01-projects/investing/plugin/skills/smart-money-watch

SKILL

investing-smart-money-watch

Multi-source smart-money surveillance. Two data sources, two cadences, one cross-source reasoning step:

  1. 13F filings (institutional, quarterly, 45-day lag) from SEC EDGAR — curated tracked-manager set
  2. Congressional PTRs (STOCK Act, monthly cadence, 30-45-day lag) from Quiver Quant — curated tracked-member set

Diffs each source vs prior period, identifies cross-manager/cross-member trend lines (multi-manager additions = darlings; multi-manager exits = avoid signals), then cross-references the two sources: when a name appears in BOTH simultaneously, that's a TIER-1 signal.

Status: v1 (2026-05-18). 13F scaffold + 2-year backfill landed 2026-05-17. Congressional data source extension + 12-month backfill landed 2026-05-18.

Why this skill exists (read every time)

Founder's edge is structural-anchor + cross-domain translation, not stock-picking. Smart-money 13Fs do two things for that edge:

  1. Idea generation — when 3+ tracked managers collectively rotate INTO a sector or name we don't yet hold a thesis on, that's a signal to investigate the structural reason. Smart money front-runs the obvious news; the 13F lets us follow them with a 45-day lag.
  2. Thesis corroboration / disconfirmation — when we have an active thesis (memory, power, space, longevity, elon-verse), checking whether the long-only quality investors are tracking with us is a real-world sanity check. If we're bullish on memory and Druckenmiller/Marathon/Berkshire are all trimming, that's a phase-marker we have to take seriously.

13F lag is real (45-day filing deadline after quarter end, so we see Q4 positions in mid-Feb). That's fine for capital-cycle theses operating on 12-36 month structural reads.

Tracked managers are LONG-ONLY positive-sum operators only. Soros / Pershing Square / Greenlight Capital are explicitly excluded — their styles (macro shorts, activist, short-biased) don't map to our operating model. Founder lock 2026-05-17.

Inputs

Invocation

/investing:smart-money-watch [quarter]

Examples:

Cron schedule

60 days after each quarter end. Quarter-end + 60d schedule:

The 45-day SEC filing deadline + 2-week buffer for amendments yields the 60-day cadence. Earlier triggers risk missing filers; later triggers delay signal.

Tracked manager set (v1, founder-locked 2026-05-17)

Manager Style CIK Notes
Duquesne Family Office (Druckenmiller) macro, long-biased 0001536411 Highest-conviction macro shifts; small portfolio, easy to read
Marathon Asset Management capital-cycle (look up) Capital-cycle framework most-aligned with our structural anchors
Berkshire Hathaway long-only quality 0001067983 Slow-moving but the canonical quality benchmark
Pabrai Investments concentrated long-only (look up) Concentrated bets; signal density per holding is high
Gardner Russo & Quinn (Tom Russo) long-only compounder (look up) Long-duration global compounders
Akre Capital Management (Chuck Akre) long-only compounder (look up) "Three-legged stool" compounder lens
Tiger Global Management growth/tech long 0001167483 Read with skepticism post-2022 drawdown; useful as growth-side cross-reference
Appaloosa LP (Tepper) macro long-biased 0001656456 Macro shifts + sector rotation reads

Excluded by founder (do not add):

CIK lookup for "(look up)" managers via SEC EDGAR full-text search at filing time. Cache resolved CIKs back into this table when surfaced.

Data source

SEC EDGAR free API. Endpoints used:

SEC EDGAR access rules:

No paid feeds (Quiver, WhaleWisdom, 13F.info) — EDGAR direct only.

Process per run

  1. Resolve quarter — determine target quarter (input or default-to-prior-filed). Compute quarter-end-date for diffing.
  2. For each tracked manager:
    • Resolve CIK if not cached
    • List 13F-HR filings, find the one covering quarter-end-date
    • If filing not yet available (manager hasn't filed for this quarter), log skip + flag for retry next cron
    • Pull the holdings XML (infotable.xml or equivalent)
    • Parse rows: CUSIP, name of issuer, class of security, value (USD thousands), shares, put/call indicator, investment discretion, voting authority
    • Map CUSIP → ticker via a cached CUSIP-to-ticker lookup (build incrementally; for unmapped, leave as CUSIP)
  3. Diff vs prior quarter for each manager:
    • New positions (in current, not prior)
    • Exited positions (in prior, not current)
    • Increased (>20% share-count growth, or value growth net of price change)
    • Decreased (>20% share-count drop)
    • Top 10 positions by value
    • Total portfolio value
  4. Cross-manager aggregation:
    • For each ticker: count how many managers added / trimmed / hold / exited
    • Surface multi-manager rotations (>=3 managers same direction same quarter)
    • Sector concentration if ticker → sector mapping available
  5. Write per-manager quarterly brief at ~/rdco-vault/01-projects/investing/anchors/smart-money/<manager-slug>/<YYYY-QN>-brief.md
  6. Write aggregate quarterly brief at ~/rdco-vault/01-projects/investing/anchors/smart-money/<YYYY-MM-DD>-aggregate-<YYYY-QN>.md
  7. Cross-ref active theses — for each thesis in theses/, check whether tracked managers are tracking or diverging from the thesis. Flag divergence loud.

