Reading the Analog Signal

There is a number I held back in the first piece on purpose. I said the load-bearing skill was attention, not warmth, and that the evidence rested at its most quantitative on a single study of a hundred-odd telephone calls in one English police force. I named the shape and withheld the figure, because a figure without its scope is the thing this whole series exists to distrust. This is the piece where I spend it. So, plainly:

In 2022, a research team coded 105 recorded telephone calls between police source handlers and their human sources — one English force, calls averaging seven minutes, inter-rater agreement of κ = .77 — and partitioned the variance in how much usable intelligence each call produced. Overall rapport tracked yield at r = .69. Then they split rapport into its parts. The attention component — focused listening, tracking what the source said against what they meant, remembering — explained sixty-nine percent of the variance. Positivity, the warmth-and-likability component, the thing every “people skills” seminar sells: four percent, and it didn’t even clear significance (r = .19, p = .051). (Nunan et al. 2022, Psychiatry, Psychology and Law)

Sit with the gap. Sixty-nine to four. The thing we all assume is the engine of getting information out of a person turns out to be, in the one place anyone bothered to measure it cleanly, almost noise.

And now the caveat, in the same breath, because the caveat is the claim: that is one study. One force, 105 calls, telephone-only, not a random sample, and the authors say so themselves — the result may not generalize to the individual handler, let alone to a discovery call over Zoom in 2026. If the attention finding rested on Nunan alone, I’d file it under “suggestive” and move on. It doesn’t rest on Nunan alone. That’s the rest of this piece.

The likability assumption, stated flat

The warmth control, dialed down. The knob everyone reaches for — and it barely moves the result.

The popular description of the skill goes like this: some people are naturals. They walk into the room, the temperature rises a few degrees, people open up to them because they’re likable. The rest of us — the ones who’d rather read the manual than work the room — are at a structural disadvantage, and the best we can do is fake enough warmth to get by.

I believed a version of this for a long time, because it’s the only version anyone teaches. It’s also, on the evidence, describing the wrong mechanism. The Nunan split is the cleanest single demonstration, but the reason I’ll say “attention, not warmth” as something stronger than a one-study hunch is that five independent literatures — built by people who were not talking to each other, measuring different outcomes in different decades — arrive at the same shape from five different directions. The convergence is the robustness. Not the single number. Here is the convergence.

Five readings of the same channel

Five jacks, one channel. Five independent literatures reading the same signal from five directions.

Inside motivational interviewing, the relational paths went quiet and the technical path carried the signal. This is the most-validated body of work that exists on getting people to say true things about themselves — more than two hundred randomized trials. In 2018, a process meta-analysis did something most of the field avoids: it tested why the method works. It put two hypotheses side by side. The technical one — the clinician’s structured skill shapes what the client says, which predicts the outcome. And the relational one — empathy and warmth drive the outcome directly. The technical path held: clinician skill predicted change-talk at r = .55. The direct relational paths — empathy to outcome, “MI spirit” to outcome — were both non-significant, r = −.04. (Magill et al. 2018, Journal of Consulting and Clinical Psychology) The warmth was in the room. It just wasn’t the part doing the work.

I want to be careful here, because this is exactly the kind of finding the triumphant version of the argument would over-read into “warmth doesn’t matter.” It does matter. The broader psychotherapy record is unambiguous that the relational stance is robustly associated with better outcomes — the therapeutic alliance correlates with outcome at r = .278 across 295 studies and more than thirty thousand patients. (Flückiger et al. 2018, Psychotherapy) That’s the foundation, and it’s real. The honest reading is the one that holds both: warmth is the room; attention is the engine. You need the room. You don’t move anything with the room.

The investigative-journalism tradition arrives at the same place, from craft rather than from data, and says it about the interviewer’s personality directly. John Sawatsky, who trained a generation of broadcast journalists, drew the distinction that does the most work: “The goal of a conversation is to exchange information; the goal of an interview is to receive information.” (Paterno, “The Question Man,” American Journalism Review, 2000) His observation about who is actually good at this is the one I keep coming back to: the best interviewers, in his training data, “have a blander personality, they’re not the life of the party.” The interviewer who fills the room with personality is crowding out the source. And then there’s Robert Caro, who put it as sharp as anyone has: “silence is the weapon, silence and people’s need to fill it — as long as the person isn’t you, the interviewer.” (Caro, Working, 2019, p. 137) Caro writes SU — Shut Up — in his notebook to stop himself from filling the silence the source needs to fill. The highest-leverage move in the craft is to not perform.

The actor-training tradition gets there from a third direction, which is the one I least expected. You’d think acting would be the home of manufactured warmth — the place where the integrity case for “fake it” lives. It isn’t. Sanford Meisner defined the craft as “living truthfully under imaginary circumstances,” and his core exercise relocates the source of a real response from the actor’s own internal recall to the other person in the scene. It trains attention outward. Sharon Carnicke’s corrective scholarship makes the same point about Stanislavski himself — that the Anglophone “Method,” with its emphasis on dredging up your own remembered feelings, distorted a system that was actually about directed attention and physical action. (Carnicke, Stanislavsky in Focus, 2009) The trainable craft underneath performed engagement is not generating heat. It’s pointing your attention at the other person and keeping it there.

