How we calculate everything

Data sources

Every position on HoldLens comes from SEC EDGAR 13F filings — quarterly disclosures required from any institutional investment manager with over $100M in assets under management. We supplement with Form 4 (insider trades) and Form 13G/13D (large position disclosures).

• • 13F-HR: Long US equity positions, filed within 45 days of quarter end
• • Form 4: Insider buys + sells, filed within 2 business days
• • 13G/13D: Large position changes (5%+), filed in real time

The 45-day lag (and why it matters)

13F filings are due 45 days after each quarter ends. Translation: when you see a position on HoldLens, the actual buy/sell happened 6 weeks to 4 months ago. We never pretend otherwise. HoldLens is for pattern recognition, not copy-trading.

Unified ConvictionScore (v4)

Every ticker is assigned ONE signed score on a single scale: +100 is the strongest possible buy, −100 the strongest possible sell, and zero is no signal. A stock can appear on EXACTLY ONE list (buys or sells) — never both — because both lists are filtered views of the same single number.

The score is built from six positive layers minus two penalty layers:

• • Smart money — manager-quality × consensus, time-decayed across 8 quarters of 13F data
• • Insider activity — CEO/CFO open-market buys (the strongest single equity signal). Routine 10b5-1 sells don't count against
• • Track record — buyer 10-year CAGR weighted by their position size in this stock
• • Trend streak — multi-quarter compounding (3 quarters in a row ≠ 1 quarter)
• • Concentration — a 15% position is weighted heavier than a 1% position
• • Contrarian bonus — under-the-radar stocks (small ownership count + tier-1 buyers)
• • − Dissent penalty — sellers subtract from the score (×1.2 because exits require more conviction than trims)
• • − Crowding penalty — when ownership count is high, the signal is already priced in

Pure sign-based: a ticker's ranking membership is determined entirely by the sign of its single signed score. Positive → buy ranking. Negative → sell ranking. Zero → neither. No dead zone. No third bucket. The same number tells you everything: direction by its sign, strength by its magnitude.

META used to be #1 on both rankings under the old dual-list scheme. Under the unified score, META has ONE conviction value (positive, ~+20) — its 9 buyers slightly outweigh its 5 sellers — so it appears in EXACTLY ONE list (buys), with a moderate score that reflects the contested nature of the stock.

How a label maps to the score

Labels are purely cosmetic — they describe how strong a signal is, but don't affect which list a ticker appears in. Only the SIGN of the score does that.

• STRONG BUY — score ≥ +70
• BUY — score ≥ +40
• WEAK BUY — score > +10
• NEUTRAL — score in [−10, +10] · still appears on buys or sells based on sign
• WEAK SELL — score < −10
• SELL — score ≤ −40
• STRONG SELL — score ≤ −70

What we don't show

• ❌ Short positions (not in 13Fs)
• ❌ Options exposure (notional only, no detail)
• ❌ Non-US equities (not in 13Fs)
• ❌ Bonds, crypto, real estate (not in 13Fs)
• ❌ Real-time positions (legal floor: 45 days)

If you need real-time, options-aware data, you need a Bloomberg terminal — or more accurately, you need to be at one of these funds. HoldLens is the best public data made beautiful.

Updates

Manager portfolios update within hours of each 13F filing. Email subscribers get a one-line move alert per filing. We never delete historical data — every quarter is preserved.

Predictive validity — what our 2026 backtest found

Why the score still matters

HoldLens is a smart-money positioning tracker, not a return predictor. The ConvictionScore is a clean composite of what the tracked portfolio managers are actually doing in their most-recent 13F filings — consensus, concentration, multi-quarter trends, insider alignment, dissent. That is legitimately useful market intelligence (people want to know what Buffett, Ackman, Burry and Druckenmiller are buying and selling).

What the score is not: a reliable guide to which stocks will outperform. Three structural reasons the backtest data points at:

• • Contrarian inversion. When tracked managers BUY a stock, it's often because the stock dropped and they're bargain-hunting; the drop continues (momentum). When they SELL, they're often taking profit on a winner that keeps winning.
• • Manager-quality drag. Several storied managers tracked on HoldLens have materially underperformed the S&P over the recent 10-year window (see per-manager ROI panels on the investor pages). Their picks drive BUY signals; their picks have underperformed.
• • 45-day filing lag. By the time we surface “smart money buying X”, the news is usually priced in.

Reproducing the backtest

The backtest script lives at scripts/backtest-conviction.ts. It pulls 2-year daily closes from Yahoo Finance for every tracked ticker + SPY, replays the ConvictionScore at each of the last 4 historical quarters, pairs it with forward return from the 13F filing date to today, and computes Pearson correlation + decile alpha spreads. Full output is logged to .claude/state/CONVICTION_BACKTEST.md. We re-run this every quarter — transparency over flattery.

Errors and corrections

Found a wrong number? Email hello@holdlens.com. Corrections logged publicly with a timestamp. Trust is the moat.