PAPER LIVE DATA MOCK AI

Glossary & FAQ

A quick reference for the terms on screen and the questions that come up first.

Glossary

TermMeaning
Paper tradingSimulated trading against an in-memory broker. No real money.
Pipeline / runOne pass of a symbol through the 12 stages.
StageOne step of the pipeline (e.g. risk sizing, the approval gate).
LONGA trade betting the price will rise (buy now, sell higher).
SHORTA trade betting the price will fall (sell now, buy back lower).
EntryThe price the thesis wants to open the trade at.
TargetThe take-profit price.
StopThe cut-the-loss price.
HorizonHow many trading sessions the thesis expects to hold.
ConvictionThe debate's confidence in the trade, 0–100%.
StanceAn analyst's lean, −1.00 (bearish) to +1.00 (bullish).
ConfidenceHow sure a single analyst is, 0–100%.
NotionalPosition size in rupees: quantity × entry.
Risk amountWhat you lose if the stop is hit — the trade's true risk.
Realized P&LProfit/loss locked in from closed trades.
Unrealized P&LPaper profit/loss on still-open positions.
MarkThe current price a position is valued at.
ExposureHow much of the account is committed to positions.
MandateThe human-set rules: allowed symbols and risk budget (stage 4).
UniverseThe list of symbols the mandate allows.
Kill switchEmergency stop: flattens positions, blocks new orders (stage 10).
SlippageThe small simulated gap between intended and filled price.
Trader IDThe identifier you enter when approving — recorded as the signer.
TokenThe secret that authorizes any action (approve, reject, kill, save).
Feature bundleThe deterministic evidence computed before the agents run: real indicators, news features, fundamentals, and a forecast. The agents reason over these numbers.
ATRAverage True Range — a measure of how much the price typically moves. Used to size the stop and target.
ATR anchorThe deterministic rule that sets the trade's prices from ATR: stop ≈ 2×ATR from entry, target at a 2R multiple. The code sets these, not the model.
AbstainWhen an analyst has no real data (e.g. no fundamentals source wired), it returns a low-confidence "abstaining" note instead of inventing a view — no AI call.
Hybrid memory / RRFThe default memory store: it blends a keyword (BM25) score with a dense vector and fuses them with Reciprocal Rank Fusion to surface genuinely similar past trades.
Price forecastA baseline expected move + band over the horizon, shown to the agents as evidence only — never used to size or route a trade.
Signal evolutionThe reviewer's verdict on how a thesis fared: Realized (win), Falsified (loss), Weakened (scratch), Strengthened, or Unknown — derived from the factual outcome.
FalsifierWhat evidence would flip the debate's chosen winner — surfaced so you know what to watch for.
Conviction calibrationThe manager's conviction is automatically knocked down when the analyst panel disagreed with the winning side.

FAQ

I clicked Approve and got "401 Authentication token required".

You did not set the token. Type it into the Token box in the header, click Set, then retry. See Getting started. The default on a fresh local setup is change-me.

I ran a symbol but it never appeared in the approval queue.

It was rejected at the mandate gate (stage 4) because the symbol is not in your universe. Add it on the Strategy screen and run again.

Is any of this real money?

No. The PAPER badge is always on and live trading is not wired in. Every fill is simulated. See the Safety model.

Why is the AI giving the same answer every time?

Offline mode uses a deterministic mock provider, so identical input gives identical output. That is intentional for a reliable sandbox. Real models only engage once an API key is configured and offline mode is off (a separate, human-owned step).

Can I trade a smaller size than the AI suggested?

Yes. On the approval screen use Modify Quantity instead of Approve, set your share count, confirm, and submit.

How do I stop everything immediately?

Open Portfolio, scroll to the red Emergency Kill Switch card, enter your Trader ID, confirm, and engage it. Every position is flattened and new orders are blocked until you reset.

What happens after a trade closes?

Stage 12 runs automatically: a reviewer agent critiques the outcome against the thesis, classifies how the signal evolved (Realized / Falsified / Weakened from the factual win/loss/scratch), and saves a structured, tagged lesson to memory, which informs future runs. The realized P&L and outcome are factual — the AI only writes the critique and lessons. You do nothing.

Where do I change which model the AI uses?

On the LLM settings screen — per role. Note offline mode forces the mock provider regardless.


Still stuck on a screen? Re-read The core workflow — it maps every screen to the step of the pipeline it belongs to.