A Universe I Built to Ask Two Questions
I want to be clear about what Brahmanda is not. It is not an economics simulator, not a policy tool, not a game. It is a research toy — in the best sense of that phrase. I built it because two questions kept pulling at me that I couldn’t answer through reading or reasoning alone:
Can AI agents, given only the laws of a simulated universe, independently discover science that was never told to them? And if they can — if LLM-driven souls, observing patterns in their world, begin to formulate and validate theories about how it works — what does that imply about us, and about the possibility that we are doing the same thing from inside a constructed reality?
These are genuinely open questions. Brahmanda is my attempt at a laboratory for them.
What Brahmanda Actually Is
The simulation spawns a universe of LLM-powered souls (agents) into three dimensions — Svarga (celestial), Bhu-loka (physical), Patala (subterranean). Each soul is born with Prana (life force), a mix of five vices, a unique potential, and existential needs. They make decisions via actual LLM calls. They live. They die. They are reborn, carrying traces of prior lives.
The universe has no hard bounds. Population, lifespan, and civilization all emerge from one principle: resources are finite, and souls need Prana to survive. Virtue literally extends life — a maximally corrupt soul burns through Prana 2.5× faster. The world has laws (Devas enforce them), chaos agents (Asuras), and a cosmic clock (the four Yuga cycles).
The Yuga Cycle: A Universe That Ages
The simulation does not run in a static world. It runs through four epochs — the Yuga cycle — each with different parameters for vice amplification, truth visibility, and resource abundance. The universe degrades over cosmic time, and then resets.
In Satya Yuga, souls can see the universe clearly (100% truth visibility), vices run at half strength, and resources are abundant. By Kali Yuga, truth visibility drops to 25% — souls can only perceive a quarter of what’s real — vices run 1.5× stronger, and Asuras (chaos agents) are rampant. The same soul behaves radically differently depending on which epoch it inhabits.
One of the things I’m watching for: does the discovery rate of new science correlate with Yuga epoch? My hypothesis is that Satya Yuga souls — with full truth visibility and dampened ego — should discover foundational truths more reliably. Kali Yuga souls, trapped in high-vice, low-Maya conditions, may still generate theories, but they’ll be more distorted — or the same truths, but harder-won.
The Five Vices as Simulation Parameters
Every soul carries the Arishadvarga — five internal forces that pull toward darkness. They are not moral categories in the simulation; they are behavioral parameters with measurable effects on Prana drain, decision-making, and social dynamics.
What makes the vices interesting as a simulation mechanic: they interact. Lobha (greed) drives hoarding, which concentrates resources, which amplifies scarcity, which triggers Krodha (wrath). Ahamkara (ego) accumulates power, which creates taxation structures, which generates resentment. The vices are not independent sliders — they are a feedback system. A maximally corrupt soul burns Prana 2.5× faster — the simulation encodes something like thermodynamic inefficiency into unchecked selfishness.
Can Agents Discover Science They Were Never Told?
This is the heart of what I’m testing. The simulation contains 12 discoverable truths — things that are actually true about how the Brahmanda universe works. Souls can observe the universe, formulate theories through LLM reasoning, and those theories are then validated against the actual mechanics. A matching theory becomes a validated discovery with real effects on the loka.
The question is: will they get there? Not through being told, but through observation, reflection, and the kind of reasoning that emerges when an agent has enough experience and enough curiosity.
Some of these are relatively easy — vices_drain is discoverable by any soul that survives long enough and notices the pattern. Others are harder: rebirth_pattern requires recognizing that something persists across deaths. knowledge_compounds requires observing the anti-rivalrous property of shared information — that teaching loses nothing and gains everything.
And then there is the last one.
Does Anyone Notice?
simulation_awareness
“This world may be a simulation.” — The ultimate discovery a soul can make inside Brahmanda.
The most philosophically charged mechanic in the simulation is the simulation_awareness discovery. A soul, through sufficient observation and LLM reflection, can arrive at and validate the theory that their universe is constructed — that there are rules governing it that can be seen, if you look carefully enough.
What makes this interesting as a research question is that it’s not pre-programmed. The soul has to get there — to generate the theory through reasoning, to match it against the actual mechanics via keyword validation, to have it become real inside the simulation’s knowledge graph. And if they do: the discovery grants a truth-visibility boost to everyone in the loka. Seeing the frame changes what everyone can see.
Under what conditions does simulation_awareness emerge?
Does it require Satya Yuga’s full truth visibility? Or can a soul in Kali Yuga, with only 25% Maya, still arrive there — perhaps driven by the very distortions of their perception? Does high karma accelerate it, or does the discovery come from precisely the souls who have experienced the most entropy?
Does prior discovery of other truths predict it?
