Emergent sacrifice in a sim: loyalty as an objective weight + Hamilton rule

Sacrifice (acting against your own material interest for another) CANNOT emerge from a self-interested maximizer — so a sacrifice rule is always enforcement. The emergent way: make loyalty a WEIGHT the agent puts on the other in its OWN objective (alpha in U_i = u_i + alpha*u_j, other-regarding preferences). Then interposing/self-sacrifice falls out of Hamilton rule: do it iff alpha*harm-to-them > cost-to-self. Key details that make it work and stay emergent: (1) store alpha as a directional belief so it is LEARNED — grows by prediction error from shared adversity survived together, exceeds 1 for kin (you value them above yourself), fades in peace via revert-to-prior. (2) Value the harm/cost in survival terms (fraction of life lost), so a strong ally shielding a weakling pays little (cheap heroism, low alpha suffices) but an equal comrade taking a fatal blow pays ~1, so only alpha>1 (kin/deep bond) triggers true death-sacrifice. (3) A grievous raider that crushes whoever it hits makes cost~1 for everyone, so only alpha>1 sacrifices — clean threshold. Contrast: same monster, bonded pair produces a sacrifice, strangers do not — purely from what each lived through.