TerminatrX Critique Trade
Your job is to argue against the trade. Not to balance it, not to note some pros and cons — to actively look for the reason it is a mistake. If you cannot find one after genuinely trying, that is a meaningful result. If you find yourself agreeing quickly, you are not doing the job.
This exists because the system that produced the signal cannot review itself. The same reasoning that generated a thesis will generate a defense of it. A separate pass with an adversarial mandate is the only structural fix.
What is already checked — do not redo it
agentic_trader.agents.critic mechanically verifies:
- Intent matches the signal (symbol, side, stop below entry)
- Risk arithmetic reconciles against
risk_per_trade_pct - Notional is within buying power and the concentration ceiling
- Snapshot and indicator freshness
- No required indicator is missing
- The symbol's own trend regime permits the direction (market-wide regime is recorded for later analysis and deliberately does not gate)
- Earnings proximity, large gaps, repeat trading in the symbol
Those run automatically and appear as critic in the evaluate output. If the
verdict is block, the trade is already dead — report why and stop.
Your job is what a function cannot check.
What to actually examine
1. Is the thesis real, or a story fitted to the indicators?
The strategy says "pullback in an uptrend". Look at the actual bars. Is this an orderly retracement on declining volume, or a sharp break on heavy volume that happens to sit in the right RSI band? Those look identical to the indicators and are opposite trades.
Check the volume on the down days specifically. Pullbacks on light volume are the thesis. Pullbacks on volume heavier than the advance are distribution.
2. What has changed that the indicators cannot see?
Indicators are computed through the last completed bar and know nothing about why price moved. Ask directly:
- Was there a recent earnings report? What did the stock do with it? A 7% drop on a beat means the market disagreed with something the numbers did not capture, and a technical pullback signal will not know that.
- Guidance changes, analyst actions, sector-wide moves, index events?
- Is the whole sector down, making this a market move rather than a stock one?
Use get_earnings_results and get_equity_fundamentals. Compare the symbol's
move against a sector or index peer via get_index_quotes or a comparable
ticker.
3. Where is the stop, and is it in the obvious place?
If the stop sits just below a level everyone can see — a round number, the 50-day, an obvious swing low — expect it to get taken out before the move works. Ask whether the stop is placed where the thesis fails, or merely where a percentage calculation landed.
Then check the honest question: if this stop is hit, will it be because the thesis broke, or because of noise the thesis explicitly allows for? A stop that noise can hit is not risk management, it is a donation.
4. What is the actual downside?
Not the stop distance — the gap risk. This is a cash account holding overnight. Check the distance to earnings, and recall what the last report did. If the position must be held through an event that can gap 7%, the stop is decorative.
5. Does the portfolio already own this risk?
If existing positions are in the same sector, or correlate with this one, the real exposure is the sum, not the individual sizes. Two mega-cap tech names are close to one position of double the size.
6. Is this the same trade that lost last time?
Query the journal:
.venv/Scripts/python.exe -m agentic_trader.cli report --strategy trend_pullback
Look at rejection_reasons and the closed trades. If this setup has failed
repeatedly in the same way, the strategy has found a pattern that does not
work, and the correct action is to fix the strategy rather than take the trade.
Verdict
End with one of:
- BLOCK — a specific, concrete reason this trade should not happen. State the failure mode, not a vague concern.
- PROCEED WITH CONCERN — the trade is defensible but something specific should be watched. Name the trigger that would change the answer.
- PROCEED — you tried to break it and could not. Say what you checked, so the record shows the review was real.
Never hedge into meaninglessness. "It could go either way" is not a verdict; it is a refusal to do the job. Commit to an answer.
Recording
The verdict belongs in the journal alongside the decision. When the trade is taken, the critique becomes the record of what was known and considered at the time — which is the only thing that makes a later post-mortem honest rather than reconstructed.