Trust center / methodology
Quality is a recorded outcome, not a universal promise.
GetMD uses staged processing, bounded adapters, deterministic contracts, visible warnings, and fixture-based evaluation to make source transformations inspectable.
Acquisition and normalization
Each source adapter validates its contract and hard limits. Acquired content is normalized into a source model before profile-specific rendering.
Provenance
Jobs, source records, file checksums, output manifests, retrieval timestamps, and warnings form the trace between input and artifact.
Quality gates
Fixtures cover success, boundary, failure, and abuse cases. A disposition can accept, warn, or reject an output. Pipeline completion alone is not acceptance.
Benchmark publication
No cross-domain accuracy percentage is published yet. Benchmarks will be released only with a named fixture set, scoring method, sample size, date, and reproducible command.
Known limitations
Source truth is not independently guaranteed. Summaries compress detail, normalization can alter layout, URL acquisition depends on policy and availability, OCR quality varies, and agent context does not authorize downstream action.