The core workflow
This is the most important page. It walks the end-to-end loop once, slowly, and tells you which screen you are on at each moment and why.
The loop at a glance
Runs -> (AI does stages 1-7) -> Approvals -> YOU decide -> Portfolio -> Review
trigger automatic waiting for sign / position & AI critique
a symbol research a human modify / live P&L saved to
reject memory
You only ever act in two places: triggering a run, and deciding at the approval gate. Everything else is the system showing you its work.
Step 1 — Trigger a run (Runs screen)
Go to Runs, type a symbol (e.g. RELIANCE), and click Run Agents.
Behind that one click, the system runs stages 1 through 7:
- Pulls the market snapshot, news, and macro data — and computes a deterministic
feature bundle (real indicators, news features, a price forecast).
- Sends it to four analyst agents in parallel (news, sentiment, technical,
fundamental), each reasoning over the computed numbers. Analysts with no real data source (sentiment, fundamental in this build) abstain rather than guess.
- Checks your memory store for relevant past trades.
- Confirms the symbol is inside your strategy mandate.
- Runs the bull-vs-bear debate — including one rebuttal round where each side
answers the other — then a manager scores a conviction (calibrated down if the analyst panel disagreed).
- Drafts a trade thesis (direction, rationale, risks), with entry/target/stop
derived from deterministic ATR/level anchors rather than invented.
- Sizes the position with deterministic risk code.
Then it stops at stage 8 and drops the result into the approval queue. You are redirected to Approvals.
If the symbol is not in your mandate's universe, the run is rejected at stage 4 and never reaches the queue. See the Strategy mandate page.
Step 2 — Review the proposal (Approval detail)
On Approvals, click Review on the pending row. This is the screen the whole system exists to produce. Top to bottom it shows:
- Order — the concrete proposal: symbol, direction (LONG/SHORT), quantity,
entry, target, stop, notional, risk amount, horizon, conviction.
- Risk decision — the deterministic verdict, with a checklist of every cap that
was tested (✓ passed / ✗ failed).
- Analyst notes — what each of the four agents concluded, with a numeric stance.
- Research debate — the bull case, the bear case, the manager's rationale, and
the trader's rationale.
Read it as: the evidence, then the risk check, then the exact order. Every field is defined on the Runs & approval gate page.
Step 3 — Decide (the action bar)
At the bottom of the review screen you have three choices:
- Approve — enter your Trader ID, tick I confirm this order, click
Approve Order. This records your signature and runs stages 9–12.
- Modify quantity — same, but you override the share count first. Useful when
you agree with the trade but want a smaller size.
- Reject — enter your Trader ID and reject. The request is closed; nothing trades.
Your signature is what makes the order legal inside the system. Without it, the execution code physically refuses to run. This is the non-bypassable gate.
Step 4 — Watch the position (Portfolio)
Approving redirects you to Portfolio, where stages 9–11 have already run:
- The order was "filled" by the paper broker (with a little simulated slippage).
- A position appears in the table with its average entry and live mark.
- The summary strip shows cash, exposure, realized and unrealized P&L.
- Alerts flag things like a stop being hit or exposure getting large.
Step 5 — The learning loop (automatic)
When a position closes, stage 12 runs a reviewer agent that critiques the actual outcome against the original thesis and writes the lesson back into the memory store. That memory is what stage 3 reads on the next run — so the system is meant to get better-informed over time. You do not have to do anything for this to happen.
That is the whole tool. Everything else in this guide is detail on a single screen.