Private workspace first
Manual reference portfolios and watchlists stay behind authentication for research and decision journaling.
AI investment research systems
An AI-assisted investment research and agent-evaluation system.
MockStock combines private portfolio research, shared evidence packets, model-assisted agents, deterministic policy, risk controls, a sandbox broker, and operational observability. This site presents the architecture and selected historical simulation results.
Arena as lab
The four-agent arena remains an internal benchmark for comparing model behavior under equal inputs.
Selected results
The historical demo shows a sanitized record of the Arena without exposing current holdings or operations.
System boundary
The model is not the trader.
Evidence Packet
Quotes, events, news, and context are frozen before analysis.
LLM Signal
Models return structured research or arena signals, not trades.
Agent Policy
Deterministic policy translates signals into bounded actions.
Risk Boundary
Rules enforce sizing, confidence, cash, and lifecycle checks.
Mock Broker
Only the simulated broker can fill sandbox orders.
Audit Trail
Queue, cost, accounting, and outcomes remain traceable.
Private by default
Reference portfolios, watchlists, thesis notes, and decisions are private workspace data.
Public pages do not depend on the current live arena, current holdings, current trades, raw market packets, raw model responses, operational job rows, or private portfolio records. Selected examples must be published as sanitized historical artifacts.
Built to evaluate
Arena results are useful because inputs, policies, costs, and accounting are traceable.
The internal benchmark lab compares four provider-backed agents under equal packets. The private committee mode can use different research mandates for useful analytical disagreement without confusing provider identity with strategy.