skills/idea-generation/references/workflow.md

# PM-Grade Idea Generation Workflow

Use this reference for full PM-pack outputs, portfolio-aware screens, watchlist refreshes, or any request where the user expects a systematic workflow rather than a quick list.

## 1. Establish mandate and universe

Capture or infer:

- Mandate: long-only, long/short, market-neutral, event-driven, Credit Markets, sector research, equity research, or multi-asset.
- Objective: research queue, trade candidate, benchmark-relative idea, hedge, watchlist, earnings setup, or sector screen.
- Direction: long, short, long/short, pair, relative value, catalyst, or watchlist.
- Time horizon: days/weeks, 1-3 months, 3-12 months, or 1-3 years.
- Geography, sector, industry, market cap, liquidity, benchmark, and instrument type.
- Constraints: minimum average daily value, max position size, restricted sectors, ESG/restricted lists, leverage, borrow, options availability, currency, country, and benchmark limits.

If unspecified, use liquid public equities and a 3-18 month fundamental horizon. State assumptions.

## 2. Build or validate the investable universe

Before scoring, resolve:

- Tickers, identifiers, exchanges, local lines, ADRs, and share classes.
- Active listings vs stale listings.
- Market cap and liquidity.
- Sector, industry, region, and benchmark membership.
- Portfolio/watchlist/coverage overlap.
- Data sufficiency and freshness.
- Exclusions: illiquid microcaps, highly binary pre-commercial biotech, special purpose vehicles, funds, non-operating shells, or companies with inadequate data, unless requested.

Flag special cases:

- Recent IPO, spin-off, merger, divestiture, bankruptcy emergence, restatement, fiscal-year change, or suspended guidance.
- Financials, REITs, E&P, biotech, insurers, and banks requiring sector-specific metrics.

## 3. Choose archetypes before scoring

Do not run one generic composite score. Select multiple archetypes based on mandate:

- Long: quality compounder, GARP, revision inflection, derated quality, self-help, capital allocation, SOTP, post-earnings overreaction.
- Short: over-earning, deteriorating revisions, quality trap, balance sheet/refinancing, narrative excess, accounting/cash-conversion.
- Relative value: peer mispricing, pair trade, long quality/short deteriorating peer, sector-neutral basket.
- Event: M&A, spin, activism, litigation, regulatory, restructuring, product launch, index event, debt refinancing.
- Watchlist: good setup missing price, catalyst, data confirmation, or risk clarity.

Use `references/screen-archetypes.md` for details.

## 4. Score the relevant dimensions

Score qualitatively or quantitatively, but explain the drivers. Default dimensions:

1. Valuation: absolute, historical, peer-relative, growth-adjusted, normalized, SOTP, or sector-specific.
2. Growth: rate, durability, organic quality, unit economics, operating leverage, and reinvestment runway.
3. Revisions: direction, magnitude, breadth, recency, dispersion, guidance delta, and whether revisions may continue.
4. Quality: ROIC/ROE, margins, FCF conversion, accounting quality, balance sheet, management, moat, and capital allocation.
5. Momentum: price, relative strength, volume, post-earnings drift, and fundamental momentum.
6. Catalysts: timing, probability, magnitude, path dependency, and whether priced.
7. Risk: balance sheet, liquidity, crowding, borrow, squeeze, accounting, governance, regulatory, litigation, macro, factor, and portfolio risk.
8. Portfolio fit: benchmark fit, diversifying exposure, factor impact, position capacity, and overlap with existing holdings.
9. Data quality: source freshness, completeness, comparability, and confidence.

## 5. Apply PM triage

For every top candidate, ask:

- Why did it screen well?
- Is that reason economically meaningful or just statistical?
- Is the metric distorted by accounting, cyclicality, one-time items, commodity/FX/tax effects, or stale consensus?
- Is the valuation denominator credible?
- What is the market debate?
- What is consensus likely pricing?
- Is there a real variant perception or merely a known fact?
- Is there a catalyst or reason the market will care?
- Is the risk compensated?
- Is the idea investable given liquidity, borrow, capacity, mandate, and portfolio context?
- What is the first smart reason to reject it?
- What is the next highest-value research step?

Kill weak ideas aggressively. A shorter, sharper list is better than a broad undifferentiated screen.

## 6. Classify priority

Use these buckets:

- A - immediate research candidate: strong setup, clear why-now, credible variant view, manageable risk, next work obvious.
- B - watchlist / needs trigger: interesting but missing catalyst, price, data confirmation, risk clarity, or valuation support.
- C - screen flag only: statistically interesting but weak actionability or weak variant view.
- Reject: false positive, stale data, mandate mismatch, poor liquidity, uninvestable risk, no edge, or factor trap.

Do not overstate precision. Narrative judgment outweighs a numeric score.

## 7. Build the output

Default PM-pack structure:

1. Scope and assumptions.
2. Methodology and source caveats.
3. Executive summary and top findings.
4. Ranked idea table.
5. Idea cards for top A/B candidates.
6. Rejected false positives.
7. Cross-idea themes and risk observations.
8. Recommended downstream routing.
9. Data gaps and next research asks.

Quick-scan structure:

- One-paragraph scope and caveats.
- Ranked table of 5-20 ideas.
- One-line PM rationale and rejection risk for each.
- Next step/routing.

Watchlist-refresh structure:

- Names that moved up in priority.
- Names that moved down or should be killed.
- New catalysts and revision changes.
- Price/valuation range changes.
- Thesis tracker updates and next actions.

Portfolio-aware structure:

- Additive ideas.
- Duplicative exposures.
- Hedges/pairs.
- Factor and sector implications.
- Liquidity/capacity observations.
- Names that improve portfolio construction vs just standalone appeal.

## 8. Maintain an idea log when appropriate

For recurring use, append an idea log with:

- Date generated.
- Security and ticker.
- Setup type and direction.
- Price/date if available.
- Rank and priority.
- Why surfaced.
- Variant perception.
- Catalyst path.
- Next work.
- Assigned owner if provided.
- Status: rejected, watchlist, active research, invested, exited.
- Later outcome and lessons learned.

A mature process reviews hit rate by archetype and false-positive patterns over time.

## PM Triage Upgrade

Top ideas require Actionability, Variant Wedge, Why Now, First Rejection, What Would Make It Investable, What Would Kill It, and Next Workflow. Separate interesting from actionable.