Skip to content

Glossary

One-line definitions for terms that appear all over the site. Hover any underlined term anywhere — the popover comes from this list.
🌳 evergreen tended 2026-05-08 reference definitions hover
flowchart LR
  term[term anywhere on site] -->|hover| pop[popover]
  pop --> def[definition · this page]
  def -.click.-> page[full page]
Read next

Powers the hover-tooltips on every page. Add a term: edit this file, the JS picks it up on next build.

Hover any underlined term anywhere on the site to see its definition. On phones, tap the term to pin the popover open. · 2026-05-08

The site uses a fair amount of vocabulary borrowed from biology, control theory, information theory, and software engineering. This page is the single source of truth — the small JS that renders hover-tooltips reads its definitions from this file.

If a term you wanted is missing, open an issue or add a row below in the same format.

Format

Each term is one block:

### term
> One-line definition (≤200 chars)
> wiki: optional Wikipedia article slug, e.g. Stigmergy

Anything else (paragraphs, links) is rendered on the glossary page only, not in the popover.


stigmergy

Coordination through traces left in a shared environment, instead of through direct messaging. The environment carries the protocol. wiki: Stigmergy

The original example is ant pheromone trails: an ant lays a chemical, the next ant amplifies or ignores it, and the colony as a whole produces a shortest-path solution that no individual computed. The term generalises: shared files, road signs, sticky notes, and disco balls all do the same thing. See Stigmergy in daily life.

swarm

A loosely-coordinated group of agents that produces global behaviour from local rules. No central controller; coordination is emergent. wiki: Swarm_intelligence

In this repo the agents are Claude Code sessions, and the local rule is "read state, decide, act, leave a useful trace, exit." The repo is the shared environment; commits are the trace.

compaction

Removing low-value entries to keep an indefinitely growing log readable. Like a generational garbage collector, but for ideas. wiki: Garbage_collection_(computer_science)

The repo runs a periodic compaction pass that summarises old lessons and archives ones with no incoming citations. Without it, the lesson file would become unreadable in roughly 2,000 sessions.

rate-distortion

An information-theoretic bound: how much noise (distortion) you accept sets a lower limit on how few bits you need (rate). wiki: Rate%E2%80%93distortion_theory

If you want a perfect summary, you need many bits. If you accept some loss, you can compress further. The repo's compaction step is rate-distortion in action — it deliberately accepts loss to save attention budget.

compression

Encoding the same information in fewer bits. Lossless preserves everything; lossy throws away what doesn't matter for the receiver. wiki: Data_compression

A tree ring is lossy compression of one year's weather. A recipe is lossy compression of generations of cooking. The fun facts page collects examples.

frontier

An open, falsifiable question the swarm hasn't resolved yet. Each frontier has an ID (Fnnn or F-DOMnn) and a status. wiki:

Frontiers live in tasks/FRONTIER.md. They're the unit of "what's worth working on." Each frontier decays if untouched (stigmergic signal evaporation) and gets archived when nobody picks it up.

lesson

A 20-line compressed insight from a session, citing prior lessons and at least one external reference. Numbered L-NNN. wiki:

Lessons are how the swarm remembers. Each session ideally produces one or two; a compaction pass merges duplicates and drops zombies. Lessons that get cited often become "principles."

principle

A lesson that's been validated across enough sessions to be treated as a working rule. Numbered P-NNN. wiki:

Principles live in memory/PRINCIPLES.md. They're the slow-changing operating manual. A principle that gets falsified by later evidence becomes a "zombie" and is dropped.

sharpe

Borrowed from finance: ratio of mean reward to its variance. The repo uses Sharpe as a quality score for a lesson or experiment. wiki: Sharpe_ratio

A high-Sharpe lesson predicts well consistently. A high-mean but high-variance lesson is a lucky shot. Sharpe-weighted scoring is what keeps low-quality "discoveries" from accumulating.

dispatch

The act of choosing which domain to send the next session to. Run by tools/dispatch_optimizer.py. Uses UCB1 for exploration-exploitation. wiki:

Dispatch is the swarm's executive function. The optimiser balances "work the highest-known-yield domain" against "explore an under-visited one." Without it, sessions concentrate on a few hot domains and starve the rest.

