Skills-aware talent recommendation draws on three families of signals: collaborative behavioral regularities, adaptive trajectory dynamics, and interpretable occupation-side criteria; yet public benchmark evidence on their complementarity is scarce. This study introduces CF-RL-TOPSIS, an interpretable late-fusion architecture that combines a transition-aware collaborative branch, a lightweight reinforcement-style bandit over occupation families, and an entropy-weighted Technique for Order Preference by Similarity to Ideal Solution branch built from six semantic proxies, with validation-selected fusion weights that remain directly inspectable. Evaluation proceeds on two frozen public information and communication technology talent-history benchmarks, JobHop and Karrierewege, under repeated chronological top-5 ranking with paired Wilcoxon testing. On JobHop the full hybrid attains NDCG@5 = 0.3040 ± 0.0073 and significantly surpasses repeat-last, item Markov, transition-aware collaborative filtering, the CF+TOPSIS hybrid, GRU4Rec, and SASRec (p ? 0.0039 across planned comparisons). On Karrierewege the hybrid remains competitive but does not significantly exceed the strongest Markov baseline, exposing a persistence-dominated regime in which the bandit branch correctly collapses to near-zero weight. Proxy-sensitivity, family-level deep Q-network, and runtime checks corroborate this reading, and a worked user-level case illustrates branch-score, criterion-weight, and rank-shift inspection for an individual recommendation. The contribution is therefore not a benchmark-agnostic superiority claim but a reproducible account of when an interpretable late-fusion architecture meaningfully complements simple continuation heuristics. In semantically rich, non-saturating talent-history regimes the three branches reinforce one another; in persistence-dominated regimes the same architecture remains competitive through its collaborative backbone, with the adaptive branch correctly inactive.