books/theory/THEORY.md

# The Relevance Theory

*The theory behind the project, revised July 2026 to match what the evidence actually supports. The previous version is preserved at `../archive/THEORY_v1.md`. Evidence tiers referenced throughout are defined in `EVIDENCE.md`.*

**Renamed** from "Cognitive Balance Theory." The old name described the wrong thing — the theory isn't about balancing two modes, it's about how the brain decides what deserves attention at all. The relevance-sorting claim is the original contribution; the network model is borrowed scaffolding.

---

## The claim, in one paragraph

Thinking costs energy, so the brain triages incoming information rather than processing it evenly. It evolved a rule for triage based on **proximity, confirmability, and actionability** — three tiers of relevance corresponding roughly to what you could walk to, what you could ride to, and what lay beyond. Modern media delivers information from the third tier while carrying every signal the brain uses to identify the first. The result is not distraction but misfiling: sustained, physiologically real concern about matters no action of yours can touch. The remedy is not consuming less but re-sorting deliberately.

Everything below either supports this or is scaffolding for it.

---

## Part 1 — The thrifty brain
**Tier A. This part is solid and can be asserted.**

Everything alive runs on an energy budget, and cognition is expensive. The brain therefore predicts rather than perceives, reuses models rather than constructing them, and reaches for heuristics rather than deliberation wherever it can get away with it. This is why optical illusions work at all: you are seeing a prediction, corrected by input, not input itself.

The same economy operates at high levels. Stereotypes and generalisations about people are the identical mechanism running on social categories — uncomfortable to say, but the discomfort is about the consequences, not the mechanism.

Three independent traditions converge here: predictive processing and the free energy principle (Friston, Clark), cognitive miser theory (Fiske & Taylor, 1984), and System 1 / System 2 (Kahneman). The theory doesn't need to re-derive any of this — it needs to cite it and build on it.

**The consequence that everything else rests on:** a brain this economical *must* triage. It cannot afford to evaluate every input on its merits. So it needs a cheap rule for deciding what to spend on. Part 2 is a proposal about what that rule is.

---

## Part 2 — The three tiers of relevance
**Tier C. This is the original contribution and it is untested. Both halves of that sentence matter.**

Nobody has tested this model. It emerged from reasoning about what information environments looked like for most of human history, and it may be wrong. It is presented as a hypothesis that generates predictions, not as a finding.

**Tier one — within a day's walk.** Direct effect on you, personally confirmable, and actionable. A flooded road, a neighbour's illness, a fire. Two properties always travel together here: you can check it, and you can do something about it. The proposal is that the salience machinery was calibrated on this tier.

**Tier two — within a day's ride.** Indirect effect, arriving through people you know, usually unverifiable but reasonably trusted. Prices, disputes, news from the next valley. Not directly actionable, but it is what context is made of — it's how tier-one facts acquire meaning. Trust here is social rather than evidential, which is a mechanism rather than a defect.

**Tier three — beyond the horizon.** Unconfirmable and unactionable. Wars, famines, rulers, catastrophes.

**The historical observation the theory rests on:** every culture before ours had a tier three, and handled it deliberately — story, myth, rumour, ballad, sermon. That handling was not a failed attempt at accurate reporting. *The form was the filing.* Distant news arrived wrapped in markers of its own distance, and you knew from how it reached you that it was not something to act on before nightfall.

---

## Part 3 — The collapse
**Tier C.**

Modern media delivers tier-three information carrying tier-one signals: present tense, high definition, real time, human faces at conversational distance, and notifications indistinguishable from ones about your own family. Every marker the machinery used to separate near from far has been stripped or inverted.

This did not happen through malice or even primarily through commercial design. It happened through the ordinary logic of media getting better. **Improved delivery is the mechanism.** A war photographed and transmitted in seconds is a genuine technical achievement that also happens to defeat a filtering system built over a hundred thousand years.

This reframes the standard critique of social media. The reward-hacking account is true but shallow. The deeper harm is that these platforms destroy the tier distinction and then require you to switch between tiers several times a minute — a war, a friend's baby, a work message, a stranger's opinion, all arriving through the same slot with the same weight.

---

## Part 4 — The actionability gap
**Tier C. The least developed idea here and possibly the most valuable.**

The distinctive modern injury is not distraction. It is **caring intensely about what you cannot touch** — genuine, physiologically real concern about matters no action of yours can affect, repeated daily, for years.

Tier-one information always came with something to do. Tier-three information historically came without urgency. Modern tier-three information arrives with the urgency and without the action. That specific combination — high salience, zero agency — has no evolutionary precedent and no inherited cultural handling.

What it produces: helplessness, compassion fatigue, moral exhaustion, and the particular shame of scrolling past a catastrophe.

**The necessary guardrail:** this is not an argument for caring less about distant suffering. It is an argument that caring without any channel for action is a state we have engineered at civilisational scale and never designed for. The response is to build channels, or to consciously accept some knowledge as knowledge rather than as summons — not to narrow the circle of concern. Any version of this theory that ends in "stop caring about far-away people" has gone wrong.

