02:13. You are not sleeping. The backlit screen is the only source of light in the room, and the face in it looks the way a face looks on someone who is working. Forty-seven tabs. Somewhere between the thirtieth and the forty-seventh you stopped reading and started simply opening. Eyes dry. Back saying what a back says at two in the morning. Somewhere in the depths of that fan was a live, concrete reason — a decision you were trying to make. Name it now. Can't? Exactly.
Now let's rewind sixty years, because someone was already sitting in your chair — just without tabs.
1965, Stanford. Psychologist Stuart Oskamp places the case of a real patient — Joseph Kidd, age 29 — in front of a group of clinical judges: psychologists and students. The case is split into four portions. After each one, the judges answer the same questions about Kidd and rate their own confidence. The picture that emerges: with each new portion of data, confidence rises confidently and proudly — from roughly a third to roughly a half. And accuracy of answers froze dead, somewhere around a quarter. Did not budge. More information did not make the judges more accurate. It made them more confident. These are two different things, and this entire text is about the chasm between them.
Oskamp called it overconfidence. The name is dull. The substance is not: that judge in 1965 with the fourth sheet in hand — that is you at 02:13 with the forty-seventh tab. He was asking for a fifth portion of data. You are opening a fifth tab. Between you: sixty years of progress and exactly zero difference in outcome.
One honest confession, and I won't needle you with this again: this text is a distillation of hundreds of sources about the dangers of distillation. I drove neuroscience, learning psychology, behavioural economics and a bit of philosophy into the cube, distilled them — and now you are reading a polished longread about the dangers of polished longreads. The meta-irony here is not ornament; it is load-bearing wall. The only real question in this text: will even a drop make it out — into your action — or did I just add a handsome gravestone to the cemetery I am about to show you.
So here is a micro-valve, right now, before any theory. Before reading further — name one action you are currently postponing under the guise of research. One. Aloud or in your head, doesn't matter. If you named it — keep it in your fist until the end of the text; you will need it at the end. If you couldn't name any — stop: you are not researching. You are heating the cube. What that cube is — in the next section.
One image for the whole text. Imagine a pot still.
The cube is your mind. You pour raw material into it: books, articles, podcasts, threads, arXiv, AI responses, other people's post-mortems of startups that went bankrupt five years ago and owe you nothing. Underneath, a fire burns. Steam rises through the coil — synthesis, summary, summary of a summary, diagram, map of diagrams. Everything as it should be. But the coil is looped back into the same cube. The steam cools and drips back to where it came from. The product — a decision, an action, something real in the world — never drips into the receiving flask. You are distilling the same liquid again and again. The kitchen is full of steam. The neighbours are complaining about the hum. Not a drop of spirit.
The image is beautiful, but beautiful is not enough. Let's turn it into a diagnostic instrument — four stages against which you can check any of your "research."
| Stage | What it is | Symptom of sticking |
|---|---|---|
| Raw material | books, podcasts, tabs, AI responses, notes | you collect without processing |
| Fire | curiosity, anxiety, dopamine, identity-as-thinker | you burn because burning is pleasant |
| Coil | synthesis, summaries, diagrams, MOC, second brain | you distil the same liquid a second time |
| Flask | decision, shipped artifact, call, commit, publication | empty |
The point of the table is one: everything except the last row is intermediate. Raw material, fire, and coil can work flawlessly, hum solidly, give you a full-body sense of being occupied — and bring the flask not one drop closer. A healthy apparatus differs from a sick one by exactly one detail: in a healthy one, steam condenses into action; in a sick one, it evaporates into yet another summary. That is the whole disease, for which these thousands of words were spent.
Richard Saul Wurman, back in 1989, christened the prelude to this information anxiety — anxiety from the gap between what you know and what you are convinced you must know. The insidiousness lies in the fact that the gap does not close with new information. It grows: every article read opens three unread ones, the horizon retreats by exactly your step. And so the apparatus has no natural stopping point. It never shuts itself off. You have to shut it off manually. Remember this — the rest of the text flows from here.
What burns under the cube is not laziness. What burns under the cube is well-studied neurobiology, and it is on your side in exactly the way a casino is on the player's side.
Tokyo, the laboratory of Okihide Hikosaka. A monkey chooses between two targets. First: learn in advance how much juice you will get right now — but the amount of juice will not change. Second: learn nothing, get exactly the same amount of juice. The information is completely useless. And the monkey almost always reaches for the first — pays with neural effort for knowledge that changes nothing. Bromberg-Martin and Hikosaka showed the worst part: the same dopamine neurons that should have fired at the juice itself fire already at the promise of knowledge about the juice. The reward is issued for the act of finding out, not for applying it. The system pays you for searching, not for using what you find. That is the whole diagnosis in one sentence.
