Strange fee calculation

Consider these two commitment TXs generated by LND:

TX 1
Includes HTLC 0 of amount 10,000,000 millsats
Input: 100,000 sats (funding)
Output 0: 10,000 sats for HTLC 0
Output 1: 89,104 sats
fee: 896 sats = Commitment weight: 724 + 172 * num-untrimmed-htlc-outputs

TX 2
Includes HTLC 0 of amount 10,000,001 millsats
Input: 100,000 sats (funding)
Output 0: 10,000 sats for HTLC 0
Output 1: 89,103 sats
fee: 897 sats

The fee for TX 2 is 897 sats when it should be 896 according to the formula.

[Roasbeef]

According to what formula? You need to consider that things like signatures aren't a fixed size and can vary.

[hxsquid]

The formula in Bolt 3:

This yields the following expected weights (details of the computation in Appendix A):
Commitment weight: 724 + 172 * num-untrimmed-htlc-outputs

[halseth]

@RQ732P @hxsquid Are you saying that lnd calculates a fee that is incompatible with the BOLT in these cases? You must remember that any trimmed and rounded outputs will also go towards the fee.

[hxsquid]

@halseth OK, thanks, let me reread the docs on how rounded outputs affect the fee. That must be the explanation because by increasing the HTLC by 1 millisat the fee jumps by 1 sat.