The mechanism at the core of large language model technology — statistical token prediction via transformer-based matrix multiplication — does not structurally support the characterisations of reasoning, comprehension, or knowledge retrieval under which these systems are commercially offered and institutionally adopted; this gap is the governance gap this paper addresses. Three evidence streams document it: peer-reviewed literature establishing architectural limitations, a practitioner field record of reproducible performance failures spanning 2024–2026, and system-sourced testimony confirming the technology's operational ceiling in its own terms. In low- and middle-income country health systems, four structural conditions compound simultaneously — training distribution boundary, cognitive fluency effect, verification capacity gap, and regulatory gap — producing a risk profile highest precisely where health system vulnerability is greatest. Six recommendations are directed at mechanism-layer governance: mandatory mechanism disclosure at procurement, calibrated uncertainty signalling, verification cost inclusion in productivity assessments, retirement of hallucination rate as a safety metric, LMIC-specific performance requirements, and alignment of the Independent International Scientific Panel on AI with mechanism-layer health risk.

To align health benefit package design with post-disbursement, this framework pools payer, provider, and consumer records at the episode-of-care level, generating structured evidence for benefit package revision cycles. It unifies four elements into one architecture: episode of care as unit; payer, provider, and consumer as sources; population, service, and cost as pooled dimensions; and a combined record as output. Records are linked through three methods: direct linkage using shared identifiers between payer and provider; composite matching on visit date, facility, and service category where no identifier exists; and administrativeboundary matching where identifiers are unavailable. Each method draws on independent precedent; this framework proposes combining them into one three-party post-disbursement coverage record. The record establishes three fixed correspondences linking each dimension to its governance lever: population coverage to entitlement administration, service coverage to provider arrangements, and cost coverage to reimbursement-enabling those units to act on the evidence. The same unit, sources, and dimensions read whatever content a tax-funded, social insurance, national insurance, or out-of-pocket-dominant system generates in its institutional form, requiring no new data collection beyond what the revision cycle already convenes, and is operationalised through a ministry's monitoring function, Joint Learning Network technical assistance, or research-partner arrangements. Linkage completeness is reported alongside the record, indicating how fully existing records can be linked-both a baseline of current capacity and a target for future strengthening. This record enters the same policy dialogue step as an add-on layer of data and evidence for the benefit package revision cycle that countries convene.

The WHO universal health coverage cube separates population, service, and cost coverage, yet at the point of care the three arrive as one episode: an entitlement, a service, and a claim met from public funds. That episode needs a unit; hospital cost studies each carry one dimension, in different cost units; this paper proposes the episode-of-care window, with a prespecified index episode and accumulation window, as the unit of analysis. Within the window, each dimension is operationalised as a vector: Entitlement for population (eleven variables, from age to days in hospital), Service Delivery for service (six cost categories, kept separate), and Reimbursement for cost coverage (insurance status, a binary non-cost variable). Because the dimensions arrive together, the three vectors enter one Gamma GLM with a log link to Direct Medical Costs, the episode's total cost. The exponentiated coefficients form the health cost driver map: the weight each variable carries within one episode. Applied to 125 households' episodes, procedure costs carry the largest z, with pharmacy, ICU, test and laboratory costs, disease type, prior disease, days in hospital, and discharge type significant; consultation fee and insurance are not. A separate robust run on the Entitlement and Reimbursement vectors is reported. The map is offered to ministries, insurers, and international bodies as a shared place to stand, informing their conversation without settling it, with each country bringing its own content. Held together at the point of care, the three dimensions of the UHC cube yield one map each institution can draw.