docs(copy): drop hard-number anchors in framework framing, use 'swarms' instead

Author: jddunn

docs/ARCHITECTURE.md

@@ -283,7 +283,7 @@ In Paracosm, HEXACO influences:
 
 ### Parallel Execution
 
-Pair runs (exactly 2 commanders) fan out via `Promise.all` in `pair-runner.ts`. Cohort runs (3 to 300 commanders) fan out through a bounded worker pool in `runBatchSimulations`, sized to `economics.batch.maxConcurrency` (default 8) so a 300-actor run lands as ~38 batches while staying within provider rate limits. Within each commander's turn, all department analyses also run in parallel. This produces independent timelines from the same starting conditions:
+Pair runs (exactly 2 commanders) fan out via `Promise.all` in `pair-runner.ts`. Cohort runs (3+ commanders) fan out through a bounded worker pool in `runBatchSimulations`, sized to `economics.batch.maxConcurrency` (default 8) so the swarm lands as a sequence of batches that stay within provider rate limits regardless of how many leaders are in flight. Within each commander's turn, all department analyses also run in parallel. This produces independent timelines from the same starting conditions:
 
 ```
 Turn N:

src/cli/sim-config.ts

@@ -520,13 +520,14 @@ export function applyDemoCaps(config: NormalizedSimulationConfig): NormalizedSim
 export function normalizeSimulationConfig(input: SimulationSetupPayload): NormalizedSimulationConfig {
   // Fork setup takes exactly one actor (the override for the forked
   // branch). Regular pair setup takes exactly two and dispatches to
-  // runPairSimulations with verdict comparison. Cohort setups (3 to
-  // 300 actors) dispatch to runBatchSimulations and skip the verdict
+  // runPairSimulations with verdict comparison. Swarm setups (3+
+  // actors) dispatch to runBatchSimulations and skip the verdict
   // because verdicts compare exactly two leaders and would be
-  // ambiguous across N >= 3. The cohort upper bound matches the
-  // Quickstart UI's clamp and the economics-profile design target;
-  // batches of `economics.batch.maxConcurrency` (default 8) keep
-  // provider rate-limit pressure bounded at any scale in the range.
+  // ambiguous across N >= 3. The upper bound is a defense-in-depth
+  // sanity check matching the Quickstart UI's clamp and the
+  // economics-profile design target; batches of
+  // `economics.batch.maxConcurrency` (default 8) keep provider
+  // rate-limit pressure bounded at any swarm size.
   if (input.forkFrom) {
     if (!Array.isArray(input.actors) || input.actors.length !== 1) {
       throw new Error('Fork setup requires exactly one actor');

src/dashboard/src/components/about/AboutPage.tsx

@@ -304,8 +304,8 @@ export function AboutPage() {
             <div className={styles.hostedLabel}>Open Core + Hosted Roadmap</div>
             <h3 id="hosted-heading" className={styles.hostedTitle}>Planned Hosted Packaging</h3>
             <p className={styles.hostedPara}>
-              The open-source engine supports unlimited leaders and simulations via the API today. The dashboard demo runs up to 300 leaders
-              side-by-side, auto-switching to a constellation view past 50. The planned hosted product targets organizations that need to run dozens or hundreds of simulations in parallel.
+              The open-source engine supports unlimited leaders and simulations via the API today. The dashboard runs swarms of leaders
+              side-by-side, auto-switching to a constellation view as the cohort grows. The planned hosted product targets organizations that need to run swarms of simulations in parallel.
             </p>
             <p className={styles.hostedPara}>
               Defense agencies stress-testing doctrine across leadership profiles. Corporations modeling executive decision-making