The InsiderWire consensus score: How we combine insider, fund, and Congressional signals
The consensus score is the single number that ranks every signal on InsiderWire's wire. It's the answer to 'of all the activity on this ticker, how strong is the cross-source evidence?' Here's how the number is constructed.
The four ingredients
Every consensus score combines four factors:
1. Source quality weight. Each source (SEC Form 4 insider buys, 13F filings, Congressional trades, news/Stocktwits trending) gets a static quality weight in the config — currently 1.0 for Form 4, 1.0 for 13F, 0.85 for Congress, 1.0 for news.
2. Same-source signal count (log curve). If the same source fires N signals on a ticker in the window, we apply a log2(N+1) curve so that 5 Form 4 buys on AAPL score more than 1, but a runaway 50-signal cluster doesn't dominate.
3. Recency half-life. Each signal's contribution decays with age. Default half-life is 7 days: a signal today counts 1.0, a 7-day-old signal counts 0.5, a 14-day-old signal 0.25.
4. Cross-source bonus. When signals from 2+ distinct sources all fire on the same ticker in the same window, we add an extra bonus (+0.5 for 2 sources, +1.0 for 3, +1.5 for 4).
The formula
Per-source contribution: sourceWeight × log2(N+1) × avg_recency_multiplier
Base score: sum of all per-source contributions
Final score: base + crossSourceBonus[distinct_source_count]
All four knobs (source weights, log curve mode, recency half-life, bonus table) live in config.ts and can be tuned without redeploying scoring logic.
Worked example
Suppose NVDA in the last 7 days has: 2 SEC Form 4 buys (today, 3 days ago), 1 Congressional trade (today), 3 news mentions (1, 2, 5 days ago).
Form 4: 1.0 × log2(3) × mean(1.0, 0.74) = 1.0 × 1.58 × 0.87 = 1.37
Congress: 0.85 × log2(2) × 1.0 = 0.85
News: 1.0 × log2(4) × mean(0.91, 0.82, 0.61) = 1.0 × 2.0 × 0.78 = 1.56
Base = 1.37 + 0.85 + 1.56 = 3.78. Three distinct sources → +1.0 bonus. Final = 4.78.
Why not just count signals?
A naive count would treat 5 trending mentions identically to 5 insider buys. Source weighting fixes that. A naive count would also reward stale activity equally — recency decay fixes that. And a naive count would miss the qualitative jump from 'one source talks' to 'multiple sources agree' — the cross-source bonus fixes that.
The result is a number that captures both the depth of evidence within each source and the breadth of agreement across sources, in a single comparable scalar.
Keep reading.
Educational content. Not investment advice. InsiderWire aggregates publicly-available SEC and Congressional filings — past activity is not a guarantee of future returns.