Emotional support conversation (ESC) requires responses that recognize users’ feelings and situational meaning while balancing relational support with practical orientation. Yet it remains unclear whether improvements from synthetic post-training reflect structured empathy signals or merely stronger rewritten responses. We introduce structured empathic supervision, representing each turn with Feeling/Meaning understanding targets and Subjective/Interpersonal/Objective (S/I/O) response plans. On a coverage-oriented diagnostic suite of 219 turns from 20 ESConv dialogues, Full-SIO-SFT significantly outperforms a decoding- and budget-matched Rewrite-SFT control on aggregate support quality, perceived empathy, counseling quality, and cognitive depth. Controlled perspective analyses show target-aligned specialization: SI favors relational support, whereas IO favors informational support. Starting from Full-SIO-SFT, balanced DPO further improves the aggregate six-dimensional support objective, while directional objectives induce distinct support profiles. Blinded comparative ratings from two psychology-background annotators corroborate the structured-supervision gain and the directional profile shift. These results support structured supervision and preference weighting as complementary mechanisms for controllable ESC post-training.