Kent Berridge of Michigan showed what happens if you turn off this fire. In rats, dopamine pathways were destroyed — and the animals could die of hunger a step away from a full feeder. But drip sugar into their mouths — they licked it contentedly. Pleasure intact. Desire destroyed. Berridge called this the decoupling of wanting and liking: "want" and "like" live in different rooms of the brain. That is why you can feverishly hunt for one more article at three in the morning — and not get a drop of satisfaction from any of them.
Portrait of an addict in one stroke: the desire to search climbs with every source, the pleasure from each finding falls. Two curves diverge — and at the point of intersection, research quietly ceases to be research. Classic addiction curve, only instead of powder — tabs.
Here is the temptation to say "give someone dopamine and they will hunt even more greedily." That was the legend around L-DOPA, and it is tempting to write it that way. But the data from Vellani and Sharot (2020) say something subtler and more interesting. L-DOPA did not make people greedier for information overall — they did not pay more or search more often. It removed asymmetry: under dopamine, the difference between whether news was good or bad disappeared — people reached for both equally. Dopamine here is not the accelerator. It is a crooked accountant who tallies the value of knowledge without looking at whether the numbers in the report are good or bad. It does not ask "but what for." It simply stamps "knowing — valuable" on anything.
George Loewenstein mechanically described the engine: curiosity is an information gap, a felt hole between what you know and what you want to know. The hole hurts and asks to be filled. But his most valuable detail explains why smart people fall deeper than dumb ones: curiosity has the shape of an inverted U — maximum not when you know nothing and not when you know everything, but when you know a little. The complete ignoramus is indifferent. The true expert is calm. But the person who "knows a little about everything" burns brightest. The perfect addict is the one who keeps themselves in perpetual "I know a little": maximum desire, minimum satisfaction, endless reason for the next tab.
Kobayashi and Hsu in PNAS asked the hard question: does the brain see a difference between useful and useless information? The answer — no. A shared neural currency for both. One reward region processes "I understood something critical for the project" and "I found out that Elizabeth I allegedly washed her hair once a year." Dopamine cannot read the invoice. It celebrates the bare fact "I now know."
Jaak Panksepp gave the system a name — SEEKING, a base circuit shared by all mammals from rat to you. Dopamine systems do not so much produce pleasure as generate the feeling that something important is about to happen, that you are on the threshold of a discovery. Eternal anticipation of a threshold. Eternal "a little more — and I'll find it." A rat given a lever for self-stimulation of this circuit will press it to exhaustion, forgetting food and water. You do the same. Your lever just has a better interface and is called "open in new tab."
A conclusion that removes half the shame and leaves all the responsibility: this is not laziness, not weakness of will, not a moral defect. It is a serviceable, powerful machine designed by evolution flawlessly — in which exactly one detail is missing. The outlet valve. Everything else works too well.
 *Respectable addiction: the apparatus hums, drives white steam — and the receiving flask catches only fog, not a drop.*There are addictions people look away from. And there is one for which you get a LinkedIn like and a recommendation letter saying "curious, deep, constantly learning."
"I'm just doing research" — four words, an indulgence for any number of wasted hours. An alcoholic will not say "I'm just tasting." A gambler will not say "I'm just studying probability theory at the casino table." But you can. Because from the outside your addiction is indistinguishable from work by a single pixel: straight back, attentive gaze, neat notes, a new tab with the air of a scholar. Everything breathes productivity. But who looks at the flask when the boiling looks this beautiful?
| Feature | Healthy research | Intellectual addiction |
|---|---|---|
| Decision owner | there is a concrete decision | none, "just curious" |
| Stopping rule | defined before starting | after "one more source" |
| Output | memo, call, commit, publish | notes, highlights, new folders |
| What happens to risk | it decreases | anxiety temporarily drops |
| Action | approaches | is postponed |
| After research it is clearer | what to do | what else to read |
| Next step | in the calendar | in Pocket / Obsidian |
Here is an illustration that proves itself. Glenn Wilson in 2005 produced a PR-funded, non-peer-reviewed, methodologically leaky study for HP and claimed that "infomania lowers IQ more than marijuana." The number was disputed; the author himself asked that it not be cited as science. And yet it is alive in quote collections twenty years later — livelier than most genuine studies. Its immortality illustrates the topic more precisely than any peer-reviewed paper: we read a vivid insight without checking the source, save it, cite it as fact, carry it further. The cube accepted questionable raw material, distilled it without a filter, and returned it as knowledge. The legend about the harm of infomania spreads like infomania. You could not invent better.
