Simulate
Simulate tests a hypothesis before anything is proposed for approval. Open it with
Simulate in the bottom navigation (/simulate). The page has three tabs, in the
order you would use them:
| Tab | For | Link |
|---|---|---|
| Metasearch Trends | Searching the open web for a term, and checking text against the grounding rules. | /simulate?tab=trends |
| Causal Matrix | The cause-and-effect structure found by the last simulation. | /simulate?tab=causal |
| Swarm Intelligence | Running simulations and handing the results to Twins. This is the default tab. | /simulate |
Older /lab links redirect here.
Swarm Intelligence
The CONSENSUS BRIEF at the top summarises what the last simulation concluded and what would be handed to the twin queue. It shows:
- IMPACT SUMMARY, RECOMMENDATION and CONFIDENCE;
- SIMULATED IMPACTS, with metric, change, risk and horizon;
- DECISION LEVERS;
- a validation verdict.
The verdict reads PASS or FAIL, Validating…, The validation couldn’t run or Not validated. Validation starts by itself when a result arrives. You can also start it with Validate, or with Retry after a failure. No simulation yet means no simulation has run.
A switch chooses between two modes: Active Scenarios and New Simulation.
Active Scenarios
| Panel | Shows |
|---|---|
| SWARM POPULATION | One agent per inferred scenario, with the counts AGENTS, RUNNING, REPORTED, FAILED and STANDBY, and each agent’s state by domain. |
| AGENT CONVERGENCE | The nodes that more than one agent’s evidence named. |
| Domain Nodes, Synapses, Open Rules | Counts of each. |
| ACTIVE SIMULATIONS | Running scenarios with a progress bar. SIMULATION SCENARIOS below lists every scenario by domain. Tap a row to open that scenario’s detail page. |
| SAVINGS SUMMARY | The Total Identified Savings (USD) and a breakdown by category. |
The savings total tells four cases apart:
| You see | Meaning |
|---|---|
| Measuring… | Still loading. |
| Not available | The request failed. Nothing shown is a measurement. |
| Not measured | No summary has been requested yet. |
| No savings identified in the current action set | Measured, and the result was nothing. |
Could not load scenarios means the request failed. It shows the error and a Retry button. No scenarios yet means the list is genuinely empty.
Run a simulation
Live agent activity, at the bottom of the tab, streams every step the agents emit. It is the same feed as on Reconcile.
Scenario detail
Tapping a scenario row opens its detail page (/lab/scenario/<id>). The page shows
SCENARIO DETAILS, the REASONING (the rules and variables it used) and, once it
has run, its confidence, AI RECOMMENDATION and SIMULATED IMPACTS. A scenario
that has not run says No simulation results available yet. Please run the simulation
to see impacts and recommendations. Scenario could not be found. means the id is
unknown.
Metasearch Trends
- TERM SEARCH runs a live query against the sovereign metasearch engine. Type a term (Search a regulatory, market or operational term…) and select Search. Each result shows its host, its ENGINE SCORE (a ranking score, not a confidence), a snippet and Open source. No results means the engine ran and the web returned nothing. The search couldn’t run means it failed.
- GROUNDING GATE runs your text through the same OpenBias rules that guard the agent call. Nothing is sent and nothing is logged. Type the text and select Check this text. The verdict is Would be refused or No rule fired, with the number of rules evaluated. Show the {count} rules lists them.
- LONGITUDINAL TRENDS charts the forecast and anomalies from the last simulation. Until you run one, it says No series analysed.
Causal Matrix
The matrix is built from the evidence traces of the last simulation.
- CAUSAL TOPOLOGY shows which node moves which, and how strongly. SUB-DOMAIN CAUSAL MATRIX tabulates the same links by SOURCE, RELATION, TARGET and SIGNIFICANCE. No causal evidence yet means no simulation has run.
- INTERVENTION sets one node and reads the effect on another. Pick Intervene on, enter Set to, pick Read effect on, and select Compute effect. The result shows EFFECT and NEW TARGET VALUE, and the note under it explains how the effect was computed.
- BAYESIAN POSTERIOR gives the posterior probabilities for a target node. Enter an observed state for any node, leave the rest blank, choose the TARGET, and select Infer posterior. The inference returned no probabilities means it ran and returned nothing.
RPCs behind this page
HypothesisService.GetHypothesisVariables,InteractiveSimulate(bidirectional stream),ValidateHypothesis,SendToTwins.AnalysisService.AnalyzeTimeSeries,AnalyzeDescriptiveStats,AnalyzeCausal,InferBayesian.MetasearchService.SearchTerms,EvaluateBias,ListBiasRules.TwinsService.GetSavingsSummary.ActivityService.SubscribeAgentActivity.
See the API reference.