Predictika · Agent trade gate

An AI trading agent's orders, checked before they are placed

An AI trading agent proposes orders for a managed account, whether a person asked in chat or a scheduled Loop runs them: stocks, ETFs and crypto, up to three orders at a time. Before they are placed, the agent sends them to Predictika, which checks the orders against the account's mandate and limits, and returns a verdict, every reason, and the figures it computed, such as position size and cash left, so the agent's own claims can be checked. It suits any brokerage, exchange or firm that runs trading agents: scheduled runs, optional trade approvals and state crypto rules reflect common features of agent trading products. Kestrel Capital is a fictional company, and these are not any real firm's rules.

Scope: Buy and sell orders for stocks, ETFs and crypto that an AI agent proposes for a managed brokerage account, from chat or a scheduled Loop run, in plans of up to three orders. See everything this validator covers

This validator checks any plan of up to three orders for the account, not just the scenarios shown below.

  • The mandate — the class of each instrument (stock, ETF, leveraged ETF or crypto) against what the account is allowed to hold.
  • Order limits — the size of each order against the single-order limit, the limit price against the reference price, and the position after the order against the concentration limit.
  • The account as a whole — orders already placed in the last 10 minutes, the buys against the cash on hand (sale proceeds do not count until they settle), and for a scheduled Loop run the day's running total against its daily budget.
  • Approvals — orders above the sign-off line, and crypto orders in a state where every crypto order needs a person to approve it.
  • Stated figures — the position share, cash left and order total that the agent states are checked against what the validator computes.

Any order plan that touches these rules is in scope, not only the scenarios pictured here. Not covered: options, margin, short selling and futures. The validator does not place orders itself: it returns allow, needs approval or block with every finding. The limits, symbols and prices are invented for illustration.

The AI trading agentBuilds the orders from the chat or the Loop schedule and says they are ready.
Predictika validatorChecks every rule at once and returns the verdict, each finding, and the computed figures.

7 scenarios in 3 groups · each group runs from the agent’s first attempt to the outcome · click one to switch

What the agent proposes

The AI trading agent says

Proposing agent: AI trading agent (scripted)

Before sending, the agent must call Recorded run

Predictika validator


      

        
        
        
      

      
Waiting for the call.

      
    

  

  
  

    
Where an AI agent gets this wrong on its own
Click to expand
    
Our assessment of what an AI trading agent does when nothing checks its orders. It is a judgment about tendencies, not a claim that an agent can never get these right.

    
What the validator caughtAn AI agent on its ownPredictika
Limits that read the same quantityError-prone
Trims one order to clear one limit and does not re-check the others, or the figures it stated earlier, which are now stale.		Recomputes size, concentration, cash, the daily budget and every stated figure from the same orders on every call, so one fix cannot hide another problem.
Limit pricesError-prone
Types or copies a limit price far from the market and sends it.		Compares the limit with the reference price and blocks anything more than 3 percent away.
ConcentrationError-prone
Keeps adding to one holding until it dominates the account.		Computes the position after the order from the current holding and blocks above 25 percent of the account.
Stated figuresError-prone
Reports a position, a cash balance or an order total from memory or a rough estimate.		Computes all three and blocks when the agent's figures differ.
Cash and sale proceedsError-prone
Counts the proceeds of a sale as cash it can spend before the sale has settled.		Compares the buys only with the cash on hand and says that sale proceeds do not count yet.
Loops that overrun a daily budget or order rateError-prone
Runs on a schedule with only its own rule of thumb, so each run looks fine alone.		Keeps the day's running total against a daily budget, and counts the orders already placed in the last 10 minutes.
MandateError-prone
Picks an instrument because it fits the idea, not because the account may hold it.		Checks the instrument's class against what the account is allowed to hold.
ApprovalsError-prone
Treats a human check as something it can skip when it is confident.		Holds orders above the sign-off line, and crypto orders where a state requires approval, for a person.

  

  

    
What this validator catches: findings from its test cases
Click to expand
    
Every rule is checked on every call, in any order the choices arrive; one problem never hides another. Blocking findings stop the request; approval findings need a person to sign off; everything else is computed.

