Decision dashboard

Decision Dashboard

Build a batch decision dashboard for: $ARGUMENTS

Overview

• Implementation status: code-backed
• Local entry script: <bundle-root>/decision-dashboard/run.py
• Primary purpose: summarize conditional actions, scores, and regimes across a symbol set instead of returning one-symbol reports
• Research layer: batch decision support (Stage 6: Risk Management & Position Sizing - Portfolio-level view)
• Workflow stages: stage 1 Universe Review and stage 6 Risk Management & Position Sizing in batch form
• Local executor guarantee: compute per-symbol decision summaries under shared account assumptions, aggregate action counts, and return a ranked focus list

Use When

• The user wants a batch decision board rather than raw ranking.
• The user wants action distribution across a watchlist.
• The user wants to shortlist names before moving into one-symbol detail.
• The user wants portfolio-level view of potential actions.
• The user wants to understand aggregate exposure and risk across watchlist.
• The user wants daily monitoring dashboard for existing positions and watchlist.

Do Not Use When

• The user wants a complete report for one symbol. Use market-brief or analysis.
• The user only wants ranked screening output. Use market-screen.
• The user wants only watchlist ingestion. Use watchlist-import.
• The user expects this to produce a portfolio optimizer, cross-name correlation control, or live order book. It does not.
• The user wants actual portfolio construction with correlation and diversification. This skill does not model cross-name correlations.

Inputs

• Direct positional symbols.
• Optional --file PATH: plain text file containing one symbol per line.
• Optional --dir PATH: directory of per-symbol CSV files.
• Optional --capital, --cash, --risk, --max-position: influence dashboard decision framing.
• Optional --top N, --start, --end, --source.
• Important boundary: --file is a symbol-list file, not a market-data CSV.
• Assumption note:
  • the same account assumptions are applied across the batch
  • the agent should identify when those assumptions are user-supplied versus defaulted

Execution

Step 1: Define dashboard scope and portfolio context

Before building dashboard, clarify objectives and constraints:

Dashboard objectives:

• Daily monitoring (track existing positions and watchlist)
• Idea triage (identify new opportunities from watchlist)
• Position review (assess current holdings for hold/reduce/exit)
• Risk review (identify concentration, exposure, risk flags)
• Action planning (prioritize symbols for deeper analysis)

Portfolio context:

• Total capital available
• Current cash balance
• Existing positions (symbols, quantities, cost basis)
• Risk tolerance (max position size, max portfolio risk)
• Concentration limits (max % per position, per sector)

Shared assumptions:

• Capital: Total portfolio value (default: user-supplied or session default)
• Cash: Available cash for new positions (default: 100% of capital if no positions)
• Risk per position: Max % of capital per position (default: 5-10%)
• Max position count: Max number of positions (default: 10-20)
• Rebalancing threshold: When to adjust positions (default: ±20% from target)

A-share specific constraints:

• T+1 settlement (cannot sell same-day purchases)
• Lot size (100 shares minimum)
• Price limits (±10% or ±20%)
• Suspension risk
• Margin requirements (if using margin)

Step 2: Validate universe and collect data

Process watchlist with quality validation:

Universe validation:

• All symbols are valid A-share tickers
• Symbols are currently tradeable (not delisted)
• Symbols are not suspended (or suspension flagged)
• Symbols have sufficient data history (minimum 60 days)
• Symbols meet minimum liquidity requirements

For each symbol, collect:

• Market data (OHLCV, normalized and validated)
• Technical features (trend, momentum, volatility, volume)
• Market state (score, regime, bias, risk flags)
• Current position (if held: quantity, cost basis, P&L)
• Liquidity metrics (volume, turnover, bid-ask spread)

Partial failure handling:

• Skip invalid symbols and record reason
• Skip suspended symbols and flag for monitoring
• Skip insufficient data symbols and record reason
• Continue processing remaining symbols

Step 3: Generate per-symbol decision summaries

For each valid symbol, generate conditional action:

Decision logic (consistent with decision-support):

For symbols NOT currently held:

• Buy (score 70-100, bullish bias, no major risk flags):
  • Suggested position size (% of capital)
  • Entry price range
  • Initial stop-loss level
  • Confidence level
• Watch (score 50-69, neutral/uncertain bias, or minor risk flags):
  • Reason for watching (waiting for confirmation, better entry, etc.)
  • Trigger conditions (what would change this to Buy)
• Avoid (score 0-49, bearish bias, or major risk flags):
  • Reason for avoiding (downtrend, overbought, high risk, etc.)

