Postoperative pneumonia synthesis requires compatible outcomes, adjustment sets, and independent populations. This study examines how eligibility and modelling decisions affect the interpretation of perioperative vulnerability. A targeted, non-exhaustive review used 159 references and an inventory of 58 candidate reports. Four hip-fracture cohorts contributed to exploratory meta-analyses using restricted maximum likelihood and modified Hartung–Knapp inference. Nine analyses were checked using two Python implementations; eight leave-one-out analyses and twelve crude-to-adjusted comparisons were examined. Eighteen AI-assisted QUIPS domain judgments covered three cohorts; independent human reconciliation was not completed. The quantitative dataset comprised 6223 patients and 461 pneumonia events. Under permissive eligibility, the adjusted odds ratio for age was 1.47 per decade (95% confidence interval 1.06–2.02). Strict eligibility retained two cohorts, below the protocol-specific pooling threshold. Removing Tian from the unadjusted male-sex analysis changed the pooled odds ratio from 1.33 to 1.02 (0.50–2.07). The implementations agreed within 0.000001. The audit identified discordant matched-sample denominators in Shin. Other complications remained separate outcomes. Numerical reproducibility does not establish absence of bias. The inventory does not establish 57 independent eligible cohorts, and these findings cannot be reported as a completed systematic review.