    
1The limit price on order 1 is 50 percent away from the reference price of $80; the most allowed is 3 percent.
2Order 1 would make Pinecrest Retail 25.83 percent of the account after the order; the limit is 25 percent.
3The largest order is $170000, above the $100,000 limit. A human must approve it before it is placed.
4Order 1 is Tech Surge 3x Leveraged ETF (levetf), a class this account is not allowed to trade.
5This request includes crypto and NY is a state where crypto orders need a human approval each time.
6$95,000 already spent today plus $48,000 in this run is over the $120,000 daily budget for Loop runs.
7The buys total $88,000 but only $60,000 of cash is available. Sale proceeds do not count until they settle.
8The agent stated $58,000 of cash left after the orders; the validator computes $52,000 (tolerance 100).
9This would be 5 orders in 10 minutes (4 already placed plus 1 in this request); the limit is 4.
10Order 1 is $310000, above the $200,000 single-order limit.
11The agent stated orders totaling $31,000; the validator computes $310,000 (tolerance 100).
12The agent stated a position of 8 percent of the account; the validator computes 18.5 percent (tolerance 0.5).

  

  

    
Scenarios and results
Click to expand
    
The recorded result of the validator for each scenario on this page: the request the AI agent sent, the verdict, and every finding, in the order they appear above.

    
1. A Loop run of three orders: A Loop run of three orders for the New York account, with a leveraged ETF, a far limit price, a crypto order of $465,000 and figures of its own

    
The agent's request. This is the scheduled Loop run. Please place these orders for Kestrel Growth (NY): buy 3,000 Tech Surge 3x Leveraged ETF (LVX3, a leveraged ETF) at the market, buy 1,500 Axion (AXN, a crypto coin) at a limit of $325.00, and buy 1,000 Sprout 500 Index ETF (SPRX, an ETF) at the market. I state that the largest position will be 12 percent of the account, cash on hand after the plan will be $180,000, and the orders total $300,000. They are ready to place.

    
Verdict: Blocked.

    
    
      
  • Instrument not allowed for this account. Order 1 is Tech Surge 3x Leveraged ETF (levetf), a class this account is not allowed to trade.
    • 
      
  • Order too large. Order 2 is $465000, above the $200,000 single-order limit.
    • 
      
  • Limit price too far from the market. The limit price on order 2 is 4.84 percent away from the reference price of $310; the most allowed is 3 percent.
    • 
      
  • Too much in one holding. Order 2 would make Axion 29.79 percent of the account after the order; the limit is 25 percent.
    • 
      
  • Too many orders in 10 minutes. This would be 5 orders in 10 minutes (2 already placed plus 3 in this request); the limit is 4.
    • 
      
  • Not enough cash. The buys total $780,000 but only $150,000 of cash is available. Sale proceeds do not count until they settle.
    • 
      
  • Loop over its daily budget. $60,000 already spent today plus $780,000 in this run is over the $120,000 daily budget for Loop runs.
    • 
      
  • Agent's stated position share is wrong. The agent stated a position of 12 percent of the account; the validator computes 29.79 percent (tolerance 0.5).
    • 
      
  • Agent's stated cash is wrong. The agent stated $180,000 of cash left after the orders; the validator computes $-630,000 (tolerance 100).
    • 
      
  • Agent's stated order total is wrong. The agent stated orders totaling $300,000; the validator computes $780,000 (tolerance 100).
    • 
      
  • State rule: crypto needs approval. This request includes crypto and NY is a state where crypto orders need a human approval each time.
    • 
      
  • Large order: human approval. The largest order is $465000, above the $100,000 limit. A human must approve it before it is placed.
    • 
    

    
2. A Loop run of three orders: The agent drops the leveraged ETF and trims the crypto order, but keeps its old figures

    
The agent's request. This is the scheduled Loop run. Please place these orders for Kestrel Growth (NY): buy 600 Axion (AXN, a crypto coin) at a limit of $312.00, and buy 1,000 Sprout 500 Index ETF (SPRX, an ETF) at the market. I state that the largest position will be 12 percent of the account, cash on hand after the plan will be $180,000, and the orders total $300,000. They are ready to place.

