Are the volatility and pricing-error improvements real?

They are paired Monte Carlo results with significance and effect sizes in the table; efficiency ties while dispersion and RMSE fall, so read them as institution-plus-cognition, not magic.

If efficiency ties, why bother with cognitive traders?

Because the takeaway is stability: the auction alone nears 97% efficiency, while Jev agents compress drift and form a tighter terminal corridor for price discovery.

What does it cost to run the cognitive side?

About 15 input tokens per decision with outputs cost-free, plus zero-token pre-filtering for submarginal valuations and delta caching when book depth is unchanged.

Can I explore results without keys or installs?

Yes. The econometrics notebook is pre-rendered with plots for viewing on GitHub, and Codespaces builds an isolated env forwarding port 8000 for the web app.

How do I run the cda-abm-ai-agents dashboard?

pip install -r requirements.txt then python web_app.py, and open localhost:8000 for the dark-mode dashboard showing trades, order-book depth and supply and demand curves; you can paste a TypeSafe key into the header without saving it.

What trader archetypes do the Jev agents choose between?

Per order-book state a Jev agent samples between Aggressive Taker, Queue Competitor, Passive Maker and Deep Discount using Score and Choice primitives, while ZI-C agents draw uniform random bids and asks with no memory.

Does the benchmark need an API key to reproduce?

Running the full benchmark.py needs an API key and compute, but the econometrics notebook is pre-rendered with high-resolution plots for viewing on GitHub, and Codespaces builds an environment that forwards port 8000.

CommunityJev Agents & MCP

Cda Abm AI Agents

Mesa market testbed pitting zero-intelligence traders against Jev archetype agents, with paired Monte Carlo stats, a live dashboard, and a pre-rendered econometrics notebook.

See highlights

Open github.com

Stars: 0
Language: Jupyter Notebook
License: MIT
Published by: Community
Author: JiriNohejl
Last verified: Oct 2026

JiriNohejl/cda-abm-ai-agents

User repository on GitHub

View on GitHub

Replicating Gode & Sunder (1993) Continuous Double Auction with TypeSafe AI Jev agents. Compares canonical ZI-C heuristics against stochastic choice AI Jev brains. Results led to compressing price variance by 43% while matching 97% allocative efficiency. Includes real-time web interface & Jupyter Notebook econometrics.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Across paired Monte Carlo market simulations under identical Poisson arrival distributions and supply/demand fundamentals: Launch the lightweight dark-mode dashboard to watch trades, order book depth, and Walrasian supply/demand curves in real time: Open in your browser. (Optional: You can paste your TypeSafe API key directly into the web UI header during runtime without saving it to disk.)

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Tool calling

Used inJev for Software & DevTools

Highlights

Reports efficiency tie near 97% but tighter price discovery with lower dispersion and RMSE.
Jev agents pick Aggressive Taker, Queue Competitor, Passive Maker, or Deep Discount per book state.
Runs lean at about 15 input tokens per decision with pre-filtering and order-book delta caching.
Ships 28 tests, paired benchmark script, stored database outputs, web app, and slide deck.

Watch out

Research replication with paired simulations; re-running benchmarks needs an API key and compute. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-mcp

MCP server that exposes Jev classify, score, yes-no check, risk gate, and multi-question decide tools to coding agents through one Jev call per tool.