Does ExpressRoute Global Reach guarantee primary↔primary / secondary↔secondary MSEE pairing between two circuits?

We're modeling ExpressRoute topology for network discovery/visualization and need to understand the deterministic behavior of Global Reach across redundant links.

Setup: Circuit A and Circuit B have Global Reach enabled between their private peerings. Each circuit has the standard redundant pair — a primary link and a secondary link, each terminating on a different MSEE router, both active simultaneously (active-active, per Designing for High Availability).

Question: When a packet arrives from on-prem Site A at Circuit A's primary MSEE link and needs to cross via Global Reach to reach on-prem Site B behind Circuit B:

1. Does it always exit via Circuit B's primary link — i.e., is primary↔primary / secondary↔secondary pairing an architecturally guaranteed behavior across the Global Reach connection?
2. Or is this just the current/typical outcome of Microsoft's backbone routing, with no guarantee — meaning it could exit via either of Circuit B's links depending on internal ECMP/path selection at the time?
3. If it's not guaranteed: is this pairing stable in steady state but subject to change during planned maintenance (e.g. AS-path prepending described in the FAQ), or can it shift at any time without a maintenance trigger?

Why we're asking: We've found one independent (non-Microsoft) technical write-up (cloudtrooper.net, 2021) showing primary↔primary and secondary↔secondary iBGP peering between two circuits' MSEEs in a lab test, but no official Microsoft documentation confirms this is guaranteed rather than incidental. We want to know whether we can safely model a 1:1 link pairing for topology/path visualization, or whether we should instead represent both of Circuit B's links as equally valid candidate paths for traffic arriving via Global Reach.