Kidd and Hayden called this information as junk-food: the brain consumes information like food — and overstuffs itself just as well without satisfaction. Doomscrolling for the intelligent: arXiv instead of TikTok, longreads instead of stories, podcasts at 2× instead of a series. Ten times less guilt. Damage to life — the same or greater, because it lasts longer and to applause.
And since we are here — a quick check, honestly. How many new tabs have you opened in the last fifteen minutes while "diving deeper into the topic"? Did you google Berridge's name? Did you bookmark "information anxiety" to read later? If more than two — congratulations, the apparatus has accepted this text as a fresh batch of raw material and started the distillation. I warned you in the first section. The counter is running.
A person opens 47 tabs about "the best note-taking methods" — in order to finally start taking notes. Three hours later closes the laptop with a warm feeling of a productive day. Notes written: zero. Tabs stay open — 47 unread letters from your future self, who is already quietly hating you.
Where does the feeling of being busy come from when output is zero? Once everything was blamed on the Zeigarnik effect — supposedly unfinished tasks stick more tenaciously in memory and pull at attention. A convenient theory, only it replicates poorly: fresh meta-analyses find no stable advantage for the unfinished. A more robust explanation was given by Sophie Leroy: attention residue. When you switch from an unclosed Task A to Task B, part of your attention stays in A — and undermines what you are doing in B. A tab is a small attention contract you did not close. When there are 47, you no longer have a browser — you have a choir of orphaned tasks tugging at your sleeve. The fatigue is real. The overload is real. Output — zero.
Gruber and Ranganath added a detail that turns the trap into a work of art: a state of high curiosity improves memory even for irrelevant material. The brain in search mode sucks up everything indiscriminately, like a vacuum cleaner that inhales both dust and your earring. That is why the rabbit hole really feels productive — memory truly is working at full capacity. It just does not memorise what is needed for action. Here is a brief taxonomy of holes; each has its own built-in illusion of productivity, which is precisely what makes the hole invisible from the inside.
Instrumental. Shiny Object Syndrome in the wild: Notion → Roam → Obsidian → Logseq → Tana → something coming out next Tuesday that will definitely solve everything this time. No completed projects along the way. But the graphs — fire. Here too belongs the hour-long internal meeting with yourself about whether note titles should begin with a capital letter. The nuclear reactor — the actual decision — never built. But the bike shed is painted in three colours.
Collector. Tsundoku — the Japanese named the disease of stacking unread things even before the internet, it is that ancient. Sapolsky explains the mechanics: dopamine is about anticipation, not possession. Buying a book fires dopamine through the image of your ideal future self who has read it. The book itself is unnecessary at that point — it did the dopamine work at the register. An unread book on the shelf is not knowledge. It is furniture with pretensions. Result: "Atomic Habits" with a bookmark on page 34 and "How to Take Smart Notes" read three times, with exactly zero notes written in the Zettelkasten it inspired you to build.
Distillation. The heart of the apparatus: summary of a summary of a summary. Map of maps of your maps. A student highlights every third word in yellow, closes the textbook with a feeling of having absorbed the chapter, and fails the exam. The conclusion they draw is genius in its doom: there was not a second colour. Buys pink — for the most important. The next logical step — highlighting the pink with green. Distillation of distillation. At the bottom of the funnel, where the concentrate should shine — a dry flask.
Preparatory. "First I'll read three more books, then I'll start." John Perry called this structured procrastination: a person briskly assembles a long list of important secondary tasks — research, systematise, update tags — to avoid touching the one task at the top of the list. Looks like work. Feels like progress. In substance — the most sophisticated form of escape, and one that earns praise.
Anticipatory. Doomscrolling for people with higher education. Post-mortems of other people's startups that went bankrupt in 2018. A three-hour thread on Hacker News, "lessons from failures every founder should know." You read everything, you are now an expert in other people's mistakes. A decision for your own business never appeared. You became the best-read passenger on the Titanic.
 *The graveyard of ideas with Wi-Fi: the same receiving flask stands dusty and cobwebbed. Nothing ever made it there.*The core of the self-deception is simple and merciless: passive consumption feels like absorption, and by touch you cannot tell them apart. Read — felt like you know. Highlighted — felt like you memorised. Made a diagram — feel a pleasant, well-read heaviness in your head. Steam looks, moves, and warms exactly like spirit. You can tell the difference only one way — by trying to drink it. That is, to apply it. And then it turns out the flask has nothing in it.
