What is the HoldLens ConvictionScore?
One signed score per stock on a single scale: +100 is the strongest possible buy, −100 the strongest possible sell.
The problem with dual buy / sell rankings
Every 13F aggregator ranks "top buys" and "top sells" as two independent lists. The naive version counts managers: META bought by 9, sold by 5. That's #1 on both lists — which makes the signal meaningless. If a stock is the strongest possible buy AND the strongest possible sell, the algorithm is measuring trading volume, not conviction.
HoldLens fixes this by collapsing buys and sells into ONE signed number per stock. A ticker appears on exactly one list — buys if its score is above zero, sells if below. META's 9 buyers slightly outweigh its 5 sellers, so it shows up once, on the buy side, at a moderate positive score that reflects the contested nature of the stock.
The seven signal layers
Every stock is scored by seven positive layers minus two penalty layers. The first six are time-decayed across 8 quarters of 13F data; the seventh reads SEC 8-K material events from the last 90 days. Recent moves count more than older ones at every layer.
As of v5 (April 2026), the model reads from all three SEC filing surfaces — 13F holdings (45-day lag), Form 4 insider trades (T+2 lag), and 8-K material events (T+4 lag) — completing the SEC Signals trilogy. No other public investing tool synthesizes all three into a single score.
- Smart money. Manager quality × consensus. A Kantesaria buy (18.7% CAGR track record) weighs more than a Pabrai buy.
- Insider activity. CEO or CFO open-market buys — the strongest single equity signal there is. Routine 10b5-1 sells don't count against. Reads from /insiders/ (curated + 3,001 EDGAR Form 4 transactions).
- Track record. Each buyer's realized 10-year CAGR weighted by their position size in this particular stock.
- Trend streak. Multi-quarter compounding. 3 quarters in a row of adding ≠ a single-quarter buy.
- Concentration. A 15% position is a conviction bet. A 1% position is portfolio filler. The model weights them accordingly.
- Contrarian bonus. Under-the-radar stocks with tier-1 buyers get extra credit — that's where alpha lives, not in the already-crowded names.
- Event signal (v5). SEC 8-K material events from the last 90 days adjust the score asymmetrically: bankruptcy (Item 1.03) subtracts 15 points, restatement (Item 4.02) subtracts 12, impairment (Item 2.06) subtracts 10, delisting (Item 3.01) subtracts 10. Positive events (officer promotion, positive Reg FD disclosure) add up to 5 points. The asymmetry is intentional — one bankruptcy correctly overrides 30 superinvestor buy signals. Reads from /events/.
- − Dissent penalty. Every seller subtracts from the score, weighted ×1.6 because exits require more conviction than trims.
- − Crowding penalty. When 15+ managers already own a stock, the signal is already priced in. The model discounts it.
Reading the number
The score is signed, so the number itself tells you everything: direction by the sign, strength by the magnitude. Labels are cosmetic — they don't affect which list a ticker appears in, only how strong the signal reads.
- +70 → +100 — STRONG BUY. Unambiguous tier-1 consensus with trend + insider support.
- +40 → +70 — BUY. Clear positive signal, solid manager backing.
- +10 → +40 — WEAK BUY. Positive signal but muted by dissent or crowding.
- −10 → +10 — NEUTRAL label, still appears on buys or sells by sign.
- −10 → −40 — WEAK SELL. Negative signal but not catastrophic.
- −40 → −70 — SELL. Clear distribution from tracked managers.
- −70 → −100 — STRONG SELL. Multi-quarter exit pressure from elite managers.
Why the score tells the truth on Buffett
The older HoldLens model assigned Buffett a hand-curated quality score of 10. The unified ConvictionScore uses Berkshire's realized 10-year CAGR instead. Over 2015–2024, BRK trailed the S&P 500 — so Buffett's computed quality score is 5.9, not 10. A Buffett buy now weighs less than a Kantesaria buy (quality 8.2) because the last decade of numbers backs it up. Reputation doesn't flatter the score — track record grounds it.
See it live
The score is already running. Browse the ranked lists, the per-ticker dossier, or the manager leaderboard to see it in action.
Where this scoring model came from
Conviction scoring borrows from three traditions: Graham's margin-of-safety, Buffett's circle-of-competence, and modern factor research. These books are the primary sources.
Amazon affiliate links — HoldLens earns a small commission if you buy, at no extra cost to you. These are the books we actually recommend. Always do your own research.
The full seven-layer formula lives in the project repo and is described in detail on the methodology page. Not investment advice.