Summary

Simulation runs — described in WorldMonitor's product as AI-driven scenario forecasting — can only be triggered by a human operator running node scripts/process-simulation-tasks.mjs --once directly in the Railway environment. No HTTP endpoint exists. Agents and API consumers can read simulation outcomes but cannot initiate a simulation run. For a feature marketed as agent-native AI analysis, the AI cannot actually initiate the analysis.

The gap

enqueueSimulationTask(runId) and runSimulationWorker are exported from scripts/seed-forecasts.mjs and are callable from Node processes only. The HTTP API has no surface to trigger them.

Capability matrix:

Related: the runId filter in getSimulationOutcome is also documented as a no-op (todo #29), meaning even if an agent could trigger a run, it cannot verify its specific run completed.

Impact on the product claim

WorldMonitor's AI Market Implications panel and agent-native MCP tools are built around the premise that an AI agent can analyze a developing situation and run forward simulations. With no HTTP trigger, the workflow is:

1. Agent detects an event
2. Agent reads the latest (stale) simulation outcome
3. Agent cannot trigger a fresh simulation

The simulation system is read-only for every consumer except a human with Railway terminal access.

Proposed fix

Add POST /api/forecast/v1/trigger-simulation following the existing RPC handler pattern:

// server/worldmonitor/forecast/v1/trigger-simulation.ts
export default async function handler(ctx) {
  const isPremium = await isCallerPremium(ctx.request);
  if (!isPremium) return errorResponse('Premium required', 403);

  // Rate limit: 1 trigger per 5 minutes per identity
  const limiter = Ratelimit.slidingWindow(1, '5 m');
  const { success } = await limiter.limit(getCallerIdentity(ctx.request));
  if (!success) return errorResponse('Rate limited', 429);

  const latest = await redis.get(SIMULATION_PACKAGE_LATEST_KEY);
  if (!latest) return jsonResponse({ queued: false, reason: 'no_package' });

  const runId = (latest as any).runId ?? crypto.randomUUID();
  await enqueueSimulationTask(runId);
  return jsonResponse({ queued: true, runId });
}

The actual execution remains Railway-side (existing poll loop picks it up). This endpoint only enqueues the task and returns the runId for polling.