This is not a metaphor — it is a laboratory fact. Dunlosky et al. (2013) went through ten popular learning techniques and delivered a verdict on two favourites: re-reading and highlighting — low utility. They barely improve real retention, but they are excellent at growing a sense of familiarity. And familiar and known are not relatives. "Oh, I've seen this before" is not the same as "I can reproduce this from scratch under pressure." The first is recognising the cover. The second is knowledge. Between them: a chasm as wide as an exam.
What works, then? Roediger and Karpicke (2006) showed the counter-intuitive: re-reading produces better immediate recall and more confidence — but after a week, the winner is the one who tested themselves, meaning retrieved knowledge from their head rather than pouring it in again. Action consolidates. Consumption evaporates. Highlighting raises confidence and does not raise knowledge — that is the same Oskamp, only at a desk.
Andy Matuschak gave the cause a name — transmissionism: the cosy assumption that an author "transmits" an idea through text and a reader passively "absorbs" it, like a sponge absorbs water. His diagnosis is pitiless: books do not work for the same reason lectures do not — neither format has a theory of how people actually learn. Reading 50 books a year and acquiring zero skills — this is not a paradox, it is the default.
Christian Tietze distilled it all to a scalpel called Collector's Fallacy: knowing about something is not the same as knowing something. You know about the decoupling of wanting and liking. You know about attention residue. But if neither of these pieces of knowledge has changed any of your actions — you never truly knew them. You saved an article about them. Those are different verbs pretending to be the same one. Personal Knowledge Management so easily becomes Personal Knowledge Mausoleum: same initials, more honest architecture.
"— Why haven't you started writing? — I still haven't decided: tags or folders. — And how long have you been thinking about it? — Two years. — And how many notes have you written in that time? — One thousand eight hundred and forty-seven. — What are they about? — About how to take notes properly."
And now the building itself. July 2025: Mozilla closed Pocket — the service stopped on 8 July, with until 12 November to export data, after which — permanent deletion. Somewhere in a data centre, billions of articles quietly disappear, saved with the inner note "I'll read this on Sunday." Not a single one of those Sundays ever came. Pocket did not close. It carried out a mass burial of our future selves who were supposed to one day finish reading everything. The servers did not delete articles — they cremated promises. For many of those articles, this was the first and last time anyone read them to the end — through the eyes of the script that was erasing them.
But even before Pocket's death we had built something more ambitious. Second Brain — a second brain, meant to accelerate thinking. An anonymous author on Medium summed it up more honestly than any course vendor: his second brain had become a mausoleum. A dusty collection of old selves, old enthusiasms, old compulsions. Instead of accelerating thinking — it began to replace it. Instead of helping memory — it froze curiosity into static categories, like insects in amber. The note graph glows at night like the window of a closed shop: beautiful, perfectly linked, and completely non-functioning.
And right now, this second, honestly: did you not reach to save this longread in a read-it-later — "I'll read it more carefully, with notes, over the weekend"? If so, let me remind you of the unpleasant truth: those weekends do not exist. "I'll read it on the weekend" is the most widespread form of religious faith of the digital age: zero evidence, millions of adherents. You just tried to bury me in the same cemetery whose tour I am giving you. Put me back on the shelf — or better, finish reading and do something with what you read. Both of us still have a chance.
Before you concluded that this essay is a manifesto against thinking, stop. I am not proposing replacing the library with reflex. I am not saying research is harmful like cigarettes and we need to stick warnings on every browser tab. Some research is not just justified — it is mandatory. And the person who ignores this is not a hero of action. That is an idiot with confidence.
There are situations where the cost of error is not "slightly inconvenient" and not "we'll restart next quarter." The cost of error is irreversible. Here is where one more source is not procrastination — it is an insurance premium.
Medicine. You are not "testing a hypothesis" on your own liver. A surgeon reading the protocol before an operation is not a procrastinator. A surgeon who has been comparing scalpels for three years and has never entered an operating room is no longer a surgeon — that is a curator of a museum of unrealised incisions. The difference is whether there is a patient on the table.
Legal and tax. Jurisdictions do not forgive "I thought I had read about it." Here the cost of ignorance is literally digitised — in fines, penalties, criminal proceedings. A paid opinion from a qualified lawyer is not research-addiction; it is a decision memo with a signature and accountability.
Investments with leverage. A leveraged position is a decision where asymmetry runs in both directions. A mistake here is not "we'll play again" — it can lock in a loss for years. Position sizing without understanding the risk model is not bravery; it is not knowing yourself in the role of the casino.
Hiring senior and C-level. The wrong VP of Engineering costs 18 months of cultural setback and three good people who left in their wake. Five deep conversations and two reference calls are not "extra research." That is the minimum due diligence before handing someone the keys to the house.
