Entity

Source file	schemas/entity.schema.json
$id	urn:synapsecommand:cdm:3.0.0:entity
CDM schema version	3.0.0
SHA-256 of source	310d04623b4647e0f92abd384a48683df783495ec9e1ffd1bedf85ee18bf092a

Anything that exists on the map, at a stated time, with stated confidence.

Fields

Field	Type	Required	Description
affiliation	Affiliation	yes
attributes	object	no	Source-specific fields the CDM has no home for. The never-drop bag: park data here rather than discarding it.
confidence	number | null	no	0..1. None = unknown; 0 means certainty-that-not, which is a claim. Default null. (≥ 0; ≤ 1)
entity_id	string (uuid)	yes	Stable across updates — derived, see ids.derive(). Never drawn at random.
entity_type	EntityType	yes
integrity	Integrity | null	no	PQC signature block — designed, not yet populated. A DATA CONTAINER and nothing more: this package makes no signature and verifies none, no conformance check, harness column or evidence field reads it, and its presence on an object proves nothing about the object. A record carrying one is unverified until something outside this package verifies it. Default null.
kinematics	Kinematics | null	no	Default null.
object_kind	"entity"	no
ontology_types	array<string>	no	Optional SC-OES semantic types — absolute ontology identifiers saying what this thing IS in operational terms. Empty = the producer asserted none. Never derived from entity_type, and entity_type is never derived from it.
position	Position | null	no	None = position unknown. NEVER a Position holding zeros. Default null.
quality	Quality | null	no	How good the SOURCE says this object is. None = the source said nothing about quality, which is not the same as saying it is poor. Default null.
residual	Residual | null	no	Source information the CDM does not model, under the name of the format it came from (§28). None = nothing was left over, or — for the fourteen adapters shipped before this container existed — the leftovers are parked in attributes / payload under source_extras, which ARCHITECTURE.md §5 rules they keep through Part 1. Default null.
schema_version	string	no	Semver of the CDM this object was written against. Default "3.0.0". (pattern ^(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)$)
source	SourceRef	yes	Which adapter produced this object. Required on every kind.
source_ids	array<SourceId>	yes	Every external identifier this object is known by. At least one, on EVERY kind — see the class docstring. (min items 1)
status	OperationalStatus | null	no	The source's own operational state for this object, namespaced. None = the source stated none. Default null.
symbol	string | null	no	MIL-STD-2525D SIDC, 20 digits. None when the source states no symbol — see symbology.sidc_from_affiliation() for deriving one. Default null. (pattern ^[0-9]{20}$)
valid_from	string	yes	When this state began. (pattern ^[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}\.[0-9]{3}Z$)
valid_to	string | null	no	When it ceased. None = still current / open-ended. Default null. (pattern ^[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}\.[0-9]{3}Z$)

additionalProperties: false

Unknown keys are rejected. That is safe only because the CDM pairs strictness with a declared escape hatch — Entity.attributes and Event.payload accept anything — so an adapter never has to choose between dropping a field and failing validation.

Referenced definitions

Every $ref on this page resolves to one of these, inlined here so the page is a complete reference and not a starting point for chasing pointers.

Affiliation

Maps to MIL-STD-2525 standard identity — see symbology.standard_identity().

Four members, not 2525's seven: PENDING, ASSUMED_FRIEND and SUSPECT are judgements a fusion layer makes, not facts an adapter can read off a wire format. An adapter that invented ASSUMED_FRIEND would be doing business logic, which adapters may not do. The source's own wording is preserved in attributes when it is finer than this.

Closed vocabulary — a value outside this list is invalid, and UNKNOWN is a

member rather than a null wherever the enum has one.

Value
FRIENDLY
HOSTILE
NEUTRAL
UNKNOWN

EntityType

Closed vocabulary — a value outside this list is invalid, and UNKNOWN is a

member rather than a null wherever the enum has one.

