Wedge100BF-32X on SONiC: 100G LR4 links back to back but not through the transport unless TX is patched first

We are handing a 100G circuit off to a core router through a leased optical transport system, and the port on our side refuses to come up. The same pair of modules links instantly on the bench, which is what makes this so irritating.

• Wedge100BF-32X (wedge100bf_32x) running SONiC
• InnoLight TR-FC13L-N00, 100G LR4 QSFP28, 10 km
• a Juniper BTI transport system between us and the core router

admin@sonic:~$ show interfaces status
  Interface    Speed    Admin    Oper
  Ethernet0     100G       up    down

What we have gone through:

• the same two modules connected back to back on the bench come up and stay up, no complaints from either side
• swapped the module on our side and replaced both patch cords
• the transport people confirm they see light from us on their ingress port

The one thing that does work: if we pull both strands and put them back in sequence, TX first, wait a few seconds, then RX, the port comes up and then stays up indefinitely. That is not something I want written into a runbook for a production handoff, and it does not survive a maintenance window on the transport side either.

So what is going on in those few seconds? It reads like both ends are sitting there waiting on each other's fault signalling and neither one starts transmitting, but I have no way to see that from this platform. Has anyone chased this down on a Wedge running SONiC against a transport handoff, and is there anything tunable on the switch side?