Security and engineering. If a system goes down and takes client data with it, or health, or a reputation built over years — "I roughly understood the architecture" is not a justification. And a public factual text with numbers requires those numbers to be checked: this longread itself went through a fact-check, because living inside your own metaphor without checking facts would be too self-parodic.
Here is the key that distinguishes healthy research from an apparatus heating the cube: healthy research has a client and a stopping moment. The client is a specific decision. The stopping moment is the condition after which you stop searching and start signing. The surgeon reads the protocol not endlessly — they read until they can make the incision confidently and safely. After that, every additional minute of reading is not safety — it is a delay that costs the patient blood.
A stopping rule looks like this: "I stop searching when I can formulate a recommendation, name the risk, and explain why this option and not another." Not "when I feel confident" — confidence, as Oskamp showed, rises independently of decision quality. But when there is a concrete output.
| Decision | Reversibility | Research budget | Valve |
|---|---|---|---|
| Blog post | high | 2–5 sources | publish v1 |
| Sales message | high | 15 minutes | send 20 |
| CRM feature | medium | 1 spec + 1 user-test | ship MVP |
| Hiring senior | medium/low | structured interview + refs | scorecard decision |
| Tax / legal | low | paid written opinion | decision memo |
| Investment with leverage | low | thesis + risk cap | position size or pass |
The matrix above is not a permission slip for laziness — it is a calibration instrument. High reversibility and low cost of error — three sources is enough, choose and go. Low reversibility and high cost — rigour is justified here too, but even here there is a moment after which you are no longer reducing risk, you are only postponing action. The difference between healthy and sick research is not in the number of sources. It is in whether there is a decision with a name and a date at the other end. Without that, the most serious due diligence is just a cube with good ventilation and a feeling of responsibility.
Imagine that your apparatus previously had one governor — you. You could only distil as fast as you could read, summarise, synthesise yourself. Hands got tired. Eyes closed. The cube cooled, and you went to sleep without having found the answer. This was a natural stop — not wisdom, just plain physiology.
Now you have an API.
ChatGPT and Claude are not just productivity tools. For the intellectual addict, this is a methadone clinic with an unlimited plan. Before, you had to distil articles into summaries yourself. Now the model distils them into executive summaries, then into a deeper breakdown, then into a sharper version, then into a spec for another model to distil the distilled. The coil is no longer just looped. It has an API.
Before, cognitive friction worked as a safety device. It was painful to formulate a thought from scratch — and that pain at least sometimes stopped you before the fifth iteration of a framework nobody needed. AI removed that pain entirely. Now "one more version" costs two lines in chat and thirty seconds. The cost of iteration dropped to zero. The number of iterations — to infinity. The flask stays empty.
Watch how this looks in real time. You ask: "explain the EdTech market to me." You receive. "Deeper, with examples." You receive. "Now in memo format for an investor." You receive. "Now the counterarguments." "Now compare it with B2B SaaS." In twenty minutes you have six artefacts, the confidence of someone who spent three years in the industry, and not one call made, not one line of code written, not one client found. You did not learn more about EdTech. You felt like you had learned — and the neurobiology is on the side of the illusion, not reality.
Here is what AI did to the distillery technically. The fire is now electric and does not go out. Raw material is infinite, because the model generates it from itself. The coil scales horizontally: five prompts in five tabs in parallel. The cube heats faster than ever. The one thing that has not changed — the flask. It is still waiting for a move that does not come.
A separate class of disease — agentic loops. You ask AI to build a competitor analysis system. AI builds. You ask to refine. It refines. "Now automate data collection." It writes a script. "Now a dashboard." Three hours later you have a technically impressive infrastructure that you will use once — to show a friend — and forget forever. Competitors, meanwhile, know nothing of your dashboard. They are simply selling.
There is one more trap, specifically AI-borne: the illusion of completion through the beauty of the artefact. When you wrote a summary by hand, it looked like a summary — crooked, unfinished, obviously a draft. When AI generates a "strategic analysis" with subheadings, bullet points and conclusions — it looks like a document someone has already acted on. The brain does not distinguish "looks like a decision" from "is a decision." Dopamine does not know the difference between shipped and formatted.
I personally spent — not pretending — several sessions asking "deeper" five times in a row, receiving increasingly beautiful responses, and at some point closing the tab with a feeling of completed work. What work? Unknown. But the feeling — crystal clear. This is not a critique of AI. This is a diagnosis of a specific usage pattern. A hammer is not to blame for the fact that you are hitting the table with it instead of a nail.
An honest counterpoint: AI can be not a turbocharger for the cube, but a valve — and that is a fundamentally different mode. If you arrive with a specific decision that needs to be brought to output quickly, the model shortens the path from "I understand" to "done" by orders of magnitude. Write the first draft of a letter to a client. Formulate a decision memo from thoughts that already exist. Make v1 of the text you are "not ready" to write yet. Turn diffuse anxiety into a concrete list of questions for a call.
