Skip to main content

v4.27

Release Date: September 17, 2026

Highlights

  • MCP Updates: We have added new text-to-insight optimizations. Queries are now returning in 5–7 seconds (down from 17–22s), ~3.2x faster, with no quality regression and improved insight selection.
  • Lenses Model Change: Lenses is enhanced and now runs on a mix of models.
  • Website: Refreshed data links and navigation, plus an improved view of privacy-policy tables.
  • New Data Releases: Mobile App data (CA0013) now available in Block, Securities Ratings (CA0092) available on Platform.
  • Pipeline Enhancements: Credit Card – Health Spend (CA0058) and Card – US Complete Panel (CA0056) underwent pipeline updates to improve stability and speed.

Key Releases

Text-to-Insight Tool Optimizations

We have added new text-to-insight optimizations. Queries are now returning in 5–7 seconds (down from 17–22s), ~3.2x faster, with no quality regression and improved Insight selection.

Mobile App data (CA0013) now in Block

Mobile app data from an opt-in panel of approximately one million users, covering downloads, active users, revenue, and usage across 4,500+ apps and 300+ tickers. Useful for tracking adoption velocity, engagement, and app-level competitive position — now available in Block. Please feel free to reach out to support@carbonarc.co with any questions related to access.

Securities Ratings (CA0092) now in Platform

Daily 1 to 5 investment ratings for roughly 5,200 US-listed stocks, 1,140 ETFs, and 7,610 mutual funds, derived from footnote-level analysis of company filings. Useful for quality and valuation screening, accounting-quality overlays, and fund due diligence — now available in Platform. Please feel free to reach out to support@carbonarc.co with any questions related to access.

Card – US Complete Panel (CA0056) Pipeline Enhancements

The CA0056 pipeline is now faster and more stable. Each transaction is now assigned to a single retailer category, so brand-category spend can read up to ~60% lower on some tables while transaction counts stay flat; that gap is double counting coming out, not missing data. Categories are assigned brand-first, with MCC used only for untagged transactions. Company categories have been removed. Distinct card users is now an approximate count, within roughly 1% on large counts and exact on small ones, while spend and transaction counts remain exact.

Credit Card – Health Spend (CA0058) Pipeline Enhancements

The CA0058 pipeline is now faster and more stable. Categories were cleaned so each transaction maps to a single category, removing double counting that previously inflated some category views. Spend and transaction counts are unchanged — if category totals look lower, that is double counting coming out, not missing data. Company categories have been removed.

Lenses Model Change

We are updating the underlying LLM model for Lenses from Sonnet 4.5 to a mix of models (primarily an upgrading to Sonnet 5). We expect MCP server users to see improved output quality, and will continue evaluating opportunities to improve our MCP server experience.