risk-debate

name	risk-debate
description	Three-perspective risk evaluation (Aggressive / Conservative / Neutral) of a draft Atlas pick. After the bull/bear debate settles a rating, three orthogonal risk personas each respond to the investment plan. A Portfolio Manager synthesizes their views into a final structured recommendation with entry/exit/stop/account placement. Pattern lifted from TauricResearch/TradingAgents.
triggers	["risk debate","three perspective risk","aggressive conservative neutral","position sizing","downside math","portfolio manager review","risk panel","should I really enter","enter at this size","how much should I buy","what size","oversized position"]
tier	research
dependencies	["research-debate","realised-return-reflection"]
origin	TauricResearch/TradingAgents (patterns)

Risk Debate — 3-Perspective Risk Panel

Purpose

After the bull/bear research debate settles a rating, this skill stress-tests how the pick should actually be sized and placed. Three personas with orthogonal mandates each evaluate the same pick:

• Aggressive — champions the upside, presses for bolder position size
• Conservative — protects capital, presses for smaller size or no entry, anchored to CC's "capital preservation > tax optimization > compounding" doctrine
• Neutral — balanced middle path, presses for staged entry and proper account placement (TFSA/RRSP/FHSA)

A Portfolio Manager synthesizes the three views into a final 5-tier rating + entry/exit/stop/account/horizon.

When to Trigger

• Always runs alongside research-debate when debate_rounds > 0
• Standalone via from research.debate import run_risk_debate when CC asks "should I really enter at this size"
• When a pick's downside is >15% of position value
• When account placement is non-obvious (e.g. growth stock with US dividend → TFSA vs RRSP trade-off)

How It Works

Investment Plan (from research-debate)
            │
            ▼
    Aggressive  ◄──┐
            │      │ N rounds
            ▼      │
    Conservative  ─┤
            │      │
            ▼      │
    Neutral   ◄────┘
            │
            ▼
    Portfolio Manager (deep_think tier)
            │
            ▼
    Final rating + entry/exit/stop/account/horizon

The PM receives all three persona transcripts plus past Atlas reflections on the ticker (cross-ticker lessons included) and produces a final structured recommendation.

Output Format

• Rating — one of Buy / Overweight / Hold / Underweight / Sell
• Final recommendation — 4-8 sentence prose with concrete entry zone, exit target, stop level, account placement, time horizon
• Revisit conditions — 1-2 sentence "watch list" — what would make Atlas re-rate
• Disclaimer — Atlas is research, not financial advice; CC executes manually

Atlas-Specific Anchoring

The Conservative prompt is anchored to CC's runway floor ($10K CAD when he moves to Montreal). A 30% drawdown on a 5K non-registered position is treated as a real impact to runway, not abstract risk.

The Neutral prompt is anchored to CC's account inventory: TFSA growth, RRSP US-dividend treaty, FHSA first-home equities, non-registered for short-term.

Cost Note

3 personas × N rounds × quick_think + 1 × deep_think synthesis. With max_rounds=1 and the default tier mapping (Haiku quick / Opus deep), a single risk debate costs roughly 4× a single LLM call.

Internal References

• [[risk_aggressive]] — aggressive persona prompt
• [[risk_conservative]] — conservative persona prompt with CC's runway anchor
• [[risk_neutral]] — neutral persona prompt with account-placement anchor
• [[portfolio_manager]] — synthesis prompt with 5-tier rating
• [[debate]] — Python orchestrator
• [[risk-aggressive]] — agent file
• [[risk-conservative]] — agent file
• [[risk-neutral]] — agent file