Spot on @User. There is a second-order effect to certain features (specifically, ones related to cross cutting concerns). If we lock ourselves into the backward compatibility fallacy of minimizing risk to LinkedIn deployments, then we're also accepting the fact that the reason for this risk is also what caused it in the first place. I would add contrast to @User's point that breaking backward compatibility should be avoided at all costs. If the platform was built under the right constraints, then there would be less fragility in upgrades. Ultimately it is LinkedIn's responsibility to canary each upgrade so that their rollouts minimize the risk of a widespread impact to users. But more importantly, to @User's point, it's not about breaking changes... the challenge is making the risk an explicit cost to an operator that is performing the upgrade.