Council, society & simulation

A tour of the swarm's surfaces. Navigation is the pet rail down the left edge — one animal per surface, and the animal is the only thing on screen that identifies it: the bunny is Forecast, the dog Council Reactions, the hamster Society Pulse, the turtle Readback, the chick Simulation, and the cat — sitting slightly apart — the Canon. The active pet grows a step and wears its name; the rest step back to icons. Every pet is alive: it breathes, blinks and watches your cursor.

Forecast

The WMTR (Wealth = Material × Time × Relational) survival projection under shocks, shown as four panels: wealth trajectory (25–75 band + mean), survival probability S(t), the outcome distribution, and the M/T/R components.

From here you can Forecast & convene to run a council on the projection.

The projection models a community: people, and the money, time and relationships that carry them. A scenario that isn't one (a fund's allocation, an insurer's capital decision, a reserving estimate or a pricing model's fit) gets no forecast. The tab says so, and the council judges the scenario as stated: trust means it holds up on what is given, distrust means something it states is wrong, and uncertain means there isn't enough to judge. There are no parameter interventions on such a run, since there is no forecast to adjust.

Council reactions

The heart of the deliberation:

• A force graph of agents (Finance, Investor, Accountant, Actuary, Psychologist, Lawyer, …), coloured and bordered by their final stance.
• A readback Sankey: profession → trust the forecast (or the scenario)? → confidence band.
• A decision sidebar — click an agent to see its per-round reasoning, key risk, and final vote — or open an audit interview with them; click a profession to see the group's aggregate. The panel icon in its header collapses it to a slim handle carrying the same icon and its name; click the handle to bring it back.

Recommended interventions

When an agent recommends changing a model parameter, it's shown as a tidy card (e.g. ↑ α_M · large with a one-line rationale), not raw JSON. On the strip you can apply & re-simulate a consensus intervention to see how the forecast shifts.

Society pulse

A broader simulated society's reaction to the same scenario, with configurable society parameters (income mix, education mix, employment mix, culture). A cluster's member list opens individual members — each of whom you can interview about the sentiment they recorded.

Audit interviews

Trust, but interview. On Council reactions and Society pulse you can put any member back on the record and converse with them about the position they recorded — watching whether they stay consistent with it.

• Pick a member. 🎲 in the header band picks one at random; shuffle in the drawer moves on to the next; or choose someone specific from the agent inspector (council) or a cluster's member list (society).
• The drawer shows the record. The member's card carries who they are and exactly what they recorded: all three rounds with confidences, the final vote, key risk and recommended intervention — and, for professionals, the justification they wrote, with its framework, citations and formulas.
• Every reply is checked. The member answers from inside the original brief — the same persona, IAAI canon and WMTR evidence they deliberated with — and each reply is verified against the recorded verdict and badged consistent, drift or unverified, with a running tally. Ready-made probes test the position from different angles.
• It is all audit trail. Transcripts persist to the swarm's chat log, and GET /api/run/:id/interviews indexes who has been interviewed, with their consistency counts.

The recorded verdict stands — an interview never rewrites a vote. Professionals defend their position with the theory on their record; society members answer as ordinary people, in plain language.

Simulation

The Simulation surface: scenario chips, drugs/compounds, sample size and population, then the macro impact tiles. The animated illustration needs JavaScript.

A standalone population simulator:

1. Pick a scenario (a medical or social shock) — or paste your own.
2. Set the drugs / compounds, sample size, and population.
3. Run simulation. A progress panel walks the pipeline: Resolving compound references → Sampling the population → Simulating agent outcomes → Scaling macro impact.
4. Results: macro impact tiles (workdays lost, GDP drag, excess mortality, severe/critical, hospital admissions/cost, insurer claims, out-of-pocket), treatment-uptake bars, and distributional tables by age and comorbidity.

What the progress panel is, and isn't

/api/simulate is a single request, so the four phases advance on a fixed cadence rather than reporting server state. Only the elapsed timer is live. The phases tell you what the pipeline does, not how far along it is.

Drug references are resolved live via PubChem, OpenFDA, and ChEMBL.

IAAI Canon

The IAAI Canon editor: reference works with import for JSON or BibTeX. The animated illustration needs JavaScript.

The reference works (title + takeaway) that get injected into every agent's system prompt under "IAAI Canon — apply where relevant". Add, edit, or import works (JSON or BibTeX). If empty, agents are told to say so — no fabricated citations.

A fresh install starts with an empty canon. The sample button fills the box with a worked example for whichever format is selected, so you can see the shape before importing your own.

State persists across surfaces

Switching surfaces keeps your scenario, slider values, and results — you resume exactly where you left off.