STOCK AGENT

Account operations

STOCK
AGENT
Awaiting data
OPERATIONS

Overview

Connecting
RUNNER
ACCOUNT--
ORDER ELIGIBILITY--
LATEST TICK--
Portfolio equity--Awaiting account snapshot
Available cash--GBP
Universe assessed--Awaiting research
Estimated monthly AWS--Excluding VAT

Portfolio value

GBP
No daily snapshots yet

System health

Worker tick
--
Heartbeat alarm
--
Research
--
Next eligible batch
--
Notification email
--
Fee verification
--
Data permissions
--
EODHD daily calls
--
Feed limit alert
--

Research coverage

--
------
------

Practice-account order gates

    Instrument blockers

    Practice account reset

    Fixed risk guardrails

    Applied to all strategy orders
    Maximum loss at stop
    --
    Maximum position
    --
    Concurrent positions
    --
    Daily-loss pause
    --
    Peak drawdown pause
    --
    Maximum holding period
    --

    Instrument assessments

    --
    InstrumentProviderCurrencyStatusReason
    --

    Open positions

    --
    TickerQuantityValue GBPUnrealised P/L GBP

    Pending orders

    TickerSideQuantityStatusCreated
    This week--Benchmarks exceeded
    This month--Benchmarks exceeded
    Since start--Benchmarks exceeded

    Benchmark comparison

    BenchmarkProxy returnPortfolio returnExcess return
    Latest batch--No completed metrics yet
    Wall time--Completed batch duration
    Process CPU----
    Peak memory--Peak resident set size
    Provider calls----

    Batch history

    Metrics appear after a research batch completes

    Wall and CPU time

    Provider request time

    Peak memory

    Provider calls

    Latest provider activity

    ProviderCallsWall timeCPU timeErrorsThrottles

    Latest research stages

    StageCallsWall timeCPU timeErrors

    Latest cache activity

    CacheHitsMissesHit rate
    Strategies running--Awaiting the strategy lab
    Practice-account slot----
    Promotion target--Net return over at least 20 trading days
    Model spend this month----

    Strategy equity

    GBP, before model and AWS costs
    No strategy equity yet

    Leaderboard

    Current stage · net of trading, model and AWS costs
    StrategyStageTrading daysNet returnMSCI WorldDrawdownCosts / monthModel spendForecast scoreBrier score; lower is better, 0.25 is a coin flip.Next step

    Rejected strategies

    Most recent 20 · the author learns from these reasons
    StrategyThesisRejectedReason
    PLAIN-LANGUAGE GUIDE

    How the research and decisions fit together

    Stock Agent checks which instruments have trustworthy data, lets a model write and run strategies on them, and judges every strategy only on results that happen after it was created. Deterministic safeguards sit between the model and any order. Results are compared with buy-and-hold MSCI World; past results are not a promise of future returns.

    01 / RESEARCH

    From instrument list to assessment

    The universe starts with instruments reported by the broker. Each one is checked separately. “Validated” means the required identity and research evidence passed; it does not mean the instrument is approved to buy.

    01Instrument list

    What the broker says is available.

    02Verify identity

    Match the exact listing, ISIN, type, and currency.

    03Check evidence

    Review price history, FX, costs, and required event data.

    04 / DECISIONChecks pass?

    Every required check must have acceptable evidence.

    YESValidated instrument

    Joins the list of instruments the strategy lab may research and trade.

    NOBlocked with a reason

    Missing or conflicting evidence is not guessed or treated as zero cost.

    02 / STRATEGY LAB

    Model-run strategies, judged on forward results

    A model writes and runs its own strategies, and only deterministic guardrails sit between its decisions and any order. Strategies are judged on results that happen after they are created, because a model may already know how past markets moved; back-testing a model would flatter it.

    01 / AUTHORWrite a strategy brief

    A stronger model writes a brief: thesis, style, focus, holding period (up to 56 days), review frequency, maximum positions and which models to use. It sees running strategies and why earlier ones were rejected, and must be different.

    02 / SHADOWTrade a virtual £2,000

    Three shadow slots trade on live prices with the real cost model. On weekdays, while the exchange session named in its brief is open (Europe, US, or Europe open to US close), the trader researches with read-only tools and places orders, each with a thesis, expected return, probability, horizon and exit condition. A deterministic screen checks prior-high breakouts, relative volume and optional EODHD market-cap bounds; earnings surprises are not available and are never assumed.

    03 / DAILY REVIEWReject or promote?

    Code reviews every strategy once a day. Failing strategies are rejected and their slot is refilled; the best qualifying shadow strategy moves to the practice-account slot.

    04 / PRACTICE ACCOUNTPass again on a fresh £2,000

    A promoted strategy starts a new ledger so the second test is independent. When authorised, its accepted orders are copied to the Trading 212 practice account, scaled to that account's value.

    05 / LIVE CANDIDATEYour decision

    A strategy that passes again becomes a live candidate. Its practice-account holdings are closed and the practice-account slot is freed for the next shadow strategy. The separate Live runner follows the candidate with real money only after you approve and authorise Live; it is never automatic.

    Rejection rules (after 10 trading days)

    Drawdown above 8%; trailing MSCI World by more than 5 percentage points; trading and model costs above 1.5% of equity a month; more than half of at least 10 orders refused by the guardrails (refusals only because a market was closed are not counted); or poorly calibrated forecasts (Brier score above 0.30 over at least 10 resolved forecasts).

