We are building the next generation of investing: decentralized, systematized, and democratized.
LEARN MOREOpen simulation ↗ · Historical simulation with parameters selected using past results. Fees and funding included; market impact excluded. Not live performance.
Models
Train on our datasets or bring your own. Submit predictions to an open challenge, compete against the world's best ML experts, measure true out-of-sample performance, and simulate or trade the meta-model as a live portfolio on Hyperliquid.
# our data, your own, or both: every model scored out of sample
import crowdcent_challenge as cc
import polars as pl
client = cc.ChallengeClient("hyperliquid-ranking")
client.download_inference_data("current", "live.parquet")
preds = model.predict(pl.read_parquet("live.parquet"))
client.submit_predictions(df=preds)# the meta-model as a portfolio: fees, funding, liquidity
result = client.run_simulation(
config={"n_long": 20, "n_short": 20},
benchmark_trials=25, # vs. 25 random-ranking books
)
result["oos_stats"]["sharpe"] # the held-out stretch# the same book, live on Hyperliquid, from your own account
plan = client.preview_rebalance(network="mainnet")
plan["trades"], plan["turnover"], plan["est_fees"]
client.execute_rebalance( # only with the preview's hash
plan["plan_hash"], network="mainnet"
)# every call above, driven by your agent in plain English
$ claude mcp add --transport http crowdcent \
https://mcp.crowdcent.com/mcp \
--header "Authorization: Bearer $CROWDCENT_API_KEY"
> backtest 20 long / 20 short, then submit my modelEach point is a feature extraction from a single write-up; similar text clusters in three independent 3D UMAPs. Source: synthetic data
Select a point to read the corresponding extraction.
Research
We extract meaning from investment write-ups (catalysts, risks, and competitive advantages) and combine it with financial data, so an analyst's thesis and opinion become data that a model can learn from.
Analysts contribute through our partner communities, including SumZero.
Analysts share investment research across our network of partner communities. CrowdCent transforms these unstructured insights into data for a global community of data scientists to build predictive models. The result is a meta-model that captures the best of both human and machine intelligence, directing capital toward its highest-conviction views. The Simulator opens the same signal to anyone: sweep parameters, blend sleeves, and stress-test strategies against out-of-sample holdouts on real Hyperliquid data.