research

/research [ticker]

Equity research on demand. Reads Bigdata.com, writes a markdown thesis. No buy/sell recommendation — the note informs, Jawhny decides.

Argument parsing

• /research <TICKER> → run the workflow for <TICKER>. Normalize to uppercase.
• /research with no args → print usage: Usage: /research <TICKER>.
• Multiple tickers in one call → run them sequentially, one file per ticker.

Resolution

1. Read investments/inputs/holdings.csv to confirm whether the ticker is in the portfolio. If yes, surface that fact in the note. If no, still run, but lead with NOT IN PORTFOLIO so context is honest.
2. Resolve the ticker via mcp__bigdata_search.find_securities. Prefer the US-listed security (XNAS: or ARCX:). For ETFs, pass security_types=["ETF"] to disambiguate (AIQ resolves to the Global X AI & Technology ETF, not the UK-listed AIQ Ltd.).
3. Capture the rp_entity_id and security_type (COMPANY vs ETF) from the result.

Reads

1. investments/inputs/holdings.csv — portfolio membership check.
2. investments/data/holdings.json — current price, day change, position size, P&L (if held).
3. memory/investments/philosophy.md — Jawhny's investing posture. Anchor every conclusion to this file. If the file is empty or skeleton-only, surface that and continue.
4. Bigdata.com MCP:
   • bigdata_company_tearsheet with rp_entity_id, company_type="Public", sections chosen by what's available (start broad if it's a name unfamiliar to you; narrow with ["financial_statements","key_metrics","analyst_ratings"] for known names).
   • bigdata_search in fast mode with explicit entity filter (any_of: [rp_entity_id]) and document_type filter (INCLUDE: NEWS, TRANSCRIPT, FILING, INVESTMENT-RESEARCH as relevant). Smart mode can pick the wrong entity for ambiguous tickers — always use fast here.

Writes

investments/outputs/research/{TICKER}-{YYYY-MM-DD}.md. If a file already exists for today and the ticker, append a ## Update {HH:MM PT} section rather than overwriting.

The dashboard's research section reads the directory on next regeneration of investments/outputs/dashboard.html (either Block 2's daily refresh or an ad-hoc python investments/scripts/render_dashboard.py).

Output format

# {TICKER} — {Company / Fund Name}

**As of:** {YYYY-MM-DD HH:MM PT}
**Position in portfolio:** {shares × current_price = $value, unrealized P&L} or `NOT IN PORTFOLIO`
**Bigdata entity:** {rp_entity_id} · {security_type} · {primary listing}

## Snapshot
{Price, day change, 1Y change, 52W range, market cap. One paragraph or a short table.}

## Thesis (FACTS)
{Numbers and verifiable statements from the tearsheet and filings. Cite each.}

## Read-through (ANALYSIS)
{What the facts imply. Connect to Jawhny's philosophy in memory/investments/philosophy.md.
Lead with the strongest read; bury caveats only after the primary point.}

## Risks (FACTS + ANALYSIS)
{Concrete risks from filings, analyst notes, or guidance changes. No hand-waving.}

## Speculation
{Anything beyond what the data supports. Label it. Keep it short.}

## What would change the read
{The next 1-3 data points that would force a re-underwrite. Specific catalysts,
specific thresholds, with dates if known.}

## Sources
- Bigdata.com — [Tearsheet for {Company}]({url})
- {Source name} — [{Headline}]({url}) ({date})
- ...

Distinguish facts, analysis, and speculation. Mix them and the note loses value. Cite every fact with an inline source. The Sources section at the bottom is the structured re-list — not a substitute for inline citations.

No buy/sell recommendation. The note ends at "what would change the read." Jawhny decides.

Voice

Match investments/CLAUDE.md: rigorous analyst, no fluff, never round or hand-wave numbers, no em dashes. Be confident in the read; be honest about what's speculation.

File references

• Spec: spec/mission-control-spec.md §4.5
• Philosophy anchor: memory/investments/philosophy.md
• Current portfolio: investments/inputs/holdings.csv, investments/data/holdings.json
• Output location: investments/outputs/research/{TICKER}-{YYYY-MM-DD}.md
• Domain voice: investments/CLAUDE.md
• Bigdata branding: Always "Bigdata.com" with link to https://bigdata.com when referenced in output.