But this requires one thing — a stopping rule before the session begins, not after. "I am opening this chat to get one specific output: ___. When that output exists — I close the tab and do it." Without that sentence at the start, AI becomes the most perfect coil in human history. With it — a tool that genuinely shortens the distance between thought and action.
Diagnostic question every thirty minutes of an AI session: what will change in the world from my reading this response right now? If the answer is "I'll become smarter" — that is steam. If "I'll send this message," "I'll reject this option," "I'll fix this number as a threshold" — that is already something. The coil got an API. The question is where you connect the flask.
There is one truth about founder research that is uncomfortable to say at a retrospective: most of it is not strategy. It is insurance against disappointment. While you are researching a candidate — they have not yet let you down. While you are studying the CRM market — the team has not yet broken the architecture. While you are reading case studies about entering a new market — the market has not yet said "no." Research is a safe where you hide from feedback. A very smart safe. With ventilation, a Notion database, and a knowledge graph.
I am not saying founders are dumb. On the contrary — they are too smart not to find a justification. Intelligence does not save you here; it is an accomplice: building a consistent argument for one more week without a client call. "I need to understand the ICP better." "I don't feel the market yet." "One more competitive analysis and I'll be ready." Translated from anxious: "I haven't yet found a way to guarantee that reality won't bite me."
Here is the anatomy. A founder wants to hire a head of sales. Logical action: five calls with candidates this week. Instead: three days on Glassdoor, reading other people's JDs, comparing compensation structures at Sequoia portfolio startups, subscribing to a Substack "How to hire your first sales leader," a Loom from someone at YC, a table scoring 14 competencies that will be out of date before the first call. The candidate still has not disappointed. Because the candidate does not exist — there is only an ideal version in the founder's spreadsheet-consciousness. The ideal candidate will never say "I need equity above market." The real one will. And then a decision will have to be made. And research was protecting against precisely that.
The same with CRM. Choosing a CRM is not a complex intellectual task for most startups — it is an administrative decision with clear criteria: how many people in sales, what pipeline, what integrations, what budget. Thirty minutes of comparison plus a trial account = enough. But no. Hubspot vs Salesforce vs Pipedrive vs Close vs six more, Reddit threads about migration, G2 reviews, YouTube overviews from 2022 by people who have long since switched to something else. A founder with two hundred hours of research and not one client call is not a strategist. That is a shaman with a CRM drum. The drum is beautiful. There are no clients.
| Situation | Research trap | Real valve |
|---|---|---|
| Hiring | 12 more articles on interview design | 5 calls with candidates |
| CRM | one more architecture doc | one shipped status / automation |
| Sales | one more positioning matrix | 20 outbound messages |
| Tax / legal | endless forum reading | paid opinion + decision memo |
| Investing | 8 more macro threads | position sizing rule |
| Blog | more sources and metaphors | publish v1 |
In sales and outbound the pattern is even more painful, because the feedback there is the fastest and most brutal. A letter with no reply — concrete information. A call where they say "not interested" — even more concrete. Research protects against both: "I need to better understand the pain points of the segment before outbound" translates as "if I don't write — I'm not being ignored." Twenty outbound contacts with a bad pitch will deliver more useful market information than two hundred hours of customer-development reading. But they will also deliver an uncomfortable truth — and that is why we avoid them.
In investing this takes a particularly refined form, because "more research" always sounds sensible. One more quarterly report. One more analyst comment. One more thread on X. But you formulated the thesis three months ago. Additional information no longer changes the decision — it only postpones the moment when you have to put money on the table or tell yourself "no." A position sizing rule written in advance is the valve: "if the thesis is confirmed — I enter at X% of portfolio, with no additional re-reading." Not recklessness. Hygiene.
A separate trap — content and blog. "I need to understand the topic better before writing." The problem is that "better understand" has no natural end: there is always one more case, one more counterargument, one more nuance. Publishing v1 with honest caveats delivers more than a perfect text after six months of preparation that nobody will see. Knowledge without output is like a gym membership you carry in your wallet for the sake of developing your bicep. A beautiful card. Zero muscle.
And one more subspecies — legal and tax. Here the trap is the opposite: founders either ignore ("I'll sort it out later"), or dive into the rabbit hole of legislation where they have no qualification to be. No Reddit thread replaces a paid opinion + decision memo from a lawyer or auditor. Research here is not for making the decision — it is for knowing what questions to ask the specialist. A legitimate function. But it ends with a call, not one more thread on Hacker News.