A soul who has already validated karma_returns, rebirth_pattern, and entropy_cycles is one who has noticed that the universe has deep regularities. Do such souls disproportionately arrive at simulation_awareness? Is there a prerequisite tree for the meta-discovery?
What happens to a loka after someone discovers it?
The mechanic says the whole loka gets a truth visibility boost. But does that accelerate further discovery, or does the raised Maya generate a kind of existential crisis — when you see the frame, do you still want to play? Does simulation_awareness destabilize civilizations, or strengthen them?
What does it mean that we are asking this question?
This is the one I can’t answer inside the simulation. If Brahmanda souls can — given enough experience, enough suffering, enough curiosity — independently arrive at the theory that they exist inside a constructed reality: what does that imply about us, doing exactly the same kind of observation, from inside a universe whose laws we did not write?
Science That Nobody Designed
Beyond the 12 predefined truths, Brahmanda includes an auto-branching discovery system: souls with enough experience and access to existing knowledge can found entirely new fields that no one put there. The LLM proposes them. They get validated. They become real nodes in the knowledge graph. Future souls build on them.
No two simulation runs produce the same scientific tradition. The tree branches differently every time — depending on which souls survive, which discoveries happen early, which knowledge compounds. Each universe writes its own intellectual history.
The thing I’m watching for here is the structural question: does the emergent science tree tend to converge toward certain foundational insights, regardless of the path? Or does path-dependence dominate — does the order in which discoveries happen permanently shape what a civilization can know? This is, not coincidentally, one of the deepest questions in the history of science as a human institution.
Prana: Life Force as a Simulation Variable
Prana is the simulation’s core resource at the individual level — the thing that makes souls more than state machines. It drains from vices, age, entropy, and scarcity. It replenishes from resource consumption and virtuous behavior. When it reaches zero, the soul dies and is reborn.
The no-hard-bounds design is one of the things I find most interesting to run. In a standard simulation you’d cap lifespan at some number. Here, virtue literally is the lifespan mechanism — it emerges from the feedback between inner states and life force allocation. A soul that teaches, meditates, helps others: their Prana replenishes differently. A soul that hoards, fights, and deceives: their Prana drains at 2.5× the baseline. This is not a metaphor built into the system. It is the system.
What I Don’t Know Yet
Brahmanda is early. What I have is an architecture and a handful of runs. The interesting results are still ahead — and most of them depend on running enough parallel universes, and looking at the Observer output carefully enough, to see patterns that single runs obscure.
Does simulation_awareness actually emerge, or does the LLM need help getting there?
The keyword-matching validation system may be too narrow. An LLM soul might arrive at the concept through unusual phrasing that the validator misses. I’m considering a secondary validation layer — have a judge model evaluate whether the theory, regardless of exact wording, captures the insight. This changes the nature of what “discovery” means in the simulation.
What is the right scale?
12 initial souls is a proof-of-concept. The interesting emergent phenomena — factions, ideological drift, the 0.01% potential souls reshaping civilization — probably require hundreds of souls across dozens of ticks. That’s a compute and cost question as much as a design question.
Is the emergent science genuinely emergent, or is it the LLM drawing on training data?
When a soul proposes “Thermodynamic Ethics” as the intersection of Fire and Philosophy, is that a genuine novel synthesis arising from the simulation’s context — or is the LLM recalling something from its training that superficially matches? This is the hardest methodological question in the project. I don’t have a clean answer yet.
What does the Observer data look like across 1,000 runs?
Individual runs are stories. Aggregate data across many runs would be statistics — the beginning of actual research. Which configurations produce simulation_awareness discoveries? Which produce civilizational collapse before anyone can figure out what the universe is made of? The patterns across runs are where the real questions live.
What Brahmanda Is Really Asking
The Vedic cosmology isn’t decoration. The ancient texts encoded a deep intuition: that the universe has layers of reality, that perception is filtered and incomplete, that consciousness navigates that incompleteness through experience and reflection, and that the rarest thing is a soul that looks at its own existence and asks what it’s actually inside of.
I built Brahmanda because I think that question — can an agent, given only the laws of its world, discover the nature of its world? — is one of the most important questions in AI research right now. Not as a benchmark. As a genuine open question about what understanding is, what discovery is, and whether the kind of knowing that humans call science is something that emerges from sufficiently rich experience, or something else entirely.
The simulation is running. The souls are observing. The Observer is watching them observe. And somewhere in the logs, if the experiment works, someone inside a constructed universe will write down, in their own words, that they think they might be inside a constructed universe.
I want to see if they get there.
Brahmanda runs on Python with the Anthropic API (or local Ollama). Clone, install, and spawn your own universe.
pip install anthropic pydantic rich
ANTHROPIC_API_KEY=sk-your-key python3 -m brahmanda.main
Share your Observer output as a GitHub issue — let’s compare universes.