Nomos

The chronicle is what makes Nomos different from a typed-agent ABM with stats panels. As the simulation runs, the browser detects significant events and routes each one to whichever theorist's lens has the most concrete purchase on it. One Mistral call per event produces one short paragraph in that theorist's voice. Across a run you hear every voice; no single moment gets buried under ten parallel takes.

The ten theorists

Each observer has a name, era, lens, sees, and watches description (in OBSERVER_INFO in lib/config.ts). The sees shapes their worldview; the watches tells the LLM what to actively look for. Both feed into the system prompt for the per-event call.

theorist	lens	what they uniquely surface
Marx	class, surplus, consciousness	Surplus extraction, alienation, contradictions. Private IOUs as new chains; the commons devoured by accumulation.
Polanyi	embedded economy, great transformation	When trade disembeds from kinship into impersonal price. Fictitious commodity stress. Promises hardening into circulating money — and, when confidence goes, dissolving in a run.
Bourdieu	capital, field, habitus	How habitus reproduces across generations even as trait drift moves agents. Capital converting between forms. Symbolic violence.
Granovetter	embeddedness, weak ties	Brokerage between clusters. Embedded trust shielding from predation. An issuer's notes travelling through the network until strangers accept. Trust centrality as leadership without a role.
Schelling	thresholds, segregation	Tipping points. Spatial sorting. The moment when one issuer's notes tip from a one-off favour into circulating money.
Turchin	elite overproduction	Structural-demographic preconditions of crisis. Soil's carrying capacity declining beneath population. Financialisation as late-cycle marker.
Farmer	complexity economics	Emergent price formation. Endogenous money creation. Discount priced by default risk. Positive feedback concentrating wealth.
Epstein	generative agent-based modelling	"If you didn't grow it, you didn't explain it." Money issued by agents, classes that cluster out of trait variance, commons worn thin from rational harvest.
Flack	slow variables, policing	What stabilises a system across time. Consensus on issuer trust as a slow variable. Policing through shame. What fails when slow variables erode.
Axelrod	evolution of cooperation, tit-for-tat	The shadow of the future. Whether retaliation against defectors is swift and visible. An issuer's reputation letting their promises circulate. Predation as the test the cooperative strategy was built for, not collapse.

Adding Axelrod was deliberate. Without him every coercion event read as decline (Marx and Flack's natural framing). With him the same predation-and-sanction data can read as cooperation being defended, not collapsing — same event, opposite interpretation. That disagreement is the point.

How significant events are detected

The detector lives in lib/events.ts. Every tick, the latest snapshot is compared against a roughly 8-turn history window and a set of latch flags. The full event roster:

event	trigger	notes
`founding`	turn = 0	once per run
`inequality_surge`	Gini delta ≥ +0.05 over the window
`leveling`	Gini delta ≤ −0.05
`stratification`	Gini crossing up through 0.5	qualitative threshold
`population_crash`	alive ≤ 75% of window-ago
`population_boom`	alive ≥ 135% of window-ago AND delta ≥ +15
`market_forming`	first time trade volume ≥ 12 per turn	once per run
`price_shock`	price ratio 1.6× or 0.625× over window	requires active market
`collapse`	alive ≤ 18% of run peak
`segregation`	clustering index crossing 0.18, re-arm at 0.12
`motivation_shift`	one motivation gains ≥ 5% share AND now ≥ 40%	dominance check
`coercion_wave`	seizures this turn ≥ max(3, 0.4% of alive)	60-turn per-kind cooldown
`cooperation_thickens`	≥ 1 issuer with ≥ 3 holders OR ≥ 50% of attacks sanctioned	60-turn per-kind cooldown
`network_fracture`	isolate share rising ≥ 15% to ≥ 40%	trade web dissolving
`extreme_inequality`	Gini ≥ 0.6 for 80+ consecutive turns	sustained-state
`oligarchy`	top-decile wealth share ≥ 80% for 80+ turns	sustained-state
`shock_blight`	engine fires blight	rate = f(land degradation)
`shock_plague`	engine fires plague	rate = f(density)
`leadership_emerges`	top agent's inbound tie weight ≥ 24	latched; re-arms at 14
`bank_run`	mistrust in top issuer crosses 0.35	60-turn cooldown
`passage`	30 turns elapsed without other events	heartbeat, max 3 in a row

Detection uses hysteresis latches for sustained-state events (Gini holds, oligarchy holds) so they fire once when the regime locks in, not on every cooldown.