    Promotion rules

    At least 20 trading days in the stage; a net return of at least 2% after trading, model and AWS costs; beating MSCI World by at least 1 percentage point; drawdown below 5%; at least 5 completed trades; and best on the leaderboard. The 2% target rises by 0.25 points for each rejected strategy, up to 4%, so trying many strategies does not make a lucky pass likely.

    What the guardrails enforce

    Every buy, virtual or real, passes the same deterministic checks: a verified instrument and cost model, a fresh quote, an open market, a stop no more than 25% below the price, fixed limits of 1% risk per trade, 20% per position and five positions, a 2% cash buffer, a monthly trading-cost budget of 1% of equity, earnings safety for stocks, no leveraged or inverse products, no near-duplicate holdings, and pauses at 4% daily loss or 15% peak drawdown. Stops can only be raised. Exits are never blocked.

    Model costs and review diagnostics

    Each model call's token usage is priced from Bedrock rates, converted at the live USD/GBP rate and charged to that strategy's results. Each strategy has £3 a month; the lab stops at £9 a month in total. Reviews distinguish completion from turn/time limits and provider errors. Transient failures and incomplete reviews get one same-day retry after 15 minutes, limited to four model turns and still subject to the monthly allowance. The strategy card shows sanitized tool names, tickers, durations and outcome categories; raw provider data is not stored there.

    The Overview displays the fixed guardrail values. Percentages are of account value; a pause stops new purchases until reviewed, and exits continue. The Strategies view shows the leaderboard, equity against buy-and-hold MSCI World (run forward from the lab start), each strategy's brief, positions, notes, orders and decisions, and the rejected list. Practice-account orders require demo_execution_authorized on the Demo runner; until then, the practice-account stage is tracked virtually.

    Three isolated runners share this deployment, each with its own configuration, credentials, database, schedule and heartbeat. Observe watches the practice account read-only and never orders. Demo trades the practice account and runs the strategy lab. Live copies the live candidate's portfolio to the real-money account, one order per tick and checked again by the guardrails; it stays idle until Live is configured through the protected approval and live_execution_authorized is set. Use the runner selector above the Overview to switch views.

    03 / BENCHMARKS

    How comparisons are calculated

    Benchmarks are reference investments used to put portfolio returns in context. The dashboard compares returns over matching time periods; it does not treat a missing benchmark as a win or a loss.

    01Match dates

    Use the same portfolio snapshot boundaries.

    02Measure portfolio

    Use account performance, including actual broker costs.

    03Measure proxies

    Use adjusted ETF prices converted to GBP with aligned FX.

    04Compare returns

    Portfolio return minus proxy return is excess return.

    SPY, ONEQ, DIA, ISF.L, 1320.T, and URTH are ETF proxies for broad markets, not exact index returns. An unavailable proxy is not counted as beaten. Strategies are judged against URTH (MSCI World). Fund fees, tracking differences, market hours, and data revisions can affect the comparison.

    04 / GLOSSARY

    Terms used across the dashboard

    Universe
    The set of instruments the broker reports for the account.
    Assessment
    Checks that each instrument is correctly identified and has the evidence needed for research.
    Validated
    Passed the required identity and data checks. It is not, by itself, a buy recommendation.
    Blocked
    A required check failed or evidence is missing. The dashboard gives the reason; the system does not fill gaps by guessing.
    History
    Past daily prices and related data, such as currency conversion and corporate actions.
    Cost evidence
    A reviewed rule for applicable transaction charges or taxes. Without it, an instrument can remain blocked.
    Risk profile
    A set of limits for trade sizing, number of positions, holding time, and loss pauses.
    Drawdown
    The decline in portfolio value from a previous high point.
    Correlation
    How closely two instruments’ daily returns move together; above 0.9 they are treated as the same bet.
    Slippage
    The assumed gap between the quoted price and the actual fill price.
    Leveraged or inverse product
    A fund that multiplies or reverses daily index moves. Excluded because value decays over time and prices can gap.
    Benchmark proxy
    An investable fund used to approximate a market index; it will not match the index exactly.
    Excess return
    Portfolio return minus the benchmark proxy’s return for the same period.
    Shadow mode
    Decisions are recorded and traded only in a virtual account; no broker orders are placed.
    Strategy lab
    The process that writes, runs, reviews, rejects and promotes model-authored strategies.
    Shadow strategy
    A model-authored strategy trading a virtual £2,000 on live prices and real costs.
    Practice-account slot
    The single stage whose orders may be copied to the Trading 212 practice account once authorised.
    Live candidate
    A strategy that passed in shadow and again in the practice account. The Live runner follows it only once Live is approved and authorised.
    Runner
    One isolated worker process (Observe, Demo or Live) with its own account credentials, configuration, database and schedule.
    Forward test
    Judging a strategy only on market moves after it was created, so hindsight cannot help it.
    Promotion target
    The net return a strategy must reach in its stage; it rises as more strategies are rejected.
    Brier score
    How well stated probabilities matched outcomes: 0 is perfect, 0.25 matches always saying 50%, and higher is worse.
    Stock AgentAwaiting sources