Known Issues
Current known issues and ongoing improvements for the Signal Database.
This page tracks known issues and ongoing improvements for the Signal Database. We update this regularly as issues are resolved and new items are identified.
Last updated: February 18, 2026
For signal availability, delivery schedules, and upcoming deliveries, see the Delivery page.
Open Issues
| Issue | Description | Signals Impacted | Est. Resolution |
|---|---|---|---|
| SEC filing signal accuracy — incorrect classifications and hallucinated details | LLM extraction workflow can misclassify signal subtypes (e.g., labeling a capex decrease as capexIncrease, or intangible asset impairment as goodwillImpairment) and generate speculative detail text not directly supported by the filing. Additionally, the LLM sometimes extracts only a single sentence of evidence, which can lose important context such as directional changes (increase vs decrease). We are implementing a validation layer that cross-checks generated signals against the source filing, tighter classification rules for directional and categorical signals, and more comprehensive excerpt extraction. | 10k, 10q, 8k, 20f, 6k | Feb 28 |
| Job title normalization | Improving disambiguation for ambiguous titles like "CMO" (Chief Marketing Officer vs Chief Medical Officer) based on company context and industry. | 10k, 10q, 8k, 20f, 6k, earnings-transcripts | Feb 6 |
| LinkedIn signals could surface more useful context | Posts currently only have basic tags; need richer metadata to create more useful signals. Sentiment, relationship (like commenter to poster), etc. | linkedin-post-company, linkedin-post-contact, linkedin-comment-contact | Feb 6 |
| Schema inconsistencies | Some fields have similar names with subtle differences (e.g., relevance vs sales_relevance). API vs Signal Database payloads have structural differences. Refer to signal-specific schema docs. | All signals | Ongoing |
Resolved
For a full history of changes, see the Changelog.
| Issue | Description | Signals Impacted | Resolved |
|---|---|---|---|
digitalTransformation subtype deprecated | Subtype was producing false positives in SEC filings and has been removed from all deliveries. | 10k, 10q, 8k, 20f, 6k | Feb 3 |
| Historical backfill timestamps | SEC filings and earnings transcripts from the January 2026 historical backfill may show detected_at timestamps from the backfill window rather than the original filing/event date. Going forward, detected_at will accurately reflect batch execution time. | 10k, 10q, 8k, 20f, 6k, earnings-transcripts | Jan 31 |
| Healthcare/pharma executive misclassification | Chief Medical Officer transitions were incorrectly classified under cmoChange. New subtype chiefMedicalOfficerChange added for proper classification. | 10k, 10q, 8k, 20f, 6k | Jan 31 |
| Social signals limited to 30 days | Current social signal data only covers past 30 days; 3-month historical backfill added. | reddit-company, linkedin-post-company, linkedin-post-contact, linkedin-comments-contact | Jan 23 |
| Global distribution gaps | Sample data skewed US-heavy; international representation added. | linkedin-post-company, reddit, glassdoor, linkedin-post-contact, product-reviews | Jan 23 |
| 8K noise reduction | Some companies file high volumes of 8Ks. Refined processing to surface actionable signals while filtering routine/administrative filings. | 8k | Jan 22 |
Questions?
If you encounter issues not listed here or need assistance, contact us at [email protected].
Updated 1 day ago