SEC Redline · Structural Diff
A 10-K is two hundred pages, of which one hundred ninety-four are identical to last year. The signal lives in the six that changed: the risk factor that appeared, the single-source foundry warning, the segment revenue breakdown. Convexity redlines filings sentence-by-sentence and grades every 8-K on statutory severity.
− We depend on a limited number of third-party foundries to manufacture our products.
+ We depend on a single third-party foundry for substantially all of our advanced products, and we have no qualified second source.
+ New or expanded export controls could restrict our ability to sell our most advanced products into certain sovereign markets.
− Pandemic-related disruptions may continue to affect our supply chain and logistics costs.
Keyword search locates sentences you already knew to ask for. It cannot alert you that a key sentence was silently altered between quarters. Convexity indexes filings into structured sections, performing deterministic pairwise sentence diffs against prior periods.
Every 10-K, 10-Q, and 8-K is parsed into structural components rather than stored as an unindexed blob. Sections are isolated by Item headings, tables extracted with document ordinals, and footnotes preserved.
Each filing is diffed against its direct predecessor sentence by sentence: added, removed, and modified. No generative model touches the diff, eliminating hallucinated edits.
An 8-K is graded by its statutory item code, not by PR headlines. A financial restatement (Item 4.02) and an earnings release (Item 2.02) represent entirely different market events.
A 10-K or 10-Q is not a flat document. Diffing whole filings creates noisy, unusable diffs due to signature blocks, pagination changes, and exhibit lists. Convexity locates strict SEC Item boundaries, comparing apples to apples across periods.
When an Item heading cannot be located in both filings, that section gracefully returns empty rather than diffing mismatched sections into false positive noise.
Important segment breakdowns are frequently buried across hundreds of pages of raw SEC HTML tables, mixed in with decorative layout tables and signature blocks.
Convexity filters tables by numeric density and statutory section headers, extracting tabular data with clean headers, row hierarchies, and permanent document ordinals for unambiguous provenance.
| Segment | FY25 | FY24 | YoY Delta |
|---|---|---|---|
| Data Center Infrastructure | $79,600M | $15,005M | +430.5% |
| Gaming & AI PC | $11,350M | $9,067M | +25.2% |
| Professional Visualization | $1,896M | $1,553M | +22.1% |
| Automotive & Robotics | $1,091M | $903M | +20.8% |
Companies file Form 8-K for routine earnings announcements and catastrophic financial restatements alike. Convexity evaluates the exact SEC item code to route alerts by true material consequence.
Comparing long financial documents via generative AI prompts introduces probabilistic drift and hallucinations. Convexity employs deterministic sequence matching over parsed sentence nodes.
Model memory or truncated chunks that fit the context window.
Two stored EDGAR XML files diffed sentence-by-sentence via sequence matcher.
Can generate plausible-sounding edits that never occurred in the text.
Zero risk. Pure deterministic text diff; nothing in the pipeline writes prose.
Two requests yield different summaries depending on temperature.
Byte-identical output across runs. Ground truth is purely mathematical.
Hallucinates an answer even when section structures diverge entirely.
Detects structural re-organizations, flags non-comparability, and gives the cause.
Sentence-level diffs across five key Item sections, structured table extraction, and 8-K severity grading, built into every security dossier.
Available on Pro and above. Founder pricing locked for life.