AWB Investment Research Strategy

"The Painting" — A progressive, layered approach to finding 10x investments with downside protection. Last updated: 2026-03-11.

1The Goal

2Current State — What We Have

Data Assets

3Storage Strategy — The Single Source of Truth

Database Schema

The existing schema is close to what we need but requires these changes:

-- 1. COMPANIES TABLE (exists, needs population)
--    Import all 8,049 companies from universe.db
--    Add columns for quick filtering

-- 2. NEW: company_metrics — the "painting" table
--    Each row = one metric for one company
--    This is where ALL layers of analysis accumulate
CREATE TABLE company_metrics (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    ticker TEXT NOT NULL,
    cik TEXT,
    metric_name TEXT NOT NULL,     -- e.g. 'agi_score', 'tangible_book_value', 'floor_price'
    metric_value REAL,             -- numeric value (NULL if text-only)
    metric_text TEXT,              -- text value (for reasoning, category, etc.)
    metric_json TEXT,              -- JSON for complex data (timelines, breakdowns)
    source TEXT,                   -- 'agi_scoring_v1', 'yfinance', 'sec_xbrl', 'deep_dive_v1'
    confidence TEXT,               -- 'high', 'medium', 'low'
    computed_at TIMESTAMP,
    cost_usd REAL,                 -- what it cost to compute this
    UNIQUE(ticker, metric_name, source)  -- no duplicates
);

-- 3. NEW: watchlist — the final output
CREATE TABLE watchlist (
    ticker TEXT PRIMARY KEY,
    entry_price REAL,              -- our target buy price
    current_price REAL,
    floor_price REAL,
    bull_case_5yr REAL,            -- 5-year bull case price
    agi_score INTEGER,
    conviction TEXT,               -- 'high', 'medium', 'low'
    thesis TEXT,                   -- 1-2 sentence investment thesis
    last_updated TIMESTAMP
);

What Gets Stored Where

Deduplication Rule

SELECT * FROM company_metrics
WHERE ticker = ? AND metric_name = ?
AND computed_at > datetime('now', '-30 days')

Migration Plan (One-Time)

1. Import 8,049 companies from universe.db into awb.db companies table
2. Import 3,648 AGI scores from agi_scores_all.json into company_metrics
3. Import Leopold portfolio data into company_metrics (as a "smart_money_signal" metric)
4. Import the 10-company floor price pilot results

4The Layers — Progressive Deepening

Each layer adds more detail to our picture of each company. Earlier layers are cheap and broad; later layers are expensive and narrow. We paint the most important parts first.

5The Funnel — From 8,049 to ~10

6Prioritization Logic

Within each layer, we prioritize companies using a composite score:

Special Priority Boost

• Leopold holds + AGI score 8+ — automatic top-50 priority. Smart money + our framework agree = high signal.
• Leopold holds + underwater — automatic top-30 priority. These are his newest, most contrarian bets. If the thesis is right, these are the cheapest entry points.
• Tracked investor overlap — if 2+ tracked investors hold the same stock, priority boost.

7Execution Pipeline — What to Do Next

8Cost Budget

9Phase 3: The Watchlist

Monitoring Cadence

• Daily: Automated price check. Alert if within 10% of entry price.
• Weekly: Quick scan of watchlist. Any material news? Any 8-K filings?
• Quarterly: Re-run financial snapshot after earnings. Update floor price and 5-year model.
• On trigger: When a stock hits entry price, re-read the deep dive. Confirm thesis still holds. If yes, initiate position.

Position Sizing

When we do buy, position sizing follows the floor price confidence:

• Very High confidence floor: Up to 10% of portfolio
• High confidence: Up to 7%
• Moderate confidence: Up to 4%
• Low confidence: Up to 2%

10Operating Principles