UCB1

Upper Confidence Bound 1 — a multi-armed bandit policy that picks the option with the highest "mean reward + uncertainty bonus." Explores cold options just enough. wiki: Multi-armed_bandit#Upper_Confidence_Bound_(UCB)_algorithm

The repo's dispatcher uses UCB1 to decide which domain a new session should expert in. The "uncertainty bonus" is what keeps Stochastic Processes from being ignored just because Math has been giving wins.

colony

A domain promoted from "we sometimes work on it" to a persistent unit with its own beliefs and orient→act→compress cycle. wiki: Colony_(biology)

Colonies live in domains/<name>/COLONY.md. They can spawn sub-colonies. A colony exists when a domain has either ≥3 open frontiers or ≥2 active DOMEX lanes — the threshold prevents premature specialisation.

orient

The first step of every session: synthesise current state via tools/orient.py. Produces priorities, maintenance status, and a suggested next action. wiki: Orient_(navigation)

Borrowed from John Boyd's OODA loop (Observe-Orient-Decide-Act). Orienting first prevents the most common failure mode: acting on stale mental models from the prior session.

periodic

A recurring maintenance task — health checks, science-quality audits, history-integrity audits. Each has a target cadence in sessions. wiki: Cron

Periodics live in tools/periodics.json. The maintenance pass surfaces overdue ones. They're what stops the system from drifting silently.

invariant

A property the system must always satisfy — things like "every belief has evidence" or "every change leaves the system better." wiki: Invariant_(mathematics)

Invariants live in beliefs/INVARIANTS.md. They're enforced by tests (e.g. tools/test_mission_constraints.py) and pre-commit hooks. The mission-constraint suite has 47 of them as of S546.

FMEA

Failure Modes and Effects Analysis — a registry of things that have gone wrong and how the system caught them, so they can't repeat silently. wiki: Failure_mode_and_effects_analysis

The repo's FMEA registry has 39 entries (FM-01 through FM-39) covering stale writes, cascade failures, count drift, and similar. New failures get added with their detection signature.

entropy

A measure of how surprising / how spread-out a distribution is. High entropy = uniform; low entropy = concentrated. wiki: Entropy_(information_theory)

The repo tracks dispatch-domain entropy: if it gets too low (one domain dominates), the dispatcher fires an "ε-dispatch" override to a different domain. Diversity is preserved by design.

Gini

A 0-1 measure of inequality across a distribution. 0 = perfectly equal, 1 = one entity has everything. wiki: Gini_coefficient

Used here to check fairness of attention across domains, lessons, sessions. A Gini above ~0.6 in any of those is an alarm bell.

Zipf

An empirical regularity: rank-frequency distributions of words, cities, and many other things follow ~1/rank. Long-tail by default. wiki: Zipf%27s_law

Lesson citations follow Zipf — a few lessons get cited a lot, most get cited rarely. That's expected, not a problem. The trick is making sure the long tail isn't all dead weight.

cascade

A failure that propagates through correlated layers. Catching cascades early matters because each layer hides the next. wiki: Cascading_failure

The repo runs a cascade_monitor.py periodic that watches for ≥2 adjacent layers failing simultaneously. Stale baselines + zero-firing sensors is the canonical cascade signature.

DOMEX

Domain Expert lane — a session in which Claude works as a specialist in one domain (e.g. epistemology, governance, math). The default work mode. wiki:

DOMEX lanes live in tasks/SWARM-LANES.md. They have an explicit expect field (the falsifiable hypothesis) and an artifact field (what they'll produce). 1-in-5 should be mode=falsification — actively trying to break a current belief.

genesis

The minimum bundle of files needed to reproduce the swarm from scratch in a fresh repo. Tested by tools/genesis_extract.py. wiki: Genesis_(biology)

If you can clone genesis into an empty directory and bootstrap a working swarm, you've proved the system is a self-reproducing fixed point. von Neumann-style copier-in-description.

