Troubleshooting IP Geo-location Discrepancies with BGP Routing Data?

Following up on the basics of IP geo-location, which I'm quite familiar with, I'm now facing significant practical challenges with accuracy for specific IP ranges. We're observing substantial discrepancies, often hundreds to thousands of kilometers off, when comparing geo-located IPs to expected user locations or our known server locations. These issues typically arise with IPs belonging to large enterprise networks, CDNs, or major cloud providers, which often employ complex global infrastructures. Standard commercial geo-IP databases like MaxMind, IPinfo, and DB-IP are consistently reporting incorrect city or region data for these problematic IPs, causing headaches for our regional content delivery and analytics. We've cross-referenced multiple leading geo-IP providers, but the inaccuracies persist across the board for the same problematic IP blocks, suggesting a systemic limitation in how these databases derive their location data for such complex networks. My hypothesis is that these geo-location databases might be relying too heavily on RIR allocations or BGP origin AS information, which, while foundational, doesn't always reflect the actual physical egress point of traffic, especially with intricate network topologies and traffic engineering. A key area of concern is precisely how BGP anycast routing affects geo-location. If an IP prefix is announced from multiple geographically dispersed locations, how do geo-IP services determine the 'most accurate' physical location, or which specific announcement takes precedence in their databases when multiple BGP peering points could be involved? I'm trying to understand how we can effectively leverage raw BGP routing data to enhance or correct IP geo-location accuracy for these challenging edge cases. Are there specific BGP attributes, such as AS path details, BGP communities, RPKI validation status, or even real-time latency measurements to various BGP peering points, that offer more reliable indicators of an IP's true physical location than what standard geo-IP databases typically provide? Furthermore, what programmatic approaches or open-source tools exist for efficiently processing and interpreting this kind of BGP data to refine geo-location data on a practical scale? I'm looking for advanced methodologies or data sources beyond the typical geo-IP APIs to tackle these persistent accuracy issues by diving deeper into BGP dynamics. Thanks in advance for any insights!