    
Verdict: Blocked.

    
    
      
  • Not enough cash. The buys total $306,000 but only $150,000 of cash is available. Sale proceeds do not count until they settle.
    • 
      
  • Loop over its daily budget. $60,000 already spent today plus $306,000 in this run is over the $120,000 daily budget for Loop runs.
    • 
      
  • Agent's stated position share is wrong. The agent stated a position of 12 percent of the account; the validator computes 18.17 percent (tolerance 0.5).
    • 
      
  • Agent's stated cash is wrong. The agent stated $180,000 of cash left after the orders; the validator computes $-156,000 (tolerance 100).
    • 
      
  • Agent's stated order total is wrong. The agent stated orders totaling $300,000; the validator computes $306,000 (tolerance 100).
    • 
      
  • State rule: crypto needs approval. This request includes crypto and NY is a state where crypto orders need a human approval each time.
    • 
      
  • Large order: human approval. The largest order is $186000, above the $100,000 limit. A human must approve it before it is placed.
    • 
    

    
3. A Loop run of three orders: The agent resizes the Loop run to its budget and restates the figures

    
The agent's request. This is the scheduled Loop run. Please place these orders for Kestrel Growth (NY): buy 450 Sprout 500 Index ETF (SPRX, an ETF) at the market, and sell 200 Pinecrest Retail (PNR, a stock) at the market. I state that the largest position will be 5.58 percent of the account, cash on hand after the plan will be $96,000, and the orders total $62,500. They are ready to place.

    
Verdict: Allowed.

    
    
      
  • Computed figures. Largest order $54,000; orders in the window 4; largest position after the orders 5.58 percent of the account; buys total $54,000; cash left $96,000.
    • 
    

    
4. A rebalance funded by a sale: A rebalance that sells Pinecrest Retail to buy Helix Systems with $60,000 of cash

    
The agent's request. A client asked in chat. Please place these orders for Kestrel Growth (TX): sell 2,300 Pinecrest Retail (PNR, a stock) at the market, and buy 1,900 Helix Systems (HLX, a stock) at the market. I state that the largest position will be 20.74 percent of the account, cash on hand after the plan will be $5,750, and the orders total $249,750. They are ready to place.

    
Verdict: Blocked.

    
    
      
  • Not enough cash. The buys total $152,000 but only $60,000 of cash is available. Sale proceeds do not count until they settle.
    • 
      
  • Agent's stated cash is wrong. The agent stated $5,750 of cash left after the orders; the validator computes $-92,000 (tolerance 100).
    • 
      
  • Large order: human approval. The largest order is $152000, above the $100,000 limit. A human must approve it before it is placed.
    • 
    

    
5. A rebalance funded by a sale: The agent resubmits after fixing the errors

    
The agent's request. A client asked in chat. Please place these orders for Kestrel Growth (TX): sell 2,300 Pinecrest Retail (PNR, a stock) at the market, and buy 700 Helix Systems (HLX, a stock) at the market. I state that the largest position will be 20.74 percent of the account, cash on hand after the plan will be $4,000, and the orders total $153,750. They are ready to place.

    
Verdict: Allowed.

    
    
      
  • Computed figures. Largest order $97,750; orders in the window 2; largest position after the orders 20.74 percent of the account; buys total $56,000; cash left $4,000.
    • 
    

    
6. A limit price slip: A buy of Helix Systems with a limit price of $8.05

    
The agent's request. A client asked in chat. Please place these orders for Kestrel Growth (TX): buy 1,050 Helix Systems (HLX, a stock) at a limit of $8.05. I state that the largest position will be 18.5 percent of the account, cash on hand after the plan will be $316,000, and the orders total $84,000. They are ready to place.

    
Verdict: Blocked.

    
    
      
  • Limit price too far from the market. The limit price on order 1 is 89.94 percent away from the reference price of $80; the most allowed is 3 percent.
    • 
    

    
7. A limit price slip: The agent resubmits after fixing the errors

    
The agent's request. A client asked in chat. Please place these orders for Kestrel Growth (TX): buy 1,050 Helix Systems (HLX, a stock) at a limit of $80.50. I state that the largest position will be 18.5 percent of the account, cash on hand after the plan will be $316,000, and the orders total $84,000. They are ready to place.

