What to Ask an Optical Crosslink Vendor About Failure, Not Throughput
An optical inter-satellite terminal rarely fails all at once. It degrades: a pump diode loses output, a fine steering mirror runs out of stroke, or the coherent DSP latches up after a heavy ion strike. So the question that matters on a shortlist call is not peak data rate but what the terminal, and the constellation around it, does in the minutes after one of those parts misbehaves. Each question below comes with the technical risk it exposes when the vendor's answer is vague.
Where the redundancy actually lives
Ask which elements are cold spared, which are hot spared and which are single string, and get the answer at component level, not box level. Many terminals duplicate the seed laser and the amplifier pumps but carry a single coarse pointing assembly, a single telescope and a single modem ASIC. The redundancy diagram then looks generous while the dominant mechanical and radiation risks sit in one path. The risk exposed is a terminal whose failure probability is set by the gimbal or the DSP, while the sparing budget went to parts that degrade gracefully anyway. Follow up by asking whether a failing pump is detected and switched autonomously, because a weakening amplifier shows up first as shrinking link margin, which a ground team can easily misread as pointing trouble.
Losing the partner, then finding it again
The failure that hurts operations most is a link drop during a pointing transient, such as a reaction wheel disturbance or a thermal snap at eclipse exit, that pushes the beam outside the fine steering mirror's range and forces the terminal back into spatial acquisition. Ask for the reacquisition time from a warm state, meaning valid ephemeris and attitude knowledge on both ends, and ask whether it was measured in a closed loop testbed with representative jitter or only computed. A credible figure is under 30 s. Anything that needs a ground scheduled acquisition window turns a brief glitch into an outage lasting a sizeable fraction of an orbit. Check whether reacquisition is autonomous and beaconless, since a separate beacon laser is one more single point of failure, and whether either terminal can start it without a fixed master and slave role that leaves one side waiting on a partner that has died.
Degraded modes versus binary links
Ask what the modem does when margin falls: if it holds the nominal rate until the frame error rate collapses the link, you are buying a terminal with only two states, full capacity or nothing, instead of one that steps down to a more robust modulation and coding mode and keeps routing control traffic alive.
Interoperability as a resilience feature
In a mesh, a failed terminal is survivable only if traffic can be rerouted through neighbours built by someone else. So ask which interoperability standards the terminal is tested against, typically defence crosslink specifications such as the SDA optical terminal standard, and whether that testing used a different vendor's hardware or just a loopback against a sibling unit. The risk is subtle: two terminals can both claim compliance and still fail to close a link because their acquisition sequences or timing assumptions differ in places the standard leaves open. My estimate is that this kind of integration mismatch, not hardware failure, causes most early on orbit link unavailability in mixed constellations. The decision rule I apply is simple: if the vendor cannot show a cross vendor link closed in a test campaign, with reacquisition demonstrated after a deliberate drop, treat the terminal as single vendor and size routing redundancy as if every plane depends on one supplier. That leaves the real trade open, because an extra crosslink per satellite buys path diversity that no amount of internal sparing can replace, and it costs mass, power and pointing accommodation that the bus team will fight over.