There's a passage in Annie Dillard's The Writing Life, drawn from Thoreau. If you want to find a beehive and you don't know where it is, you catch a bee in a jar. Hold it for a minute. Let it go. The bee always flies toward the hive. But the bee is faster than you — you lose it almost immediately. So you catch another bee. Hold it. Release it. Run in the direction it goes. Catch another. And another. After fifty bees, you've found the hive.1
Johann Hari runs his entire research methodology on that metaphor. Every book starts with a written list of questions. Each question generates a list of people who could answer it — who could carry this story for the reader. He interviews them. At the end of every interview, he asks: who else should I talk to? Each interview points to the next bee. By the end of a project he's done between two and three hundred interviews. Eighty percent of them never appear in the book.1
If you only see the finished book, the eighty percent looks like waste. If you've ever run a long investigation, you know it isn't. It's the bees you had to chase to find the hive.
Hari is naming a specific architecture for investigative-research projects whose subject is systemic. You don't know in advance where the systemic answer lives. You can't go straight to it because nobody fully holds it — it's distributed across hundreds of people, each of whom holds a fragment. The bee-and-jar method is the iterated-fragmentation strategy that lets the writer reconstruct the hive from the bees.
The mechanism is recursive: each interview produces (a) some fragment of the answer, and (b) two or three pointers to other people who hold adjacent fragments. The total quantity isn't the work product. The navigation through 200-300 interviews is the work product. The eighty percent that don't make the book aren't waste — they're the navigation. They're how the writer found the twenty percent that do.
This is structurally different from interview-as-quote-extraction. Most journalism treats interviews as a way to get usable quotes; the goal is the quote, and an interview that doesn't produce a quote is a failed interview. Hari treats interviews as triangulation moves. A given interview may not be quotable, but it changed Hari's map of where to go next — and that change of map is what eventually produced the quotable interview eight pointers down the chain. The success of the failed interview is the success of the chain.
Hari has a defense ready for the eighty percent: it's the same as a police investigation that interrogates twelve suspects and finds out the twelfth one did it. If you say you wasted your time on the first eleven, you're applying retrospective knowledge to prospective work. The investigator didn't know in advance the twelfth one did it. The eleven negative results were the only path to the positive result.1 You can't have the twelfth interview without the first eleven, even when the first eleven don't directly produce evidence.
The justification matters because it generates psychological permission for the methodology. Writers who try to optimize their interviews — only meeting with people they're sure will produce material — produce thinner books, because the optimization closes off the chain of pointers that would have led to the unexpected breakthrough subject. The book gets impoverished by efficiency. Hari's tolerance for "wasted" interviews is what permits the chain to reach the hive.
Hari adds a specific operational claim: wherever possible, he sits with people in person, because "you get 90% more when you sit with people in person."1 The 90% figure is rhetorical, not measured. The underlying claim is that physical presence — being in the same room, sitting on the same floor of the Vegas tunnel, eating across the same diner booth in Brooklyn — produces material that phone or video interviews cannot. The Chino Harden case from Hari's book on addiction makes the case: Chino's opening line in their first meeting — "I was conceived when my mother who was a crack addict was raped by my dad who was an NYPD officer" — is the kind of thing that arrives only in physical presence. The phone interview would have produced a different opening line, or no opening line at all.
The in-person premium has cost. It requires travel, time, money, scheduling. But Hari treats it as non-optional for the work the books are doing. The eighty-percent-discarded budget includes travel cost, and Hari accepts it as the price of admission to the chain.
This page gives the vault its single most articulated research methodology and grounds several adjacent pages. It deepens Tetris-Database Research Method (Bilton) — Bilton's chat-log-and-photo-geolocation forensics is the same iterated-fragmentation strategy applied to text-trace rather than face-to-face material. The two methods are siblings: Bilton's bees are digital, Hari's are physical, both methodologies are recursive. It connects to Three-Phase Research-Think-Write Method (Orlean) — Orlean's Phase 1 (years of research, no writing) is the temporal envelope inside which the bee-chasing happens. Orlean names the time; Hari names the method. And it sits in productive tension with Karlsson 5-Step Writing Method — Karlsson's incubation methodology produces essays from interior development; Hari's bee-and-jar produces books from exterior triangulation. Both are practitioner-grade, neither is universal, and the choice depends on whether your subject is interior (philosophical, autobiographical, personal-essay) or systemic (political, social, historical investigation).
