Reflect 3 Attempt 01

Portfolio Management JSON Generation Skill

Purpose

Generate structured portfolio-management decision JSONs by combining shared-environment data with request-specific constraints. The tasks span fixed-income (PF-FI-LUMEN) and multi-asset (PF-MA-HELIO) portfolios.

When to Use

• You receive a prompt asking to prepare a desk proposal, portfolio review, allocation update, risk meeting memo, or committee decision file.
• The prompt references a portfolio_id (PF-FI-LUMEN or PF-MA-HELIO) and points to input/payloads/committee_request.json (or similar request file) and input/payloads/answer_template.json.
• You must return only JSON matching a strict schema.

SOP

1. Read the Three Input Files

Always read, in this order:

1. input/prompt.txt — the narrative instruction (what the stakeholder wants).
2. input/payloads/<request>.json — the structured request (dates, instruments, bands, candidate shortlists, baseline weights, etc.).
3. input/payloads/answer_template.json — the schema contract (required top-level keys, field types, enums, precisions, ordering rules, required values).

Key insight: The answer template is a specification, not an example. Every key, enum value, precision, and ordering rule is enforced.

2. Fetch All Shared-Environment Data

Query the Asteria Investment Office environment for the endpoints relevant to the portfolio:

Fetch all five endpoints even if the prompt only names a subset; cross-referencing is usually required.

3. Map Request Requirements to Template Fields

Create a field-by-field checklist. Common mappings:

Fixed-Income (PF-FI-LUMEN) Tasks

• portfolio_id → always "PF-FI-LUMEN" (hard-coded).
• proposal_date / review_date → from request payload or prompt.
• trades / sells / buys → derived from:
  • Watchlist instruments that must be reduced/removed.
  • HY exposure that exceeds the policy band.
  • Cash injection or stale exceptions that fund new positions.
  • Candidate shortlist that provides replacement IG instruments.
• post_trade_hy_exposure_pct → recalculate after sells/buys; must respect target_hy_band_max.
• post_trade_duration → net duration after rotation; must stay inside CIO range.
• watchlist_breach_flag → true if any watchlist instrument remains post-trade.
• next_step → depends on whether all constraints are satisfied.

Multi-Asset (PF-MA-HELIO) Tasks

• portfolio_id → always "PF-MA-HELIO".
• review_quarter → often hard-coded (e.g., "Q2_2026").
• correlation_summary → compute Pearson correlation of monthly simple returns from consecutive index levels for the pairs listed in the request.
• target_sleeve_actions → derive from views and risk flags (trim overweight, add underweight, hold neutral, hedge/rotate for risk).
• allocation_views → for each opportunity set:
  1. Read prior_view from /api/environment/prior-views.
  2. Compute or read signal_score from macro signals.
  3. Derive view (UW / N / OW) from signal score and policy thresholds.
  4. Set change by comparing view to prior_view.
  5. Set conviction based on signal strength dispersion.
  6. Pick rationale_code from the allowed enum that best matches the dominant macro signal.
• rebalance_trigger → select the enum value that matches the primary reason for the committee action.
• portfolio_risk_concentration_flag → true if any correlation, band, or watchlist threshold is breached.

4. Perform Calculations Rigorously

Monthly Simple Returns

return_t = (level_t - level_{t-1}) / level_{t-1}

Use consecutive index levels from the review period. Do not annualize.

Pearson Correlation

Compute between the return series of each pair. Round to 3 decimal places.

Critical: Pairs inside the correlation_summary object must have their two index IDs sorted alphabetically.

Post-Trade Portfolio Metrics (Fixed Income)

• HY exposure % = sum of HY holdings / total portfolio value.
• Duration = weighted average of instrument durations (weight = market value).
• Cash = prior cash + proceeds from sells - cost of buys + any injection.
• Round HY exposure to 4 decimal places, duration to 2, cash to 2.

Signal Score to View Mapping (Multi-Asset)

• Typical threshold: signal_score > 0.5 → OW, < -0.5 → UW, else N. (Confirm against policy document if provided.)
• change is UP if view improves from prior, DOWN if worsens, UNCHANGED if same.

5. Apply Business Rules in Priority Order

1. Watchlist reduction first — sell watchlist instruments to the extent possible.
2. HY band compliance — if post-trade HY % > max band, sell more HY or add IG.
3. Duration band compliance — keep duration inside CIO min/max; use duration ballast candidates if needed.
4. Avoid new watchlist buys — if the request specifies avoid_new_watchlist_buy: true, exclude candidates flagged as watchlist risk.
5. Stale exception handling — stale exceptions may be swapped for cleaner candidates of similar profile.
6. Cash deployment — inject or deploy cash only after risk constraints are met.

6. Enforce Schema Compliance

Before returning JSON, verify every field against the answer template:

• All required_top_level_keys are present.
• Keys appear in the exact order specified (if an item_order is given).
• required_value fields match exactly (e.g., portfolio_id, review_quarter).
• Enum fields use only allowed values.
• Numbers match the specified precision.
• Lists have the required length.
• Nested objects contain only the declared sub-fields.
• No extra commentary, markdown, or keys outside the schema.

Common pitfall: Using "USD" in a trade action field when the template expects "buy" or "sell". Always cross-check the enum list in the template.

7. Validate Logical Consistency

• If watchlist_breach_flag is false, ensure no watchlist instruments remain.
• If cio_review_flag is true, the cio_review_reason must not be "NONE".
• hy_reduction_pct_points should equal the difference between pre- and post-trade HY %.
• target_allocations deviations must sum to zero (or near-zero within rounding).
• signal_score sign should match the direction of view (positive → OW or N, negative → UW or N).

8. Output Only Valid JSON

Return the raw JSON object. Do not wrap it in markdown code fences, do not add a preamble like "Here is the JSON:", and do not include any narrative explanation. The downstream consumer parses the response directly.

Task-Specific Cheat Sheet

Remember

• The answer template is the contract: every rule, enum, precision, and ordering constraint in it is enforced.
• Fixed-income tasks revolve around HY exposure, duration, and watchlist.
• Multi-asset tasks revolve around correlations, macro signals, and active views.
• Always fetch all five environment endpoints even when the prompt only names a subset; the data is cross-referenced implicitly.
• Return only the JSON object—no markdown, no preamble, no trailing commentary.