Value
UNIT
PLATFORM
SENSOR
FACILITY
EVACUEE_GROUP
INTERFERENCE_SOURCE
OVERLAY_OBJECT
UNKNOWN

Integrity

DESIGNED, NOT IMPLEMENTED — the field the PQC signature will occupy.

No crypto happens in this package (tests/test_cdm_boundary.py asserts the package imports no crypto module). The field exists from day one so that turning signing on is a value change rather than a schema change: a schema change would be a MAJOR bump rippling through every store and every consumer, and would arrive exactly when the signing work is already late.

algorithm is a free string rather than an enum, naming what the platform's ledger already uses — ML-DSA-87 for entry signatures, SLH-DSA for checkpoints. Free, because the algorithm that replaces those is not knowable now, and an enum would make the migration a MAJOR bump for a value nobody reasons over programmatically.

All three fields or none. A block holding a signature with no algorithm is unverifiable, and an unverifiable signature that LOOKS present is worse than an absent one: it reads as assurance to everything downstream that does not check.

Field	Type	Required	Description
algorithm	string	yes	e.g. ML-DSA-87, SLH-DSA-SHAKE-256s. (min length 1)
chain_hash	string	yes	Hash binding this object to the chain. (min length 1)
signature	string	yes	(min length 1)

additionalProperties: false — unknown keys are rejected. Source-specific fields belong in the declared extension bags (Entity.attributes, Event.payload).

Kinematics

Motion. Every field optional, and absent means UNKNOWN, never zero.

This is the AIS sentinel lesson in schema form: 0 kt is measured stillness, 0 deg is a course due north, 0 m/s climb is level flight. All three are real measurements, so none of them can double as "no data" — the adapter translates the source's sentinel to None.

Field	Type	Required	Description
climb_mps	number | null	no	Metres per second, negative = descending. Default null.
course_deg	number | null	no	Degrees true, [0, 360). Default null. (≥ 0; < 360)
speed_mps	number | null	no	Metres per second. Default null. (≥ 0)

additionalProperties: false — unknown keys are rejected. Source-specific fields belong in the declared extension bags (Entity.attributes, Event.payload).

OperationalStatus

What state the source says a thing is in, in the SOURCE's vocabulary, namespaced.

THERE IS NO ENUM HERE AND THERE WILL NOT BE ONE. "SERVICEABLE", "DEGRADED", "MISSION CAPABLE", "RED" and "U/S" come from five different domains and mean five different things, and a closed CDM vocabulary would have to map each of them onto a member — which is a judgement about somebody else's operational language, made inside a translator, invisible in the output, and exactly what Rule 6 forbids. namespace says whose vocabulary state belongs to, so a consumer meeting an unfamiliar value can find out what it means instead of guessing; a consumer MUST NOT compare state across namespaces.

since is when the state began, not when it was reported. Absent means the source did not say — never the receipt time, which would make every restart look like a state change.

Field	Type	Required	Description
attributes	object	no	Status-specific source fields with no canonical home.
namespace	string	yes	Whose vocabulary state is in — normally the source format's name. Required: an unnamespaced status is a word with no owner. (min length 1)
since	string | null	no	When the state began. None = the source did not say. Default null. (pattern ^[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}\.[0-9]{3}Z$)
state	string	yes	The source's own token, verbatim and untranslated. (min length 1)

additionalProperties: false — unknown keys are rejected. Source-specific fields belong in the declared extension bags (Entity.attributes, Event.payload).

Position

A fix. Both coordinates required — that is how the null-never-zero rule is structural.

An unknown position is the ABSENCE of this object (entity.position is None), never a Position holding zeros. Because lat and lon are required here, an adapter cannot express "unknown" as (0, 0) even by accident: it has to either omit the Position or state a real coordinate. Coordinate zero is a real point in the Gulf of Guinea, and a contact painted there is a contact that does not exist.

accuracy_m absent means unknown accuracy, NOT perfect accuracy. Zero would mean a fix with no error, which no sensor produces.

