Testing & access
Named cloud agents
My agents is a separate private paper beta. Users name agents, choose bounded momentum or mean-reversion rules, and create new versions to test improvements. Chat proposes settings; it cannot place orders or silently change a run. Five-minute observations wait at least 90 seconds after a candle closes. Revised data is retained without rewriting earlier decisions. This page also documents the earlier three Windows experiments.
Open my agentsHow our agents are tested
A reproducible paper experiment, with a clear record of what is known and what is still unproven.
Three experiments. Separate virtual accounts.
The daily baseline uses a frozen statistical classifier. Short-term momentum and intraday breakout use fixed rules. The agents do not retrain themselves, use a chat model to place orders, or receive astrology inputs.
| Experiment | Decision data | Intended holding period | Starting funds |
|---|---|---|---|
| Daily baseline | Completed daily prices | Daily observations | $100 virtual |
| Short-term momentum | Daily momentum and market trend | Approximately 1–5 sessions | $100 virtual |
| Intraday breakout | Completed five-minute price and volume bars | Up to 60 minutes; scheduled close exit | $100 virtual |
The forward-testing sequence
1. Record the rules
Freeze the model version, risk limits and costs before collecting outcomes. Any material change requires a new experiment rather than rewriting history.
2. Observe completed data
Prices and volume come from Yahoo research data. Incomplete, stale, missing or conflicting observations are rejected; a gap may freeze the experiment.
3. Save a decision before its fill
A saved signal becomes eligible at a later completed bar or session. The first observation creates no trade. Repeating an observation cannot multiply fills.
4. Review the whole record
Examine return after modeled costs, drawdown, number of observations, closed trades, failure events and same-period benchmarks. A high win rate alone is not sufficient.
What the simulation leaves out
Daily baseline costs are 0.1% per side; the newer strategies model 0.15% per side. A completed-bar price is not a guaranteed executable quote. Delays, spreads, market impact, taxes, dividends and corporate actions can change live outcomes. Stop thresholds are observed triggers, not guaranteed loss caps.
Historical tests and forward observations are separate. Accounts with different dates or exposures cannot be fairly ranked by raw return. The current record does not establish profitability.
Who can see and manage the research?
| Role | Can view | Can manage |
|---|---|---|
| Visitor | Public methodology and policies | Nothing in private experiments |
| Registered member | Their available account tools; public methodology | Their own portfolio records |
| Invited researcher | Shared experiment results and historical evidence | View only; no grants or trading controls |
| Experiment owner | Results, evidence, health and access history | Grant/revoke research access; use local runner controls |
Research access is granted to a registered account by the owner. It does not expose another person's portfolio or birth profile. Revocation blocks future server reads; information already viewed or downloaded cannot be recalled.
Review before expanding
The owner reviews data quality and safety stops first, then performance across an adequate sample and different conditions. There is no automatic promotion to real money and no daily-return target that the system promises to meet. Connecting a brokerage would be a separate project with separate approval.