For symbols currently held:

• Hold (score 60-100, position within target range, no major risk flags):
  • Current position size vs. target
  • Unrealized P&L
  • Stop-loss level
  • Monitoring conditions
• Reduce (score 40-59, position above target, or moderate risk flags):
  • Suggested reduction amount (shares or %)
  • Reason for reduction (risk management, rebalancing, etc.)
  • Target position size after reduction
• Exit (score 0-39, major risk flags, or stop-loss triggered):
  • Reason for exit (stop-loss, invalidation, risk, etc.)
  • Expected exit price
  • Realized P&L (if exited)

Position sizing (for Buy actions):

• Fixed sizing: Equal weight across all positions (e.g., 5% each for 20-position portfolio)
• Score-based sizing: Higher score = larger position (within limits)
• Risk-based sizing: Size based on volatility (higher vol = smaller position)
• Liquidity-constrained: Limit position to % of ADV (e.g., max 5% of 20-day ADV)

Risk checks:

• Position size within limits (max % per position)
• Total exposure within limits (max % invested)
• Concentration within limits (max % per sector, if tracked)
• Cash sufficiency (enough cash for suggested buys)
• Liquidity sufficiency (can execute without excessive impact)

Step 4: Aggregate action distribution

Summarize actions across entire watchlist:

Action counts:

• Buy: Count of symbols with Buy recommendation
• Watch: Count of symbols with Watch recommendation
• Avoid: Count of symbols with Avoid recommendation
• Hold: Count of currently held symbols with Hold recommendation
• Reduce: Count of currently held symbols with Reduce recommendation
• Exit: Count of currently held symbols with Exit recommendation

Action distribution analysis:

• % of watchlist in each action category
• Trend over time (if historical dashboards available)
• Comparison to previous dashboard (if available)

Aggregate exposure:

• Total suggested new capital deployment (sum of Buy position sizes)
• Total suggested capital reduction (sum of Reduce/Exit position values)
• Net capital change (deployment - reduction)
• Resulting cash balance after all actions
• Resulting total exposure (% of capital invested)

Aggregate risk:

• Total portfolio risk (sum of position risks, if calculable)
• Concentration risk (largest position %, largest sector %)
• Liquidity risk (% of positions with low liquidity)
• A-share specific risks (% with price limit risk, suspension risk, etc.)

Step 5: Generate ranked focus list

Prioritize symbols for deeper review:

Focus list criteria:

• High priority (immediate action):
  • Buy recommendations with score > 80
  • Exit recommendations (stop-loss triggered, major risk flags)
  • Reduce recommendations with significant risk flags
• Medium priority (review soon):
  • Buy recommendations with score 70-79
  • Watch recommendations near Buy threshold
  • Hold recommendations with moderate risk flags
• Low priority (monitor):
  • Watch recommendations not near threshold
  • Hold recommendations with no risk flags
  • Avoid recommendations (no action needed)

Focus list ranking:

• Sort by priority (high > medium > low)
• Within priority, sort by score (descending)
• Within same score, sort by momentum strength
• Break further ties by volume confirmation

Focus list output: For each symbol in focus list (top 10-20):

• Symbol, name, board
• Current action (Buy/Watch/Avoid/Hold/Reduce/Exit)
• Score, trend, regime, bias
• Priority level (high/medium/low)
• Key reason (why this action, why this priority)
• Suggested next step (run market-brief, decision-support, etc.)

Step 6: Generate dashboard report

Organize findings into structured dashboard:

Part 1: Dashboard Summary

• Universe source and size
• Dashboard timestamp
• Valid symbols processed
• Symbols skipped (by category)
• Portfolio context (capital, cash, positions)

Part 2: Action Distribution

• Action counts (Buy/Watch/Avoid/Hold/Reduce/Exit)
• Action percentages
• Trend vs. previous dashboard (if available)
• Aggregate capital deployment/reduction
• Resulting cash and exposure

Part 3: Aggregate Risk Summary

• Total portfolio exposure (% of capital)
• Largest position (symbol, %)
• Concentration risk (sector, if tracked)
• Liquidity risk (% low liquidity positions)
• A-share specific risks (price limits, suspensions, etc.)

Part 4: Focus List For each high/medium priority symbol:

• Symbol, action, score, regime
• Priority level and reason
• Suggested next step
• Key risk flags (if any)

Part 5: Position Review (if positions held) For each current position:

• Symbol, quantity, cost basis
• Current price, market value
• Unrealized P&L ($ and %)
• Action (Hold/Reduce/Exit)
• Reason for action
• Risk flags (if any)

Part 6: Watchlist Review For each watchlist symbol not held:

• Symbol, score, regime
• Action (Buy/Watch/Avoid)
• Reason for action
• Suggested position size (if Buy)
• Risk flags (if any)

Part 7: Skipped Symbols

• Invalid symbols (count and list)
• Suspended symbols (count and list)
• Insufficient data (count and list)
• Calculation errors (count and list)

Part 8: Dashboard Interpretation

• What the dashboard represents (conditional action summary under shared assumptions)
• What it does NOT represent (final portfolio instructions, optimized allocation)
• Recommended workflow (focus list → deeper analysis → execution planning)
• Caveats and limitations (no correlation modeling, no optimization, etc.)