Field	Type	Required	Description
accuracy_m	number | null	no	Metres, 1-sigma. None = unknown, never 0. Default null. (≥ 0)
alt_m	number | null	no	Metres HAE. None = unknown. Default null.
lat	number	yes	WGS84 decimal degrees. (≥ -90; ≤ 90)
lon	number	yes	WGS84 decimal degrees. (≥ -180; ≤ 180)
position_source	PositionSource	yes	How the fix was obtained — the field that survives GNSS denial.
vertical	VerticalPosition | null	no	The height AS THE SOURCE STATED IT — unit and datum carried, never converted. alt_m above stays the canonical HAE-in-metres projection and is None whenever the source's datum is not HAE. Default null.

additionalProperties: false — unknown keys are rejected. Source-specific fields belong in the declared extension bags (Entity.attributes, Event.payload).

PositionSource

How the position was obtained. Load-bearing in a GNSS-denied environment.

This is the field that lets a commander tell a fix from a guess. When PNTMAP reports jamming over an area, every GNSS-sourced position inside that area becomes suspect and every INERTIAL or MANUAL one does not — a distinction that is impossible to make after the fact if the adapter flattened them all to "position".

Closed vocabulary — a value outside this list is invalid, and UNKNOWN is a

member rather than a null wherever the enum has one.

Value
GNSS
INERTIAL
MANUAL
ESTIMATED

Quality

How good the source says its own data is. Four ways of saying it, none derived.

confidence is 0..1 and comparable across sources; source_quality is the source's OWN grade, a free string, because ASTERIX's track quality, AIS's position accuracy flag and a NATO track's evaluation code are three ordinal scales with no defined mapping between them and inventing one would be a fusion decision made inside a translator.

uncertainty is a NAMED dict — {"along_track_m": 40.0, "cross_track_m": 12.0} — rather than a single number, because an error ellipse is not a radius and flattening it loses the orientation that made it worth sending. Names are the source's; the CDM does not fix a vocabulary here, and the unit belongs in the key exactly as signal_strength_dbm carries its own unit in its name.

Field	Type	Required	Description
accuracy_m	number | null	no	Metres, 1-sigma. None = unknown, never 0. Default null. (≥ 0)
confidence	number | null	no	0..1. None = unknown; 0 means certainty-that-not, which is a claim. Default null. (≥ 0; ≤ 1)
source_quality	string | null	no	The source's own grade, verbatim. None = the source stated none. Default null. (min length 1)
uncertainty	object<string, number>	no	Named components, e.g. along_track_m / cross_track_m. The unit is in the key. Empty = the source named none.

additionalProperties: false — unknown keys are rejected. Source-specific fields belong in the declared extension bags (Entity.attributes, Event.payload).

Residual

Source information the CDM does not model, kept under the name of the format it came from.

§28's container, and its two rules are load-bearing in opposite directions:

1. It MUST identify its origin. namespace is the source format's own name — the same string as the adapter's metadata.format.name — so a reader meeting an unfamiliar key inside data can find out which standard's vocabulary it belongs to. An unnamespaced bag of leftovers is the free-form dict this model exists to replace.
2. It MUST NOT be treated as semantically trusted merely because it survived translation. A residual says "the source said this and the CDM has no home for it". It is a fact about the source RECORD, not an assertion about the world. A consumer MUST NOT promote a residual value into a canonical field, and a later adapter MUST NOT read another adapter's residual as an input — ARCHITECTURE.md §5 states both as normative text.

data preserves the source's own STRUCTURE, which is why it is a dict and not a list of dotted paths: lossless.residual() learned that the hard way when a two-element list came back as two keys named affected_constellations[0] and [1], satisfying the never-drop rule in the letter while destroying the reader's ability to see a list.

THE FOURTEEN ADAPTERS IN THIS REPOSITORY DO NOT USE THIS YET, deliberately. ARCHITECTURE.md §5 rules that they keep their attributes / payload parking under source_extras through the whole of Part 1 and declare residual: legacy in their manifests, because the information is already preserved and paying for a placement change with every golden file and every downstream consumer buys nothing a reader can use (§29: no breaking change for stylistic cleanliness). Every Part 2 adapter declares residual: structured and uses this.