How routing works

Each event kind has a priority list of theorists in lib/observer-routing.ts:

coercion_wave:        ["axelrod", "marx", "flack"]
cooperation_thickens: ["axelrod", "granovetter", "flack", "epstein"]
market_forming:       ["polanyi", "farmer", "granovetter"]
stratification:       ["bourdieu", "marx", "turchin"]
segregation:          ["schelling", "bourdieu", "granovetter"]
leadership_emerges:   ["granovetter", "flack", "bourdieu"]
bank_run:             ["polanyi", "farmer", "marx"]

When an event fires, pickObserver(kind, available) walks the priority list starting at a per-kind rotation offset. The first available theorist (i.e. one the user selected at setup) wins; the offset advances so the next firing of the same kind reaches for a different voice. This is why a run with eight active observers produces visible rotation across coercion_wave events instead of Marx narrating every one.

Pacing — making it readable

The chronicle has four layered throttles so it stays legible regardless of simulation speed:

Global cooldown — 12 turns minimum between any two events.
Per-kind cooldowns — coercion_wave and cooperation_thickens each have a 60-turn floor. Without these, loud recurring events dominated the chronicle.
Hysteresis latches — extreme_inequality, oligarchy, segregation each fire once when the regime locks in, and re-arm only when the underlying metric relaxes back below a lower band.
Wall-clock floor — MIN_NARRATION_INTERVAL_MS = 12000 in observer-narrator.tsx. Narrations are skipped if dispatch would land less than 12 seconds after the previous one. Skipped events re-detect next tick, so nothing important is lost as long as the condition persists. This is the gate that keeps the chronicle readable at 4× and 8× speeds.

The prompt and the call

When an event fires:

pickObserver(event.kind, activeObservers) selects the theorist.
buildSystemPrompt(observer) constructs the persona system message — name, lens, sees, watches, plus formatting rules (2–3 sentences, present tense, no jargon without self-evident meaning, no preamble).
buildUserPrompt(event, world, context) builds the user message — a one-line setting, the current motivation mix, the trade-tie snapshot, recent earlier events, and the factual event summary.
The browser POSTs to app/api/observe, which calls Mistral (mistral-small).
The returned paragraph streams into the chronicle.

If MISTRAL_API_KEY is not set, the simulation still runs; the observers stay silent.

Adding a new observer

Three places to touch:

lib/config.ts — add to the ObserverKey union, add an entry to OBSERVER_INFO with the four required fields.
lib/observer-routing.ts — add the new observer to whichever event-kind priority lists make sense (and to passage for heartbeat rotation).
lib/config.ts defaults — optionally add to DEFAULT_CONFIG.observers. (Existing users' saved configs won't auto-include them; they'd toggle on in setup.)

The UI is driven by Object.keys(OBSERVER_INFO), so the new theorist appears in setup automatically.

Adding a new event kind

Five places in lib/events.ts and one in lib/observer-routing.ts:

Extend the EventKind union.
Add a TITLES[kind] short label.
Add detection logic in detectEvent (with appropriate hysteresis or cooldown).
Optionally extend EventMetrics with new fields.
Add a summarize(kind, m) case producing the factual paragraph the theorist reads.
Add a PRIORITY[kind] list in observer-routing.ts.

The bench script is the fastest way to verify the new event fires at the rate you intended without spinning up the dev server.