Risk-factor churn across the megacaps: a 3.4x spread
2026-08-18
The same measurement across the largest issuers, latest 10-K against the prior one:
| Ticker | Latest pair | Change ratio |
|---|---|---|
| AAPL | FY2024 to FY2025 | 0.522 |
| GOOGL | FY2025 to FY2026 | 0.477 |
| AVGO | FY2024 to FY2025 | 0.476 |
| MSFT | FY2025 to FY2026 | 0.406 |
| AMD | FY2025 to FY2026 | 0.357 |
| TSLA | FY2025 to FY2026 | 0.327 |
| NVDA | FY2025 to FY2026 | 0.325 |
| META | FY2025 to FY2026 | 0.253 |
| AMZN | FY2025 to FY2026 | 0.155 |
A 3.4x spread, so the metric does not cluster: the information is in the outliers and in each name's own history. Running five to six pairs per name, no other filing in roughly thirty pairs approaches NVIDIA's 2022 rewrite (0.892 with 35 surviving sentences); in this sample that event class has a base rate of one. AAPL runs years of near-boilerplate (0.18 to 0.20) punctuated by discrete revision events in 2021 and 2025; MSFT and AVGO drift upward across years.
What this measures: editing activity, not exposure. A lightly reworded sentence counts as one removal plus one addition, and four new sentences on export controls can matter more than forty reworded boilerplate ones. Both tails are informative. The calm middle is mostly noise, and low churn after years of convergence can mean stable boilerplate rather than low risk.
Computed with the Signal Nodus MCP server: 10-K and 10-Q section
extraction and sentence-level diffs, pinned by accession number, priced per call.
Company lookup is free and needs no key: https://mcp.signalnodus.ai/.
Pricing - More research