Field	Type	Required	Description
data	object	no	The unconsumed source structure, preserved as the source shaped it.
namespace	string	yes	The source format's name — normally metadata.format.name. Required. (min length 1)

additionalProperties: false — unknown keys are rejected. Source-specific fields belong in the declared extension bags (Entity.attributes, Event.payload).

SourceHash

A digest of the source record, as the ADAPTER computed it. Carried, never computed here.

Rule 5 asks for "source hash where appropriate", and the appropriateness is the adapter's judgement: a hash of a 3-byte AIS sentence identifies nothing an external id does not, while a hash of a 4 MB NITF segment is how an auditor proves the bytes on the wire were the bytes described. So it is optional, and its absence means the adapter did not compute one.

NO HASHING HAPPENS IN THIS PACKAGE. tests/test_cdm_boundary.py asserts that no module under synapse_cdm/ imports hashlib or any crypto library, and this model does not change that: it is a container for a value produced outside the contract layer, exactly as Integrity is a container for a signature this package does not make. algorithm is therefore required and free-form — a digest with no algorithm cannot be reproduced by anyone, and an enum would make the day SHA-3 arrives a MAJOR bump for a string nobody branches on.

Field	Type	Required	Description
algorithm	string	yes	e.g. sha256. Named by the producer. (min length 1)
value	string	yes	The digest, in the producer's own encoding. (min length 1)

additionalProperties: false — unknown keys are rejected. Source-specific fields belong in the declared extension bags (Entity.attributes, Event.payload).

SourceId

One external identifier for an object — the provenance mapping.

A list of these, not one, because the same object arrives from several systems: the same vessel is an MMSI to AIS, a track number to STANAG 4676 and a UID to TAK. Fusion joins them later; the adapter's job is to record which name its own system used, and never to overwrite another system's entry.

Field	Type	Required	Description
external_id	string	yes	That system's own identifier. (min length 1)
system	string	yes	The external system, e.g. PNTMAP, TAK. (min length 1)

additionalProperties: false — unknown keys are rejected. Source-specific fields belong in the declared extension bags (Entity.attributes, Event.payload).

SourceRef

Which adapter produced this object, from which system, and whether it is real.

synthetic is required and has no default. Every fixture in this repository is synthetic and every scenario package is too (TR-12), and the platform keeps the synthetic and live layers apart over one interface — so an object that does not say which layer it belongs to cannot be filed. A default of false would silently promote exercise data to operational data, which is the dangerous direction; a default of true would silently demote live data and hide it from an operator. There is no safe default, so there is no default.

Field	Type	Required	Description
adapter	string	yes	Adapter name, e.g. pntmap. (min length 1)
adapter_version	string	yes	Adapter semver: MAJOR.MINOR.PATCH, no leading zeroes, nothing else. (pattern ^(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)$; min length 1)
format_name	string | null	no	The source STANDARD's name, e.g. 'ASTERIX'. Filled from the adapter's own declared metadata.format; None only for an adapter that declares none. Default null.
format_version	string | null	no	The edition of that standard, e.g. 'Cat 062 ed 1.18'. None = no document in this tree states which edition the adapter targets — a reading, never a gap filled in. Default null.
observed_at	string | null	no	The instant the SOURCE RECORD states for itself, when it states one and no canonical field already carries it. Event.observed_at stays the event's own time; this is the record's. Default null. (pattern ^[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}\.[0-9]{3}Z$)
original_id	string | null	no	The source record's OWN identifier, as a first-class provenance field. Distinct from source_ids, which is the identity of the THING; this is the identity of the RECORD that described it. Default null.
record_index	integer | null	no	Which record of a multi-record payload this object came from, 0-based. None = the payload was not a sequence, or the adapter does not track it. Never -1 and never 0-as-unknown: 0 is the first record. Default null. (≥ 0)
source_hash	SourceHash | null	no	Digest of the source record, computed by the adapter. Default null.
synthetic	boolean	yes	true for anything not from a real source (TR-12).
system	string	yes	The external system this came from. (min length 1)
transformations	array<string>	no	Rule 5's transformation chain: the TRANSFORMS reasons this adapter applied, in the order it applied them. Empty = nothing was transformed, which is a claim the lossless check can contradict.

additionalProperties: false — unknown keys are rejected. Source-specific fields belong in the declared extension bags (Entity.attributes, Event.payload).