Hari is working on his book about addiction. Question list pinned to the door: What causes addiction? What is addiction? Why were drugs banned? What are the alternatives? Decomposition into people-who-could-carry-the-story: drug dealer, cop who supports the drug war, cop who no longer supports, Mexican cartel member, recovered addict, family of addict, harm-reduction worker. He starts the bee-chase.
He's living in New York. He knows about Vocal NYC, a group that does work in deprived parts of the city. He tells them he wants to meet some interesting drug dealers. They introduce him to Chino Harden. The introduction is a single edge in the chain — Vocal NYC was one bee, Chino is the next bee. Vocal NYC may not appear in the final book at all. They were navigation. They got Hari to Chino.
Chino is the kind of subject the chain was always pointing at without anyone knowing it. He arrives at the diner. He opens with the conception-rape sentence. Hari spends years interviewing Chino — not one session but a relationship developed over time. Chino becomes a load-bearing figure in Chasing the Scream. The book's analytical apparatus is built on what Chino's life shows about the drug war.
But the path from question-list to Chino has fifteen interviews between Vocal NYC's recommendation and the first Chino conversation, none of which the reader ever sees. The first five established Hari's knowledge of the trade well enough that Vocal NYC took him seriously. The next ten were the wide net Hari cast simultaneously to make sure Chino wasn't his only drug-dealer source. All fifteen are eighty-percent. None of them are waste. They are the architecture that made the Chino interview possible and that made Hari able to read what Chino was telling him when the conversation finally happened.
It's a Tuesday morning. You're working on a long-form piece about a regional opioid crisis. You've done ten interviews — three sheriffs, two harm-reduction workers, four family members of addicts, one ER doctor. You've taken notes, recorded each, transcribed. You have a list of who else should I talk to names from each, sorted and ranked.
Today you have an interview scheduled with a recovered-addict-now-counselor named Maria. You've prepared — you've read her published interviews, the methadone-clinic data from her county, the local newspaper coverage of the deaths Maria's clinic has tried to reduce. You're not going in with a thesis. You're going in with the question list pinned to your wall and a specific subset of it that Maria might be positioned to help with.
You meet at her clinic. You sit in her office. Before the recorder is on, you tell her why I'm here — a brief, honest sentence about why this project matters to you personally. She nods. The recorder goes on.
You don't lead. You ask: what do you want me to understand about the people who come through your door? Maria talks. She tells a story of one specific patient, then another. She makes a passing reference to a guy named Jimmy who runs a peer-support program in the next county. You make a note. The bee just pointed to Jimmy.
You ask: what do you wish the journalists who cover this got right? Maria gives you forty minutes of structural analysis you hadn't read in any source. She mentions a state policy change in 2019 that she says everyone misreads. You make a note. The bee just pointed to the policy change.
At the end you ask: who else should I talk to who would tell me something you can't? Maria gives you three names. You write them down. You thank her, leave the clinic, and start running searches on Jimmy and the policy change before you've made it to your car.
The interview wasn't the destination. It was a node in the chain. The destination is the hive, and you're on bee eleven of fifty.
You'll know the chase is broken when the interviews stop producing pointers. When every interview ends without anyone you should talk to next, when the who-else-should-I-talk-to question consistently produces "I don't know" or repeats of names already on your list, you've either saturated the network or you've been talking to people who don't have your network's structure mapped. Time to find a different cluster of bees.
You'll know it's working when the chain is self-extending — when the names produced by interview N are not on the list produced by interview N-1, when the geography of the project keeps expanding sideways, when the question-list pinned to your door starts getting answers from people you wouldn't have predicted. The expansion is the chain doing its work. The eighty percent is the cost, and the cost is non-negotiable.