Output shape

Per-manager brief

---
date: <run-date>
type: smart-money-brief
manager: <manager-slug>
quarter: <YYYY-QN>
filing_date: <date-filed>
portfolio_value_usd: <total>
position_count: <n>
tags: [investing, smart-money, 13f]
---

# <Manager> — <Quarter> 13F

## Top 10 positions
| Rank | Ticker | Value $M | % Portfolio | Δ vs prior Q |
| --- | --- | --- | --- | --- |
...

## New positions (this quarter)
| Ticker | Value $M | % Portfolio |
...

## Exited (this quarter)
| Ticker | Prior value $M |
...

## Significantly added (>20%)
...

## Significantly trimmed (>20%)
...

## Thesis cross-ref
- Memory: <holding ANY memory names? direction?>
- Power: <...>
- Space: <...>
- Longevity: <...>
- Elon-verse: <...>

## Notes
<freeform read of the manager's quarterly motion>

Aggregate brief

---
date: <run-date>
type: smart-money-aggregate
quarter: <YYYY-QN>
managers_processed: <n>/<8>
managers_pending: [<list of managers not yet filed>]
tags: [investing, smart-money, 13f-aggregate]
---

# Smart-money aggregate — <Quarter>

## Multi-manager rotations (>=3 managers same direction)
### Collective adds
| Ticker | # Managers adding | Managers |
...
### Collective trims/exits
...

## Thesis corroboration scorecard
| Thesis | Smart-money tracking? | Notes |
| Memory | <yes/mixed/no> | <which managers, which direction> |
...

## Concentration shifts
<aggregate sector/factor reads>

## Anomalies + flags
<single-manager standout moves, surprise exits, etc.>

## Idea-generation candidates
<tickers/sectors not in our thesis set that warrant a `candidates/` lookup>

Companion Python helper

~/.claude/scripts/edgar-13f.py — does the EDGAR fetching, XML parsing, and diff math. Skill shells out to it for the data layer; this SKILL.md owns the cross-manager reasoning + thesis-corroboration + brief authoring.

Subcommands (sketch):

Backfill subcommand:

Cross-references to active theses

Each per-manager brief tags exposure to the 5 v1 buckets (memory, power, space, longevity, elon-verse). The aggregate brief includes a Thesis corroboration scorecard that for each bucket flags whether tracked managers are net-long, net-trimming, or mixed.

When a thesis-corroboration flag goes red (managers collectively trimming a name we're bullish on), surface it in the aggregate brief's Anomalies + flags section AND link the relevant thesis file. Founder reads this as a phase-marker check.

Dependencies

Constraints / caveats

Related

Congress tracking (STOCK Act PTRs) — added 2026-05-18

A second data source: congressional periodic transaction reports (PTRs). Higher cadence than 13F (30-45 day disclosure lag vs 13F's 45 days post-quarter-end), captures both sides of the political aisle, and pulls signal from the small subset of members + spouses who are active stock-pickers vs the broad passive-managed-account flow.

Data source

Quiver Quant public per-politician pages at https://www.quiverquant.com/congresstrading/politician/<Name>-<bioguideID>. Each page ships the full per-member trade history embedded in a JS tradeData variable — no auth, structured data, free.

Fallback if Quiver becomes unreliable: official House Clerk + Senate efdsearch sites publish PDFs directly. Scraping them is hairy (PDF parse, no structured records); only fall back if Quiver structurally breaks.

Why not SEC EDGAR: PTRs are NOT filed with the SEC. They go to the House Clerk's office of financial disclosures and the Senate Office of Public Records. EDGAR has no congressional disclosures.

Access rules:

Tracked member set (v1, founder-named + top-volume sweep, 2026-05-18)

Founder-named primaries (high signal density per trade):

Top-volume traders (added from Quiver active-list sweep — lower signal density per trade but useful for cross-member darling-detection):

Signal-density caveat: Pelosi at 21 trades / 12mo has higher density than Khanna at 5,083 / 12mo. Pelosi's trades are large-bracket, directional, and individually attributable. Khanna's are mostly spouse-managed small flow. Always filter to substantive notional brackets ($15k+) before reading trend.

Scraper

~/.claude/scripts/congress-trades-scraper.py (Python; uses ~/.claude/scripts/dram-spot-venv).

Subcommands:

CSV columns: filed_date, trade_date, member, bioguide, party, state, chamber, ticker, action, notional_range, asset_type, instrument, sector, owner, notional_mid_est, description.

Cron cadence

Monthly on the 15th. Each PTR has a 30-45 day filing window after the trade; the 15th sits after most members' month-end filings have closed. Each run pulls trailing 60 days (covers two PTR cycles) and diffs vs prior monthly snapshot.

Cross-source reasoning (TIER-1 rule)

When a name appears as a tracked 13F manager's recent buy AND a tracked Congress member's recent buy in the SAME window, that's a TIER-1 multi-source signal. The intuition: two independent disclosure systems pointing the same direction simultaneously narrows the false-positive surface.

Cross-source overlap analysis lives at ~/rdco-vault/01-projects/investing/anchors/smart-money/congress/<YYYY-MM-DD>-cross-source-overlap.md — produced each time a fresh 13F aggregate and fresh Congress backfill are both available.

Single-source signals are TIER-2 (one source) or TIER-3 (only one tracked entity within one source). Use TIER for thesis-corroboration weighting, not as a buy/sell trigger by itself.

Output

Changelog