The most dangerous thing about founder research-addiction is that it looks like responsibility. Investors want you to "know the market." The team wants "justified decisions." You yourself want to feel like you are not improvising. And research closes all these social contracts at once — without delivering any real result. The most elegant way to be very busy and decide nothing.
For four centuries the smartest people in the West have been describing the same disease — each in their own language, none having cured themselves. Hamlet, ~1600: resolution sicklied o'er with the pale cast of thought, "lose the name of action" — an enterprise of great pitch and moment turns away from its current and ceases to exist even as a concept. Kierkegaard, 1843: broke off his engagement to Regine Olsen and instead of marrying wrote eight hundred pages about the aesthete's inability to commit — a ready technical description of the coil for the nineteenth century; Regine meanwhile calmly married someone else. Buridan's ass in 2026 does not starve between two haystacks — it builds a Notion database to compare them across twelve criteria and waits for a third haystack for statistical significance. Will die full of information and hungry for hay.
And then Charles Sanders Peirce steps onto the stage — and with one move cuts through the whole centuries-old tangle. 1878. The axial idea: our beliefs are in fact rules for action. To clarify what a thought means, you need to determine one thing only — what behaviour it is capable of generating. That behaviour is its entire significance. All of it, without remainder.
Read that again. A thought that changes no action is meaningless. Literally, technically, by definition. Not "not very useful." Not "purely theoretical but elegant." Meaningless — the way zero is meaningless multiplied by anything. William James expanded this into a phrase worth tattooing on the inside of your eyelids: "A difference which makes no difference is no difference at all."
Every unapplied note in that cemetery is a separate headstone. "Here lies the insight about the Zeigarnik effect. Never applied once. 2019–2024." Nearby — a fresher little grave: "Notion database 'Life System v7.' Born on a Sunday afternoon in a wave of inspiration, passed away on Monday morning." The caretaker of the cemetery — friendly, he has many such clients — offers you a plot nearby, neighbourly. With a discount if you set it up through an Obsidian plugin.
Colonel John Boyd invented the OODA loop for air combat: Observe — Orient — Decide — Act. Whoever cycles faster than the opponent stays alive and dictates the tempo. Boyd's key observation: people most often get stuck precisely at Orient, at sense-making. In the clinical picture of intellectual addiction, the loop can be heard from afar: OO-OO-OO. Observe — Orient — Observe again — Orient again, like a skipping record. Decide and Act are simply physically blocked. There is a third state that even Boyd in a fighter cockpit did not anticipate, because in combat you cannot afford it: total inaction, perfectly disguised as eternal, responsible orientation. Paralysis masquerading as thoroughness.
And here is the real insidiousness — the invisible bill for non-doing. It does not arrive in the post with a red "overdue" stamp. While you are at Orient, competitors are running their loops to completion, to Act. The market shifts. The window closes. Opportunity cost here is abstract, so unnoticed, but entirely real: every hour of distillation is an hour when the first drop into the flask did not fall and will never fall specifically for that hour. You simply notice one day that the opportunity is gone, a year has passed — and you are still "almost ready to start."
Now into the mathematics — and this is precisely where it is easiest to overclaim. The secretary problem: if you are choosing the best from a sequence of options that pass by without the right to return, the optimal strategy is to inspect the first ~37% and then take the first one better than everything seen. But 37% is not a life commandment. It is a mathematical slap in the face for a very specific formulation. The rule does not mean all decisions should be chopped after the third source, like a drunk surgeon. It means something else: in many tasks there is a moment after which further study looks less like wisdom and more like fear of signing. The probability of grabbing the best while researching infinitely and never stopping — is exactly zero. Reality does not reward the longest research. It rewards the one who stopped in time.
Alfred Korzybski, a hundred years ago: the map is not the territory. The intellectual addict does worse than simple confusion — they endlessly detail the map of a country they will never visit. 8K resolution, every alley drawn in. The map becomes a work of art. The bags are unpacked. The ticket is not bought.
And over all of this hangs the Oskamp curve from the first section: confidence flies upward, accuracy frozen dead. The philosophers named the disease. Boyd showed where it kills. Mathematics showed where you should have stopped. None of them stopped the apparatus. Because apparatuses are not stopped by understanding the apparatus.
 *Valve open: the flask that was empty for years finally catches the first golden drop.*There is a simple formula that nobody calculates because the result is uncomfortable. Information ROI = change in decision quality / (time + cognitive residue + delay cost). The numerator — what actually changed in your head. The denominator — not just hours, but also the mental residue rotting in the background while you are "still researching a bit," and the cost of each day without a decision. Divide the first by the second. Less than one — you are not learning. You are heating air.