The emotional-labor literature closes the loop by showing what happens to the people who get this wrong. When you perform warmth you don’t feel — surface acting — it predicts impaired well-being at ρ between .39 and .48 across 95 studies and 494 effect sizes. (Hülsheger & Schewe 2011, Journal of Occupational Health Psychology) A 2025 replication with 1,252 service workers found surface acting drove burnout at β = 0.592 — a large effect — while deep acting, the kind grounded in actual attention to the person, went the other way, mildly protective at β = −0.087. (Üngüren et al. 2025, Frontiers in Psychology) Same outward behavior — warmth shown to another human — opposite consequences, depending entirely on whether there’s attention behind it or just a mask. And the workplace-personality record points at the same component: the second-order meta-analysis on extraversion at work found the modest advantage extraverts hold runs through the positive-emotions facet, not through sociability — “sociability itself confers few workplace benefits.” (Wilmot et al. 2019, Journal of Applied Psychology) The room-warming charm we file under “people person” is, in the data, the part that doesn’t pay.

Five literatures. Police source-handling, clinical interviewing, investigative journalism, the psychology of performed emotion, personality science. None of them built to confirm the others. All of them landing on the same thing: the operative variable is attention, and warmth is the foundation it stands on, not the force that moves the work. That convergence is the reason I’ll say it out loud. Knock out any one of the five and it still holds at four.

What this does to the quiet one

The status LED. One color, mostly off — and it turns out that's the right hardware for the job.

Here is where the finding stops being abstract, because it reverses the thing I started this project braced to confirm.

The first piece set up a claim and promised to prove it: that the “introverts are worst at this” framing isn’t wrong about the people — it’s wrong about the skill. It’s right about the popular self-help description of the work, which is warmth-display, and someone like me really is worse at performing warmth on demand. It’s wrong about the mechanism the controlled evidence actually identifies, which is attention — focused listening, tracking implication, disciplined remembering, structured questioning, holding silence instead of filling it. Re-specify the skill correctly, and the supposed deficit turns into an aptitude. The one who’d rather read the room than fill it, who is comfortable with information density and has a low tolerance for performing, is built for the operative variable, not against it.

This is not a consolation prize I awarded myself. It’s the cleanest result the corpus produced, and it’s why the candidate it disqualifies is the right one to do the work. I spent twenty-five years assuming the social layer was a tax I paid to get to the work. It turns out the part of the social layer that actually moves the outcome is the part that runs on the same hardware I’ve been debugging with the whole time — read the inputs carefully, track state against what’s claimed, don’t trust the surface, hold position when the instinct is to interrupt. The skill closer to the one I have isn’t charisma. It’s instrumentation. You read the analog signal off the person the way you’d read a trace off a misbehaving system: patiently, attending to what’s there rather than to the story the documentation tells you should be there.

Which is good, because my own warmth output has the dynamic range of a status LED. One color. Mostly off.

The signal you actually want is below the noise floor

The trace you came for sits low, near the noise floor — the part the source can't put into words.

Now the harder half, the one that survives even after you’ve accepted that attention beats charm. Reading the channel well doesn’t guarantee you the data you came for — because the most valuable thing the source knows is precisely the part they can’t transmit.

This is the tacit-knowledge problem, and every one of these literatures eventually bottoms out on it. Polanyi’s formulation is the one everybody quotes because it’s exactly right: “we can know more than we can tell.” (Polanyi, The Tacit Dimension, 1966, p. 4) The requirements-engineering literature carries the same admission in operational form — observation, BABOK notes, “is not suitable for evaluating knowledge-based activities since these are not directly observable.” The expert genuinely cannot serialize the most important thing they know. Ask them directly and you get the articulable surface — the prepared answer, the cleaned-up version, the part that was already in language. The signal under the noise floor is the part that never made it into words, and direct questioning systematically under-recovers it.

The user-research tradition has spent decades engineering around this, and the move they converge on is the same one the journalists found by instinct: do not trust self-report; reconstruct behavior. Don’t ask the person what they want, or what they do, or how they decide — those questions query a layer the person can’t reliably read out. Ask them to walk through a specific real incident, backwards from the moment it happened, anchored on concrete detail. The jobs-to-be-done interview reconstructs an actual purchase timeline rather than asking about preferences, on the stated premise that “the human mind is not designed for self-knowledge.” (Chin, “Putting the JTBD Interview to Practice,” Commoncog) Contextual inquiry changes the prompt entirely — instead of “tell me what you do,” it’s “let me watch you do it and ask in the moment.” Sawatsky’s clearest example is a reporter trying to learn whether a child went to school hungry. Ask “did you eat breakfast?” and you get a defended, false answer. Ask “what’s the first thing you did when you got up? Then what? Then what?” — walk the chronology — and the truth arrives in context, unguarded. You don’t decode the tacit signal by turning up the gain on direct questions. You change what you sample.

