Glossary

The terms these pages use, in alphabetical order. Each links the page that covers it in depth.

Archetype

The design rules every Cosyte parser shares: a lenient reader that turns recoverable deviations into warnings with stable codes rather than failures, a conservative serializer that writes spec-clean output, frozen models, and vendor behavior described in profiles. @cosyte/synth is a generator, not a parser, and keeps the rules that apply to one: its results are frozen, its errors carry stable codes, and its quirks come from the parsers' profiles. How it works explains each.

Artifact

One generated fixture as a corpus records it: its format, its kind, its wire text in content, and the codes its parser reported reading it back in warnings. See Artifact.

Corpus

A frozen set of artifacts generated from one seed, with a manifest that lists the formats, the count of each kind, and the quirks applied. Each format has one (hl7Corpus, fhirCorpus, and the others), and the formats with quirks have a quirk corpus too. Regenerating from the same seed, on the same package versions, gives the same artifacts. See Generator options, kinds and profiles.

Floor

Two things on these pages are called a floor.

• A synthetic-safety floor is the source a kind of identifying value is drawn from, chosen so the value cannot be a real one: SSNs in the 900 series, phone numbers from 555-0100 to 555-0199, NPIs that fail their check digit, and the others. The floors lists each, with the authority behind it.
• A quality floor is the lowest coverage or mutation score CI accepts, recorded in the repository's quality-floor.json. It only rises. See Testing and quality.

Intended-warning contract

The promise every quirk fixture keeps: a parse with no profile raises exactly the warning codes its quirk names, no more and no fewer. Each quirk generator parses its own output and throws SYNTH_INTENDED_WARNING_MISMATCH rather than return a fixture that breaks it. A quirk round trip reports it as intendedWarningHeld.

Kind

The label of what an artifact is within its format, such as ADT^A01, USCorePatient, ccd, 837P, B1 or Result. A corpus mix is a list of kinds. Kinds lists every one.

Peer

One of the sibling Cosyte packages @cosyte/synth builds through: @cosyte/hl7, @cosyte/fhir, @cosyte/ccda, @cosyte/x12, @cosyte/ncpdp, @cosyte/astm, and @cosyte/deid. Each is an optional peer dependency, needed only for its subpath and loaded the first time a generator uses it. A missing one throws SYNTH_PEER_NOT_INSTALLED. See Installation.

Profile

A named set of behavior. A synth profile names quirks for a quirk corpus; a parser profile tells a parser which deviations to tolerate; a US Core profile is a FHIR StructureDefinition a resource claims. Profiles lists each.

Quirk

A deviation from a standard that real vendor traffic carries and a Cosyte parser's profile system describes as tolerable, such as a site-defined Z segment in HL7 v2. A quirk fixture is an otherwise spec-clean artifact with one quirk applied. A quirk changes the structure of a fixture, never where its values come from. The quirk catalog lists every one.

Round trip

Serializing an artifact, parsing it with its own parser, and serializing the result again. Each format has a helper that does it and reports the parser's warnings and whether the bytes came out the same. The FHIR helper also validates the resource.

Seed

The number that decides every choice a generator makes. The same seed gives the same bytes on the same versions of @cosyte/synth and the parser, so a test can regenerate its fixture rather than store it. See Pin a golden fixture.

Sentinel

An identifying value in a generated artifact, at a place that holds patient data, that the @cosyte/deid pairing loop tracks, such as a name from the invented-name pool, a 555-01xx phone number, or an identifier under the synthetic assigning authority. The loop de-identifies the artifact and fails if any sentinel is still in the output. See Check a de-identifier with the pairing loop.

Spec-clean

What a round trip reports when its parser reads the artifact with no warnings and serializes it back to the same bytes. For FHIR, spec-clean means the resource validates with no error or fatal issue and serializes back to the same bytes. It is the parser's judgment of our output, so it cannot catch a misreading of a standard the builder and the parser share. Who checks the output says which formats an outside validator also grades.

Synthetic by construction

No code path emits an identifying value from anywhere but a reserved range or the shipped pool of invented names, so no fixture can hold a real person's data. The floors above are how.

Test-data marker

The field a format's own standard defines for saying a message is test data, set on every artifact whose format has one: MSH-11 in HL7 v2, the HTEST security label in FHIR, ISA-15 in X12, and field 12 of the ASTM H record. See the marker table.