Where a deal can exist
ABuyer and seller posteriors for one negotiation in the latest round. The hatched area is where both find the price plausible.
Agent beliefs
Mean posterior by behavioural type after the latest round. Same evidence, different beliefs.
Market uncertainty
Inject information
Applied at the start of the next round and stored in the run config.
What happened
Plain-language reading of the current run, followed by the mathematics with this run's numbers.
Posterior mean by behavioural type
CTransaction price over time
DAverage price per round with the min–max band. Dashed: the latent true value.
Trade volume and failed negotiations
FPosterior uncertainty by type
GAverage posterior SD σ. Narrow = confident.
Distribution of transaction prices
EProfit and loss per agent
ICash plus holdings valued at the true value. Box = middle 50%, bar = median.
Trade success rate by group
JShare of each group's negotiations that ended in a trade.
Posterior overlap vs probability of trade
KNegotiations from the last 10 rounds. Line: share accepted per overlap bin.
Trade probability: buyer type × seller type
LAccepted ÷ attempted negotiations, whole run.
Who trades with whom
MEdge width ∝ completed trades. Node size ∝ group size.
Conservative vs neutral vs aggressive
HBelief accuracy, calibration, trading and profit by group. 90% coverage = share of agents whose 90% credible interval contains θ_true (calibrated ≈ 90%).
All agents
Click a row to inspect the agent.
Controlled experiments
1–3 isolate one source of heterogeneity each. 4–10 run full markets. Market experiments can be opened in the dashboard to inspect every agent and trade.
Market replay
Step through the current run one round at a time.
Compare scenarios
Save runs as scenarios, or generate a matched pair that differs in one setting. Matched pairs share the seed.
Three sources of behavioural difference
Every label in this simulation reduces to explicit numbers in one of these layers. They can be switched on and studied separately.
Agent templates
Group mean ± between-agent spread for each parameter. Normal parameters: spread is an SD. Positive scale parameters (marked ×): spread is a coefficient of variation, value = mean·e^N(0,spread).
Market, information and trading
Model specification
Prior, likelihood, posterior, utility and decision rule exactly as implemented.
Reproducibility
The full configuration of the current run: seed, agent counts, templates, information process, true value, matching, trade and pricing rules, shocks. The same JSON always reproduces the same run.