The most-taught technique produces more false confessions

Pressure the channel hard enough and the needle pegs in the red — it manufactures a reading that was never there.

There’s a darker version of getting this wrong, and it’s the one with the highest stakes, and — this is the pattern that runs through the whole series — it’s also one of the most heavily trained methods in the field.

The first piece named the methodological stance the collection runs on: read the source, distrust the vendor firmware. The fields I’m reading are full of techniques that are institutionally embedded, certified, brand-named — and weakly validated, or validated in the wrong direction, the moment you go back to the controlled evidence. The master cargo cult is the warmth-and-charisma “people skills” framework itself, which I’ve now spent five literatures dismantling. But it has a sharper cousin worth naming because the failure mode is so stark.

The Reid technique is the canonical accusatorial interrogation method in the United States — the one that opens with “direct positive confrontation,” presenting the suspect with an assertion of guilt and pressing toward a confession. It has been taught for decades. And the controlled record on it is damning: confrontational, accusatorial approaches produce false confessions at an odds ratio of 4.41 relative to rapport-based information-gathering — that’s more than four times the false-confession rate. (Catlin et al. 2024, Campbell Systematic Reviews) Roughly 29% of DNA exonerations in the United States involved a false confession. (Innocence Project) In 2017, Wicklander-Zulawski — a training provider that says it has worked with a majority of US police departments and trained more than 200,000 officers — announced it would stop teaching the Reid method after thirty years. Its CEO’s reason was about as flat and damning as it gets: “Confrontation is not an effective way of getting truthful information.” (The Marshall Project, 2017) The most-taught method doesn’t just underperform. Pressure the channel hard enough and it manufactures signal that was never there — the source tells you what you’ve pushed them toward, and it isn’t true.

I’ll keep the caveat attached, because it’s the rule: most of that evidence comes from mock-crime laboratory paradigms, heavily weighted toward US college students, and the field-study record is thinner. The senior engineer is not running interrogations. But the lesson generalizes precisely because the mechanism does. The confrontational, theory-first approach — I already know what this is, I just need you to confirm it — doesn’t extract truth. It extracts compliance with your prior. The structured, rapport-based, hypothesis-testing approach — I’m here to learn what this is, not to prove what I think it is — is the one with the empirical support, in interrogation rooms and, by extension, in the discovery call where you’re trying to find out what the work actually is before you’ve decided what it must be.

And structure is the part that does the work, not the warmth wrapped around it. The personnel-selection record is blunt about this: structured interviews predict outcomes at ρ = .51 against .38 for unstructured “let’s just talk” — a finding that’s held for two decades, with a corrected 2022 estimate of ρ = .42 still preserving the ordering. (Schmidt & Hunter 1998; Sackett et al. 2022) Same questions, same order, calibrated against what you hear. The clinical-interview tradition operationalizes it down to a ratio — a proficient motivational interviewer reflects at least as often as they question, and listens at least twice for every question once they’re good. The discipline isn’t being warm. It’s being structured enough to hear, and structured enough not to lead.

What’s known, and what isn’t

So the shape of it, marked the way the first piece promised to mark everything — what I’m sure of held apart from what I’m guessing at.

Verified across five independent literatures: attention, not performed warmth, is the variable that does the proximate work of getting usable information out of another person. That’s the strongest finding in the whole corpus, and it’s strong because of the convergence, not because of any one number. One study, carried with its limit: the single cleanest partition of that variance — sixty-nine percent attention, four percent warmth — comes from 105 telephone calls in one English police force, and travels nowhere without that sentence attached. Well-supported and pointed in one direction: the confrontational, theory-first technique extracts compliance, not truth; the structured, hypothesis-testing one extracts more, and more reliably. Genuinely open: whether any of this transfers cleanly from the in-person, mostly-Western settings where it was measured to a video call with a client in 2026 — the evidence is in-person and telephone, the samples are WEIRD, and the transfer is inference, not measurement. I’d rather tell you that than pretend the studies were run in the room I’m actually working in.

What changed for me, reading all this, is smaller and larger than a tactic. The interface skill isn’t a personality I lack. It’s an instrument I can learn to read — and the reading is the part I was already built for. The warmth subroutine can stay off. The attention is the firmware that ships.

The signal can be read. The next problem is that it decays — the conversation is volatile memory, and what feeds the machine downstream isn’t what was said, it’s your fading reconstruction of it. That’s the piece on writing to disk.

This is the keystone of the input line — the acquisition stage. It proves the reframe asserted in You Are the Interface and pays off the statistic that piece withheld. The signal you read here has to be persisted before it corrupts: Write to Disk. The same tacit wall returns on the other direction of flow, in The Exocortex — you can’t retrieve what was never encoded. And the aptitude this piece vindicates has a cost the closing piece counts: Thermal Limits.