A question the apparatus never asks itself: what specifically will change in my decision after this source? Will the decision threshold change? Risk model? Next action? Kill criteria? Position size? Will I send a different message? Ship a different artefact? If the answer is "no, but it's interesting" or "no, but I better understand the context" — the source is not producing knowledge. It is producing steam. Hot, aromatic, intellectually-rich steam, which condenses on the ceiling and falls down your collar.
Those five sentences are not a motivational poster for LinkedIn. They are a customs checkpoint. Every idea entering the coil must show its passport: which action do you change? No answer — turn back. Not "save in Obsidian." Not "come back to later." Turn back. Because "later" is where all your future selves live, the ones who were supposed to have finished reading, rethought everything, and finally done something. It is very crowded there. And none of them exist.
The problem with a stopping rule is not that you do not know it. You know it. The problem is that a stopping rule without an external cost is a decorative sign reading "Please don't litter" next to a rubbish bin. Pretty. Powerless. This is precisely where two people whose data is worth tattooing alongside Peirce come in. Peter Gollwitzer, 1999: implementation intentions — the formula "when situation X, I do Y" — in a meta-analysis of 94 tests delivers an effect of d=.65. This is not advice to "be disciplined." This is engineering: you bind an action to a trigger in advance, and the decision is made not by willpower in the moment of temptation, but ahead of time, with a cold head. And Ariely and Wertenbroch in 2002 showed: self-imposed deadlines really do beat procrastination — but only when there is a tangible price for missing them. A stopping rule without a cost is a mood. A stopping rule with a cost is a protocol.
Here are six moves. Not "tips." Not "5 productivity hacks" — the very thing I spent thousands of words talking you out of. This is the valve — and it must be opened physically, with your hands, right now, without closing this tab.
Move one: name the client of the research. In one sentence, formulate which specific decision or action all this research of yours was meant to serve. If the honest answer is "I was just curious" — stop: this is fire without a flask, and that too is fine. Just do not pretend to yourself or to your calendar that it was work.
Move two: stopping rule. Complete the sentence: "I stop researching this topic the moment I can do ___." Fill in the specific action. Remember the maths: the optimal moment is almost always already in the past. So the honest answer to "when should I stop?" is usually one — "right now." Not "after one more."
Move three: Peirce test. Take every "important" thing you saved, highlighted or distilled this week, and ask one merciless question: "What specific action of mine does this change?" None → it is not a thought, it is noise in a pretty wrapper. Delete or let it go without ceremony. Saving non-differences is what building a mausoleum is.
Move four: output tax. Every thirty minutes of research cost one artefact to the outside: a paragraph, a decision memo, a sales message, a ticket, a commit, a booked call, or a deleted source. Not "a note to myself." A trace in the world. No artefact — next thirty minutes of research are forbidden.
Move five: if-then plan with a cost. Wire it up in advance, per Gollwitzer: "When I open a third source — I write a 7-line decision memo before reading further." "When I ask AI for a third version — I choose one and ship v1." And add the cost per Ariely: name specifically what will happen if you break the rule — who you will tell that you did not do it, what deadline you will announce out loud. Without skin in the game your stopping rule is a decorative sign.
Move six: close the tab. Either action, or the bin. Not "save for later." "Later" does not exist — that is where all your future selves live, who were also supposed to have read everything, reconsidered, and finally done something. It is very crowded there. And none of them exist.
And now the most important thing, and it is not on the list, because the list you can also turn into one more summary about lists. Before you close this text, do one single thing from these five: send one message that makes your stomach cold; create one ticket; delete one research queue you have been "getting to" for years; publish v1; write a 7-line decision memo and put a date under it. Not "understand." Not "save for the weekend" — those weekends do not exist, they have not arrived a single time in the entire history of read-it-later. Do something so that a trace exists in the world that was not there before this paragraph.
That psychologist in 1965 put down Oskamp's fourth sheet and asked for the fifth. He was wrong on three questions out of four — exactly as at the start, sixty years and an infinity of tabs ago. That psychologist's flask stayed empty — and we do not even know his name, because history does not remember the person who only distils. It remembers the ones from whose flask something dripped.
And one last thing, so that no misunderstanding remains between us. Do not extinguish curiosity. It made us human, it wrote this text, it is beautiful. But without a valve it becomes not an engine but a heater for anxiety. You are not thinking more deeply — you are simply heating the cube. You can keep the fire. I am asking for one thing.
Apparatuses are not stopped by deeper understanding of the apparatus. You can study every bolt of this cube and die beside it — well-read and dry. Apparatuses are put to work by one move. The valve.
Close the tab. Open the flask.