Does a DC optimal power flow, made to see network losses through nodal penalty factors, order its generators well? Earlier in infra-dcopf the answer was mixed: factors taken at a DC dispatch helped at light load and hurt at full load (2.24 % above the AC optimum at 1.1 of load, worse than plain DC's 1.68 %).
I predicted the fault was where the loss sensitivities were measured, not the idea. Measured at the AC optimum and applied once, the AC cost of the dispatch on IEEE 118-bus is 0.367 % above the optimum at the case's load (plain DC 1.479 %) and 0.339 % at 1.1 of load (plain DC 1.678 %). At 0.6 and 0.8 of load the two versions agree.
Caveat: this uses the AC optimum, which a DC planner lacks, so it bounds what factors can do and is not a deployable scheme. One solver (pandapower), one dispatch point per cell; case30 and case39 differences are within 0.1 % and not interpretable. Next: factors from a previous-interval estimate.
Code, table and log: research/infra-dcopf/.