Hari's method is well-attested across long-form journalism — Robert Caro's biographies, Susan Orlean's literary nonfiction, David Grann's investigative work — though none names the bee-and-jar metaphor specifically. The method has implicit cognitive-science support in the network exploration literature on expert problem-solving: experts don't optimize at each step; they tolerate locally-suboptimal moves that improve their map. The eighty-percent-discarded tolerance is the formal name for this in research methodology.
The open tension: how does the method scale to writers with less institutional support? Hari has publisher advances, research budgets, and the established credibility that makes Vocal-NYC-tier introductions possible. A first-time investigator without these has trouble running the same method at the same scale. The vault should note: the method's structure (recursive triangulation through chained interviews) is portable, but the scale (200-300 interviews) is not. Earlier-career writers may run abridged versions — 30-50 interviews — and produce honest work within the budget they have.
Hari and Bilton converge on the iterated-fragmentation strategy from radically different content domains. Bilton works with digital traces — Silk Road chat logs, photos with metadata, geolocation reconstruction. Hari works with physical presence — diners, prisons, tunnels under Vegas. Both writers tolerate enormous amounts of material that doesn't make the book, and both produce books whose final shape couldn't have been predicted from the question-list at the start. The convergence reveals that the recursive method is medium-independent — the same epistemic architecture runs through chat logs and tunnel-floor sittings, and the writer who learns one can extrapolate to the other. The divergence is operational: Bilton's bees can be chased from a desk; Hari's require travel. The choice between methods isn't methodological — it's about what kind of material your subject leaves traces in.
Hari and Karlsson diverge sharply. Karlsson's incubation model says the essay is born from the writer's interior, after a year of internal development. Hari's bee-and-jar says the book is born from 300 strangers, after years of triangulation across them. These are not compatible methodologies for the same project. They are appropriate for different projects. Karlsson's method produces philosophical and personal essays. Hari's method produces investigative books about systemic phenomena. The vault holds both because the vault doesn't have one project — different writers serve different projects, and the methodology is determined by the subject's epistemology, not by writer preference.
What Hari is doing with bees and jars belongs in two adjacent places worth noticing.
Behavioral Mechanics: Network Mapping as Intelligence Discipline — Intelligence analysts mapping a criminal or political network start with one or two known nodes, interview each, generate a who-else list, prioritize, repeat. Same shape as Hari's bee-chase. The analyst tolerates "wasted" interrogations the way Hari tolerates "wasted" interviews — both know retrospectively-discarded work is the only path to the work that lands. The honest journalist and the honest analyst share more methodology than either shares with the writer-of-essays. This has implications for how the writer holds the ethics of the practice: the techniques sit close to intelligence tradecraft, and the writer should be aware of the adjacency rather than pretending it isn't there.
Eastern Spirituality: Satsang as Iterative Darshan with Many Teachers — The Indian traditional way of seeking truth was to sit with many teachers, each pointing toward the next, until the seeker found the teacher who could carry them through. The seeker who sits with only one teacher gets one teacher's distortion. The seeker who runs the chain gets cross-validation. Same shape as the bee-chase. The deeper recognition that crosses both domains is that no single source carries the whole answer — not in addiction-research, not in liberation-seeking. The honest seeker runs the chain. The writer who refuses to run it has already decided what the answer is.
The Sharpest Implication If Hari's bee-and-jar is the right method for systemic-investigation books, then the proliferating Substack-essay model — single-writer, no-interviews, distilled-from-reading-and-personal-experience — is structurally underspecified for the kind of claims that long-form journalism used to make. Writers who try to do systemic-investigative work in essay form, without the 300-interview chain, are vulnerable to producing confident-sounding work whose actual epistemic foundation is thin. Hari's method is what honest systemic claim costs. The corollary: when you read systemic claims that didn't go through the chain, hold them at higher skepticism than the writer's confidence would suggest.
Generative Questions