---

## Part 5 — The network scaffolding
**Tier A on the anatomy, Tier C on the interpretation. Borrowed, not original. Keep it contained.**

Three networks, in plain language:

| Network | What it does | Failure mode |
|---|---|---|
| **Default mode** | Reflection, narrative, autobiographical memory, mind-wandering | Rumination, paralysis |
| **Task-positive** | Focus, execution, external monitoring | Anxiety, burnout, depletion |
| **Salience** | Detects what matters; mediates between the other two | Poor prioritisation, false urgency, switching failure |

Menon's triple-network model is well-supported, with causal evidence for the salience network's mediating role from dynamic causal modelling and related methods. But **these are functional networks that overlap and co-activate, not discrete modules with a switch.** "Switchboard" is a metaphor used knowingly. State this once, clearly, and the metaphor is then earned.

**The load-bearing interpretive claim (Tier C):** salience-network *flexibility* is the lever — not favouring reflection over doing, or the reverse.

### The contradiction this resolves

The obvious version of this theory says modern life starves us of default-mode activity and we need more of it. That version is wrong, and the evidence says so:

- Mind-wandering — the default mode's signature activity — predicted *lower* momentary happiness in Killingsworth & Gilbert's 2010 experience-sampling study, accounting for more variance in wellbeing than what people were actually doing.
- Rumination, the default mode's failure state, is one of the more reliable predictors of depression risk.
- The best experimental result in this whole literature (Bratman et al. 2015) is a finding that nature *reduced* self-referential neural activity — the good outcome came from less of it, not more.

So "reflect more" is not the prescription, and any version of this theory that lands there has misread its own evidence. The target is the capacity to switch appropriately, which is precisely what tier collapse degrades. This is the point where the theory stops resembling every other book about slowing down.

---

## Part 6 — What follows in practice
**Tier C, except where noted.**

**Retriage, not abstinence.** The practice is one question asked at the moment of contact: *is this within a day's walk?* Tier-three information isn't forbidden; it's re-filed. You may know about the war; you may not treat knowing as acting.

**The evidence supports this over quitting** (Tier B, and it favours you): review evidence on digital detox is poor quality overall, total abstinence has the weakest record of any approach, several studies find harm during the detox, and interventions that change *how* people engage outperform those that merely reduce exposure. Abstinence fails because it treats the problem as volume. It's structure.

**Restore the stripped markers.** Read distant news in text, in past tense, at a delay, in longer form. Form was always the filing system; you can reinstate it manually.

**Low-salience environments.** Tier A, and the strongest practical claim available: ninety minutes in a natural setting reduced rumination and subgenual prefrontal activation; ninety minutes in an urban setting did neither (Bratman et al. 2015). Same duration, same effort — the setting did the work. The strong version, that 5,000 steps somewhere quiet beat 10,000 in a corridor, is Tier C but sits on a solid foundation. A forest is one of the last environments most people enter where nothing is pretending to be urgent.

**Desk hunger.** A personal observation worth keeping: sustained focused work produces hunger with no exertion, while multi-day walking produces little despite heavy exertion. Chronic stress raising appetite for energy-dense food is Tier A (Adam & Epel, 2007, and the cortisol-reactivity literature). The specific interoceptive-suppression mechanism proposed here is Tier C — a guess, and should be written as one. It is good essay material and not a basis for advice.

---

## What was removed from v1, and why

- **The foreign-policy argument** (that task-positive-dominant societies bias toward interventionism). Unfalsifiable and politically loaded. One quotation from it would discredit the rest.
- **The weight-loss protocol.** A prescriptive four-anchor health regimen resting on an invented mechanism, aimed at readers who may include vulnerable people. "Weight regulation follows nervous-system safety, not willpower" is an unsupported clinical claim. The underlying observation survives as an essay; the protocol does not.
- **The diagnostic test's scoring.** Numeric thresholds derived from a single respondent create false clinical precision. The twenty questions survive as unscored reflection prompts.
- **The "I" / "me" labels.** They invert most readers' intuitions, add no explanatory power, and cost a paragraph of defence every time they appear.
- **The seasonal network mapping.** No evidence supports the specific assignment and Meyer et al. 2016 points elsewhere. It survives in Book Two as an explicitly declared design metaphor.

## What the theory still needs

1. **Any test at all of the tier model.** Even a small study — do people report different physiological and emotional responses to identical information framed as local versus distant? This is genuinely testable and nobody has done it.
2. **A worked account of the actionability gap** distinct from the existing compassion-fatigue literature, which covers adjacent ground from a clinical angle.
3. **Case studies** for the workplace and seasonal claims. Currently argument without instances.

---

## Where the theory sits relative to existing work

Ideas here with real academic lineage, to cite rather than re-derive: cognitive miser theory (Fiske & Taylor); predictive processing (Friston, Clark); System 1/2 (Kahneman); media ecology (McLuhan, Postman, Carr); the narrative-noise critique (Taleb); context collapse (boyd, Marwick, Turkle).

**Use with care:** Dunbar. The idea of a bounded social world is usable; the number is not. Lindenfors et al. 2021 re-derived it with modern methods and got confidence intervals spanning 2 to 520, meaning no specific figure is defensible.

The genuinely novel piece is the three-tier relevance model and the actionability gap that follows from it. Everything else is assembled from existing work — which is fine, and worth saying openly, because the assembly is the contribution.