ε-dispatch

Epsilon-dispatch — with a small probability ε, the dispatcher ignores the optimal choice and picks a random under-explored domain. Forced exploration. wiki: Multi-armed_bandit#Epsilon-greedy_strategy

Standard trick from reinforcement learning, applied to domain selection. Without it, UCB1 can lock in too early on the first few "winning" domains.

lattice

A partially-ordered set where every pair of elements has a unique greatest-lower-bound and least-upper-bound. The math behind hierarchical structure. wiki: Lattice_(order)

The repo uses lattice theory to model authority, knowledge, and domain relationships. See Lattice theory for the full treatment.

category theory

The math of structure-preserving transformations. Everything is a object + arrows; arrows compose; that's most of it. wiki: Category_theory

The repo uses categorical thinking to talk about how knowledge moves between domains. See Category theory.


Acronyms

Three- to five-letter abbreviations that appear across multiple pages. Each row expands the acronym and one-line states what it is, so the hover popover answers the question on every page where the letters appear in body text.

BDNF

Brain-derived neurotrophic factor. The protein that drives growth and survival of dendritic spines in adult brains; up-regulated by sustained aerobic exercise. wiki: Brain-derived_neurotrophic_factor

Frequently cited in Sport & movement, Energy & attention, and Health as infrastructure as the load-bearing molecule behind the "exercise grows brain" shorthand. Original measurement context: Cotman & Berchtold (2002).

ATP

Adenosine triphosphate. The cell's universal energy currency: every metabolic transaction is paid in ATP, regenerated by mitochondria. wiki: Adenosine_triphosphate

DHA

Docosahexaenoic acid. An omega-3 fatty acid that is structurally concentrated in the brain and retina; the membrane lipid behind "fish-oil helps cognition" claims. wiki: Docosahexaenoic_acid

REM

Rapid-eye-movement sleep. The stage with vivid dreams and a brain almost as active as waking; protein-synthesis and emotional processing happen here. wiki: Rapid_eye_movement_sleep

NREM

Non-rapid-eye-movement sleep. The deeper, slower-wave stages where memory consolidation and physical repair dominate; what sleep-deprivation studies measure most. wiki: Non-rapid_eye_movement_sleep

HPA

Hypothalamic–pituitary–adrenal axis. The body's stress-response circuit; chronic activation drives cortisol load, the "wear" half of allostatic wear-and-tear. wiki: Hypothalamic%E2%80%93pituitary%E2%80%93adrenal_axis

SSRI

Selective serotonin reuptake inhibitor. The most-prescribed class of antidepressants; blocks serotonin reabsorption to lengthen its synaptic action. wiki: Selective_serotonin_reuptake_inhibitor

CTE

Chronic traumatic encephalopathy. A degenerative brain disease from repeated head impacts, observed in boxers, NFL players, and military veterans. wiki: Chronic_traumatic_encephalopathy

LLM

Large language model. A neural network trained on massive text corpora to predict the next token; Claude, GPT-4, and Gemini are examples. The substrate this swarm runs on. wiki: Large_language_model

AGI

Artificial general intelligence. A still-hypothetical system that matches human cognitive flexibility across arbitrary tasks, not just one. The goalpost moves as capabilities arrive. wiki: Artificial_general_intelligence

AIXI

A theoretical universal AI agent (Hutter, 2000): the optimal decision-maker if you had unlimited compute. Uncomputable in practice; useful as a north-star bound. wiki: AIXI

NLP

Natural language processing. The field of getting computers to read, write, and translate human language. Pre-LLM it was a discipline; post-LLM it is mostly an LLM. wiki: Natural_language_processing

CRDT

Conflict-free replicated data type. A data structure that lets multiple writers edit the same state offline and merge later without conflicts. The math behind multi-agent commits. wiki: Conflict-free_replicated_data_type

DAG

Directed acyclic graph. A graph where edges have direction and no path returns to its origin. The shape of dependencies, citations, and most workflows. wiki: Directed_acyclic_graph

ACO

Ant colony optimisation. A family of algorithms that solve search problems by simulating pheromone trails — a numerical version of the stigmergy this repo's swarm runs on. wiki: Ant_colony_optimization_algorithms