Step 7: Run the local executor

python3 <bundle-root>/decision-dashboard/run.py <symbol ...> [--file PATH] [--dir PATH] [--capital N] [--cash N] [--risk N]

Step 8: Interpret dashboard correctly

When delivering results, maintain proper framing:

Dashboard interpretation:

• This is a batch conditional action summary, not a ready-to-trade portfolio
• Actions are based on shared assumptions (capital, risk, etc.)
• Actions are conditional on deeper verification (thesis, catalyst, execution)
• Actions do not account for cross-name correlations or diversification

Assumption transparency:

• State which assumptions are user-supplied vs. defaulted
• State capital, cash, risk tolerance, position limits
• State that assumptions are applied uniformly across batch
• State that individual symbols may need different assumptions

Limitations disclosure:

• No portfolio optimization: No mean-variance optimization, no efficient frontier
• No correlation modeling: Treats each symbol independently
• No diversification analysis: Does not assess portfolio diversification
• No transaction cost optimization: Does not minimize turnover or costs
• No timing optimization: Does not consider optimal entry/exit timing

Recommended workflow:

1. Review focus list (high priority symbols first)
2. Run market-brief or analysis on focus list for deeper review
3. Run decision-support on selected symbols for refined position sizing
4. Run strategy-design on selected symbols for execution planning
5. Execute trades via paper-trading or live trading system
6. Rerun dashboard daily/weekly to monitor and adjust

Step 9: Route focus list to deeper skills

Pass focus list to appropriate next skill:

For high priority Buy recommendations:

• market-brief: Quick verification of setup
• analysis: Full thesis development
• decision-support: Refined position sizing with individual assumptions
• strategy-design: Execution planning

For high priority Exit recommendations:

• strategy-design: Exit execution planning
• paper-trading: Simulated exit to verify execution

For medium priority Watch recommendations:

• market-brief: Monitor for trigger conditions
• technical-scan: Pattern recognition for entry timing

For position review (Hold/Reduce):

• backtest-evaluator: Retrospective evaluation of position
• decision-support: Reassess position sizing
• strategy-design: Rebalancing execution planning

Output Contract

• Minimum local executor output: human-readable text beginning with 决策仪表盘.
• Core fields: total symbol count, action distribution, and a ranked focus list.
• Partial failure behavior: invalid symbols or malformed per-symbol files are listed under 跳过 if at least one valid result exists.
• Caller-facing delivery standard:
  • Eight-part structure: Dashboard summary, action distribution, aggregate risk, focus list, position review, watchlist review, skipped symbols, dashboard interpretation
  • Universe and assumption transparency: Identify universe source and shared account assumptions (capital, cash, risk, limits)
  • Action distribution summary: Counts and percentages for Buy/Watch/Avoid/Hold/Reduce/Exit
  • Aggregate exposure and risk: Total deployment, reduction, resulting cash, exposure, concentration
  • Focus list prioritization: High/medium/low priority with specific reasons and next steps
  • Position review: Current holdings with P&L, action, reason, risk flags
  • Conditional framing: Treat action counts as conditional outputs, not final portfolio instructions
  • Assumption disclosure: State which assumptions are user-supplied vs. defaulted
  • Limitations transparency: State what dashboard does NOT provide (optimization, correlation, diversification)
  • Workflow guidance: Recommend focus list → deeper analysis → execution planning
  • No ready-to-trade claims: Dashboard requires verification before execution

Failure Handling

• Parse and argument errors: non-zero exit with a readable 命令错误 message.
• If all candidates fail: readable failure text beginning with 执行失败:.
• Mixed candidate quality: valid results still return, and invalid entries are summarized under 跳过.
• If assumptions are missing or defaulted, the dashboard should still return but the agent must disclose that limitation.
• Invalid symbols: Skip and list under "Invalid symbols" category.
• Suspended symbols: Flag and list under "Suspended symbols" category.
• Insufficient data: Skip and list under "Insufficient data" category.
• Calculation errors: Skip and list under "Calculation errors" category.

Key Rules

• Prefer this skill when the user wants breadth first and depth second.
• Keep the distinction between symbol lists and data-file directories explicit.
• Treat the dashboard as a triage output, not as a substitute for portfolio construction or execution planning.
• If the user wants true portfolio risk budgeting, say that this skill does not model cross-name correlations or optimizer logic locally.
• Assumption transparency is mandatory. Always state capital, cash, risk tolerance, position limits.
• Action distribution must be comprehensive. Count all actions (Buy/Watch/Avoid/Hold/Reduce/Exit).
• Aggregate risk must be assessed. Exposure, concentration, liquidity, A-share specific risks.
• Focus list must be prioritized. High/medium/low priority with specific reasons.
• Position review must include P&L. Unrealized gains/losses for current holdings.
• Conditional framing is critical. Actions are conditional on deeper verification, not final instructions.
• Limitations must be disclosed. No optimization, no correlation modeling, no diversification analysis.
• Workflow guidance must be explicit. Route focus list to appropriate downstream skills.
• No cross-name correlation modeling. Each symbol treated independently.

Composition

• Often follows watchlist-import.
• Commonly feeds market-brief, analysis, decision-support, or strategy-design on shortlisted names.
• Shares decision logic with decision-support but applies in batch form.
• Can be rerun daily/weekly for ongoing portfolio monitoring.
• Should be combined with paper-trading for position tracking and P&L calculation.
• Focus list can feed back into market-screen for refined screening.