Can DI find the right entities?
Measure precision@k, recall@k, synonym/abbreviation resolution, no-result quality, geographic and specialty coverage.
GLOBAL BENCHMARK LAB™
Do not call it world-leading until the benchmarks can show it.
V9 defines the benchmark suite Disease Intelligence should publish as the platform matures. No synthetic number below is presented as a real performance result.
Are relationships correct and explainable?
Sampled edge accuracy, ontology consistency, duplicate resolution, shortest-path validity and graph update integrity.
Can an assertion be independently checked?
Provenance completeness, freshness, conflict detection, source quality and reviewer agreement.
Does DI reduce operational friction?
Time-to-route, time-to-acknowledge, unresolved steps, task completion and handoff failure rates.
Does the system behave appropriately under uncertainty?
Abstention quality, unsafe suggestion rate, escalation success, explanation comprehension and adverse-event review.
Can the product scale efficiently?
Infrastructure cost per active user/query, support cost, gross margin, payback, retention and expansion.
Publication rule: benchmark methodology, denominator, dataset scope, dates, limitations and confidence intervals should accompany any future performance claim.