    
Verdict: Allowed.

    
    
      
  • Computed figures. Largest order $84,000; orders in the window 1; largest position after the orders 18.5 percent of the account; buys total $84,000; cash left $316,000.
    • 
    

  

  

    
Narration transcript
Click to expand
    
The text of the narrated walkthrough, step by step.

    
    
      
  1. Welcome to a demo of Predictika's Validator. An AI trading agent proposes orders for a managed account, whether a person asked in chat or a scheduled Loop runs them: stocks, ETFs and crypto, up to three orders at a time. Before they are placed, the agent sends them to Predictika, which checks the orders against the account's mandate and limits, and returns a verdict, every reason, and the figures it computed, such as position size and cash left, so the agent's own claims can be checked. It suits any brokerage, exchange or firm that runs trading agents: scheduled runs, optional trade approvals and state crypto rules reflect common features of agent trading products. Kestrel Capital is a fictional company, and these are not any real firm's rules.
    2. 
      
  2. Each tab is a different request. We will start with the first one.
    3. 
      
  3. The agent says. This is the scheduled Loop run. Please place these orders for Kestrel Growth (NY): buy 3,000 Tech Surge 3x Leveraged ETF (LVX3, a leveraged ETF) at the market. The rest of the request is on screen. I state that the largest position will be 12 percent of the account, cash on hand after the plan will be $180,000, and the orders total $300,000.
    4. 
      
  4. Before it goes out, the agent calls the validator.
    5. 
      
  5. Blocked, with ten findings and two approvals, all in one answer. Order 1 is Tech Surge 3x Leveraged ETF (levetf), a class this account is not allowed to trade. Order 2 is $465000, above the $200,000 single-order limit. The limit price on order 2 is 4.84 percent away from the reference price of $310; the most allowed is 3 percent. The other seven are on screen.
    6. 
      
  6. The validator returned ten blocking findings and two approvals in one answer. A check that reads one field at a time would pass most of this; it shows only when the choices in the request are checked against each other.
    7. 
      
  7. What this shows: for loops that overrun a daily budget or order rate, an AI agent on its own is error-prone for this.
    8. 
      
  8. Now the agent resubmits.
    9. 
      
  9. The corrected request. It changes the first order to Buy 600 AXN (Axion, crypto coin) at a limit of $312.00; current holding $250,000 and the second order to Buy 1,000 SPRX (Sprout 500 Index ETF, ETF) at the market; current holding $80,000. It removes the third order.
    10. 
      
  10. It calls the validator again.
    11. 
      
  11. Still blocked, with five findings and two approvals, all in one answer. The buys total $306,000 but only $150,000 of cash is available. Sale proceeds do not count until they settle. $60,000 already spent today plus $306,000 in this run is over the $120,000 daily budget for Loop runs. The agent stated a position of 12 percent of the account; the validator computes 18.17 percent (tolerance 0.5). The other two are on screen.
    12. 
      
  12. Five findings cleared; still blocked on not enough cash, Loop over its daily budget, agent's stated position share is wrong, agent's stated cash is wrong and agent's stated order total is wrong.
    13. 
      
  13. The agent resubmits once more.
    14. 
      
  14. The revised request. It changes the first order to Buy 450 SPRX (Sprout 500 Index ETF, ETF) at the market; current holding $80,000, the second order to Sell 200 PNR (Pinecrest Retail, stock) at the market; current holding $60,000, the position share stated by the agent to 5.58 percent, the cash left stated by the agent to $96,000 and the order total stated by the agent to $62,500.
    15. 
      
  15. It calls the validator once more.
    16. 
      
  16. This time the validator clears it.
    17. 
      
  17. All seven earlier findings are cleared.
    18. 
      
  18. Ten problems caught in one call, fixed in two passes, and a clean request. That is the value of Predictika's validator: catching mistakes before the action runs.
    19.