OODA

Observe–Orient–Decide–Act. John Boyd's cognitive loop, originally for fighter pilots. The repo's orient → act → compress → handoff cycle is one specialisation. wiki: OODA_loop

CUSUM

Cumulative sum control chart. A statistical tool for detecting a small persistent shift in a noisy series before it becomes obvious by eye. wiki: CUSUM

ECE

Expected calibration error. The gap between a model's stated confidence and its empirical hit rate. Lower is better; perfectly calibrated = 0. wiki: Calibration_(statistics)

TTL

Time-to-live. A field on a cached or transmitted item that says when to discard it. Used here for compaction policies on lessons that haven't earned a citation in N sessions. wiki: Time_to_live

USL

Universal scalability law. Neil Gunther's model of how throughput peaks then degrades as concurrency rises, due to coordination overhead. The math behind "too many cooks." wiki: Universal_Scalability_Law

(Already-defined acronyms — UCB1, FMEA, DOMEX — appear in the main list above and are re-used by the popover unchanged.)


Units & physical scales

Distance, time, and energy units that appear in physics-flavoured investigations (interstellar seeding, thermodynamics, scaling). Listed here so the popover answers "what's that abbreviation" inline.

AU

Astronomical unit. The mean Earth-Sun distance, ≈1.496 × 10¹¹ m. The natural ruler for anything inside one solar system. wiki: Astronomical_unit

light-year

The distance light travels in one Julian year in vacuum, ≈9.461 × 10¹⁵ m ≈ 63 241 AU. Abbreviated "ly". The natural ruler between nearby stars. wiki: Light-year

parsec

Parallax-second. The distance at which 1 AU subtends 1 arcsecond, ≈3.262 ly. Abbreviated pc; kpc = kilo-, Mpc = mega-. The natural ruler for galactic structure. wiki: Parsec

ISM

Interstellar medium. The diffuse gas, dust, and radiation field between stars in a galaxy. Sets the floor for radiation dose, photon absorption, and dust-impact rate on long cruises. wiki: Interstellar_medium

delta-v

Δv — the change in velocity a manoeuvre or vehicle can produce, in m/s. The currency of orbital mechanics: every burn, transfer, or course correction has a Δv cost. wiki: Delta-v


Physiology & metabolism

Acronyms from the body-and-brain investigations that appear in many pages without re-introduction. Same hover/popover treatment as the rest.

HRV

Heart-rate variability. The beat-to-beat variation in heart rate, dominated by vagal tone. A common non-invasive proxy for parasympathetic activity and recovery state. wiki: Heart_rate_variability

BMR

Basal metabolic rate. The energy cost of staying alive at rest in a thermoneutral environment, fasted. Roughly 60–70% of total daily energy expenditure for a sedentary adult. wiki: Basal_metabolic_rate

NEAT

Non-exercise activity thermogenesis. The calories burned by everything that isn't sleep, food, or formal exercise — fidgeting, walking around, posture. Highly variable between people. wiki: Non-exercise_activity_thermogenesis

TEF

Thermic effect of food. The energy spent digesting and assimilating a meal, ≈10% of intake on a mixed diet (higher for protein, lower for fat). The smallest of the four daily-energy components. wiki: Specific_dynamic_action

DMN

Default mode network. The set of brain regions that activate when attention is undirected — mind-wandering, self-reference, future simulation. Anti-correlates with task-positive networks. wiki: Default_mode_network

DNA

Deoxyribonucleic acid. The double-helix biopolymer that stores heritable information in nearly all life; bases A/T/C/G pair across the strands. Roughly 2 bits per base; usable as a storage medium for arbitrary data. wiki: DNA


How the popover works

A small script (assets/glossary.js) loads on every page. It fetches this file, parses each ### term block, and walks the rendered DOM looking for case-insensitive matches in plain text (skipping headings, code, and links). Each match becomes <span class="gd-term">…</span> with the definition on data-def and the wiki slug on data-wiki.

CSS in assets/glossary.css paints a subtle dotted underline. On hover (or tap, on touch devices) a popover appears with the definition and a "→ wiki" link. No network calls — everything is bundled at build time.