I think a new dependency in the way routing nodes are accounting, advertising and verifying their fees must be underscored better before consider Bidirectional as "balance-safe" for intermediary hops.

Type 2 Loop Attacks are usually defined as long-held HTLCs by the final recipient or an intermediary hops thus freezing liquidity across all the links used by the payment path. This attack is made possible due to a fundamental building block of LN, a safety CLTV delta, deduced at each hop to allow in-order settlement of incoming/outgoing HTLCs. And thus avoiding a routing node to severe a balance disequilibrium across its relay links.

This liquidity timevalue is a resource like another and to prevent its abuse, its usage should be accounted as part of a relay cost structure. Along a payment-path, CLTV timelocks are higher on the topologically first links and thus routing fees required by those nodes will be higher.

Within context of Bidirectional, we should expect new hodl_fees to bind to the above routing fees computation, which may hinder a routing node's balance equilibrium.

E.g, with the following topology:

Alice ---> Bob ---> Caroll

Alice-Bob's HTLC nLocktime is superior to Bob-Caroll's one. H.a , "Alice's hodl_fee" required from Bob is superior to H.b "Bob's hodl_fee" required from Caroll.

Bob must correct this fee difference by a) advertise an unconditional fee F such as F - H.a > H.b and b) enforce that any payment routed through him commits to a F satisfying the equation.

It sounds like mitigating against Type 2 with Bidirectional will come with some change in the routing requirements. Routing fees for a given hop must integrate both incoming/outgoing channels' channel_update rather than only the incoming channel_update.

I don't see this as a blocker, but maybe a bit tricky to get it right.