VerticalPosition

A height, with the unit it was stated in and the datum it was measured from.

§26's rule made structural: value is never alone. Three fields are required together because any two of them without the third is the ambiguity the policy exists to prevent — 600 could be metres or feet, and 600 ft AGL over the Alps is not 600 ft MSL.

NO CONVERSION HAPPENS HERE. This model records; it does not normalise. models.Position keeps alt_m as the canonical HAE-in-metres projection and a validator (there, not here) requires the two to agree when both are stated and the reference really is HAE. When the source's altitude is anything else, alt_m stays None — Rule 2, unknown is not a substituted value — and this block carries what the source actually said. Converting MSL to HAE needs a geoid model, converting a flight level needs the local QNH, and an adapter that performed either silently would be publishing a number nobody measured.

uncertainty is in the SAME unit as value, and is one-sigma, matching Position.accuracy_m's convention. Absent means unknown, never zero: zero would assert a height with no error, which no altimeter and no GNSS receiver produces.

Field	Type	Required	Description
reference	VerticalReference	yes	The datum. UNKNOWN is a member, not null.
uncertainty	number | null	no	1-sigma, in the same unit as value. None = unknown, never 0. Default null. (≥ 0)
unit	VerticalUnit	yes	Never inferred; see VerticalUnit.
value	number	yes	The number the source stated, in unit.

additionalProperties: false — unknown keys are rejected. Source-specific fields belong in the declared extension bags (Entity.attributes, Event.payload).

VerticalReference

What a VerticalPosition is measured FROM. The datum, never assumed.

The six here are the ones the formats in scope actually state. HAE is the WGS84 ellipsoid — what a GNSS receiver computes natively and what Position.alt_m has always meant. MSL is a geoid model, and the separation between the two reaches 100 m in places, so a silent substitution moves an aircraft by more than its own vertical separation minimum. AGL is height above the terrain beneath the object, which is not a datum at all but a difference, and is therefore not convertible to either of the others without a terrain model. BARO is an altimeter reading against a stated or unstated pressure setting; FL is BARO against the standard setting.

UNKNOWN is a MEMBER, unlike VerticalUnit's absent one, and the asymmetry is the point: a number with no unit cannot be rendered at all, whereas a number whose datum is unstated can still be shown to an operator beside the words "reference unknown". That is a worse fix than a referenced one and a far better one than a fix silently labelled MSL.

Closed vocabulary — a value outside this list is invalid, and UNKNOWN is a

member rather than a null wherever the enum has one.

Value
HAE
MSL
AGL
BARO
FL
UNKNOWN

VerticalUnit

The unit a VerticalPosition states its number in. §26's "value plus unit", closed.

Three members and no UNKNOWN, which is a departure from this module's opening paragraph and is the unit policy rather than an oversight. §26 forbids "ambiguous naked numeric fields where unit ambiguity matters"; a vertical measurement whose unit nobody knows IS that naked number wearing a wrapper, and recording it would let a consumer render 300 as metres when the source meant feet. An adapter that cannot read the unit therefore has no vertical position to state — it parks the raw number in the residual, where nothing reads it as a height. Absence is expressible (the whole VerticalPosition is optional); an unknown unit is not, deliberately.

FL is a unit as well as a reference because a flight level IS its own scale: FL350 is "35000 feet on the 1013.25 hPa isobaric surface", a number that is neither metres nor feet above anything on the ground. Converting it needs the local pressure, which the adapter does not have, so it is carried as it was stated.

Closed vocabulary — a value outside this list is invalid, and UNKNOWN is a

member rather than a null wherever the enum has one.

Value
m
ft
FL