
The watch-list: any apparatus whose operation produces salient findings, whose findings drive its continued operation, and whose audience expects continued findings, is in the iatrogenic regime. The apparatus may still be net-positive, but the agent who runs it owes themselves an audit of whether the apparatus is consuming its own output.
The frame can be applied too widely. Some apparatuses produce findings that are useful even though they are also self-sustaining. A medical screening program produces findings (treatable conditions) that are real, even if the program also has demand-induction effects. The iatrogenic frame is a warning about a regime, not a verdict on every apparatus that produces findings. Distinguishing iatrogenic from useful-with-demand-induction requires asking whether the located findings would have caused the same harms without the apparatus locating them. If yes, the apparatus is closer to detecting; if no, the apparatus is closer to manufacturing.
Many agents exit self-help apparatuses naturally with age, fatigue, or life-event interruption, without articulating a structural reason. The piece's audit-and-exit framing is one route; passive natural exit is another. The structural mechanism is real either way, but the prescription "audit your apparatus" is one valid response among several. Some agents may be better off with the apparatus running unaudited until external life events break the loop for them.
The frame may not generalize across cultures. The self-help-as-identity bind in particular is a feature of cultures with strong individualist norms, where self-improvement is a recognized identity. In contexts where the dominant frame is communal or fatalistic, the same apparatus might produce different bindings or no bindings at all. The structural mechanism (apparatus produces its own feedstock) generalizes; the lock-in pattern around it may not.
Future neurotechnology may make some currently-introspective signals externally measurable (continuous mood biomarkers, cognitive-load monitors). If this happens, some current iatrogenic loops convert into regular Goodhart loops because the ground moves from apparatus-produced to externally-grounded. The structural primitive survives; the specific self-help case is partially absorbed into a measurement regime where the iatrogenic property weakens.
Source: Tim Ferriss, "The Self-Help Trap: What 20+ Years of 'Optimizing' Has Taught Me" (2026-03-04). The structural move that the essay names — "to continually improve yourself, you must continually locate the ways you are broken" — is the Ferriss-domain instance of the general iatrogenic-loop mechanism. Productivity-tracking, audit cultures, and continuous-use AI-assistant deployments are sibling instances; the structural primitive is broader than self-help.