You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
UI
- Title is now the city switcher: a glass column-select (Pixelify titles); the
active city shows in the title button, clicking opens the dropdown of cities.
- "whole city" button moved directly under the title (no more overlap).
- News bubble: same opacity as the rest of the chrome; click to expand the full
set of articles (date · headline · summary) that are informing the residents.
- Loading bar spans the status bubble width, sits underneath it with spacing.
- Bigger, more readable chrome; emojis removed throughout.
- About card: notes the five cities now supported + a build credit.
Per-resident LLM chatter (replaces hardcoded SF phrases)
- New POST /branches/:bid/chatter: one short, in-character present-tense thought
per requested resident, batched into a single Sonnet call (Engine::chatter).
- Frontend requests chatter only for the sprites currently on screen and caches
it, so calls stay sparse. Falls back to a city-neutral local pool while it
loads — fixing SF chatter ("Karl the Fog", Muni, the bay) leaking into other
cities.
<!-- simulation status: city · residents · knowledge-clock (top-right) -->
@@ -89,6 +96,7 @@
89
96
<divclass="about-h">The pitch</div>
90
97
<p>What if you could predict how your customers responded to releases before they went out into the real world? Or if you could get a signal on who will win an election?</p>
91
98
<p>We simulated the population of San Francisco distributed across real US Census data and achieved near-parity with these types of real-world events.</p>
99
+
<p>It now spans <b>five US cities</b> — San Francisco plus <b>neu york</b> (New York City), <b>synth la</b> (Los Angeles), <b>cybercago</b> (Chicago), and <b>simami</b> (Miami) — each its own Census-sampled population. Switch between them from the title in the top-left.</p>
92
100
93
101
<divclass="about-h">Historically predictive</div>
94
102
<p>How do we know if our simulation of SF is accurate? The honest test needs a model whose knowledge cutoff <i>predates</i> the event, so it can't have just memorized the outcome. The older Claude models with a 2023 cutoff have since been <b>retired</b> — so for this leakage-free backtest we ran <b>GPT-4o</b> (knowledge cutoff <b>October 2023</b>) and asked the twin city to predict results it had never seen. (The live app you're using runs the current <b>Claude Sonnet 4.6</b>.)</p>
@@ -108,6 +116,10 @@
108
116
<divclass="about-h">The future</div>
109
117
<p>All of decision-making is predicated on our understanding of causality. How will the choices we make influence the world? How will people react?</p>
110
118
<p>To enable (first) SF, and then the world, to understand the causal nature of reality, we present <spanclass="about-name">sim francisco</span>.</p>
119
+
120
+
<divclass="about-credit">
121
+
Built by <b>Tejas Prabhune</b> and <b>Tanmayi Dasari</b> · 2nd place, <b>Claude Build Day</b> (June 13, 2026).
122
+
</div>
111
123
</div>
112
124
113
125
<!-- character inspector (tap a character when zoomed in) -->
0 commit comments