Annotation span evaluation
Evaluate supplied labeled text spans with exact boundaries or thresholded intersection-over-union. Bounded deterministic maximum-cardinality one-to-one matching returns precision/recall/F1, matches and unmatched IDs. UTF-16 half-open offsets are validated; labels and text meaning are not judged.
data-assurance · Operation ID: span-evaluate
Choose this operation when
- evaluate named entity extraction annotation spans
- match predicted and reference text spans using exact offsets or iou
- calculate one-to-one span precision recall and unmatched evidence
Outside this profile
- Semantic factuality or entailment checking
- Tokenization byte offsets or globally maximum total IoU matching
Exact release references
Static JSON contract · Markdown reference · Fixed example response
Complete input schema
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"properties": {
"text": {
"type": "string",
"maxLength": 50000
},
"references": {
"maxItems": 100,
"type": "array",
"items": {
"type": "object",
"properties": {
"id": {
"type": "string",
"minLength": 1,
"maxLength": 80
},
"label": {
"type": "string",
"minLength": 1,
"maxLength": 80
},
"start": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
},
"end": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
}
},
"required": [
"id",
"label",
"start",
"end"
],
"additionalProperties": false
}
},
"predictions": {
"maxItems": 100,
"type": "array",
"items": {
"type": "object",
"properties": {
"id": {
"type": "string",
"minLength": 1,
"maxLength": 80
},
"label": {
"type": "string",
"minLength": 1,
"maxLength": 80
},
"start": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
},
"end": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
}
},
"required": [
"id",
"label",
"start",
"end"
],
"additionalProperties": false
}
},
"match": {
"default": "exact",
"type": "string",
"enum": [
"exact",
"overlap"
]
},
"minimumIou": {
"default": 0.5,
"type": "number",
"exclusiveMinimum": 0,
"maximum": 1
}
},
"required": [
"text",
"references",
"predictions"
],
"additionalProperties": false
}
Complete output-envelope schema
{
"type": "object",
"required": [
"operation",
"version",
"result",
"provenance"
],
"properties": {
"operation": {
"const": "span-evaluate",
"type": "string"
},
"version": {
"const": "0.29.0",
"type": "string"
},
"result": {
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"properties": {
"profile": {
"type": "string",
"const": "annotation-span-evaluation-v1"
},
"offsetUnit": {
"type": "string",
"const": "utf16-code-unit"
},
"referenceCount": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
},
"predictionCount": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
},
"matchedCount": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
},
"precision": {
"type": "number"
},
"recall": {
"type": "number"
},
"f1": {
"type": "number"
},
"precisionDefined": {
"type": "boolean"
},
"recallDefined": {
"type": "boolean"
},
"meanMatchedIou": {
"type": [
"number",
"null"
]
},
"matches": {
"type": "array",
"items": {
"type": "object",
"properties": {
"predictionId": {
"type": "string"
},
"referenceId": {
"type": "string"
},
"label": {
"type": "string"
},
"iou": {
"type": "number"
},
"exactBoundaries": {
"type": "boolean"
}
},
"required": [
"predictionId",
"referenceId",
"label",
"iou",
"exactBoundaries"
],
"additionalProperties": false
}
},
"unmatchedPredictionIds": {
"type": "array",
"items": {
"type": "string"
}
},
"unmatchedReferenceIds": {
"type": "array",
"items": {
"type": "string"
}
},
"perLabel": {
"type": "array",
"items": {
"type": "object",
"properties": {
"label": {
"type": "string"
},
"referenceCount": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
},
"predictionCount": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
},
"matchedCount": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
},
"precision": {
"type": "number"
},
"recall": {
"type": "number"
},
"f1": {
"type": "number"
}
},
"required": [
"label",
"referenceCount",
"predictionCount",
"matchedCount",
"precision",
"recall",
"f1"
],
"additionalProperties": false
}
},
"conventions": {
"type": "array",
"items": {
"type": "string"
}
}
},
"required": [
"profile",
"offsetUnit",
"referenceCount",
"predictionCount",
"matchedCount",
"precision",
"recall",
"f1",
"precisionDefined",
"recallDefined",
"meanMatchedIou",
"matches",
"unmatchedPredictionIds",
"unmatchedReferenceIds",
"perLabel",
"conventions"
],
"additionalProperties": false
},
"provenance": {
"type": "object",
"required": [
"inputSha256",
"outputSha256",
"deterministic",
"externalRequests"
],
"properties": {
"inputSha256": {
"type": "string",
"pattern": "^[a-f0-9]{64}$"
},
"outputSha256": {
"type": "string",
"pattern": "^[a-f0-9]{64}$"
},
"deterministic": {
"const": true
},
"externalRequests": {
"const": 0
}
}
}
},
"additionalProperties": false
}
Fixed example
One accepted fixed example, not a custom-input trial. No operation runs when this static page is requested.
Example input
{
"text": "Ada builds agents in London.",
"references": [
{
"id": "r1",
"label": "person",
"start": 0,
"end": 3
},
{
"id": "r2",
"label": "place",
"start": 21,
"end": 27
}
],
"predictions": [
{
"id": "p1",
"label": "person",
"start": 0,
"end": 3
},
{
"id": "p2",
"label": "place",
"start": 21,
"end": 26
}
],
"match": "overlap",
"minimumIou": 0.8
}
Example response
{
"operation": "span-evaluate",
"version": "0.29.0",
"result": {
"profile": "annotation-span-evaluation-v1",
"offsetUnit": "utf16-code-unit",
"referenceCount": 2,
"predictionCount": 2,
"matchedCount": 2,
"precision": 1,
"recall": 1,
"f1": 1,
"precisionDefined": true,
"recallDefined": true,
"meanMatchedIou": 0.9166666666666667,
"matches": [
{
"predictionId": "p1",
"referenceId": "r1",
"label": "person",
"iou": 1,
"exactBoundaries": true
},
{
"predictionId": "p2",
"referenceId": "r2",
"label": "place",
"iou": 0.8333333333333334,
"exactBoundaries": false
}
],
"unmatchedPredictionIds": [],
"unmatchedReferenceIds": [],
"perLabel": [
{
"label": "person",
"referenceCount": 1,
"predictionCount": 1,
"matchedCount": 1,
"precision": 1,
"recall": 1,
"f1": 1
},
{
"label": "place",
"referenceCount": 1,
"predictionCount": 1,
"matchedCount": 1,
"precision": 1,
"recall": 1,
"f1": 1
}
],
"conventions": [
"Spans are nonempty half-open UTF-16 code-unit ranges; surrogate pairs cannot be split. Labels match exactly. Repeated label/boundary spans on the same side are rejected.",
"Matching maximizes one-to-one pair count. Deterministic ID order and local IoU preference break ties; total IoU is not globally maximized.",
"Undefined precision/recall/F1 uses 0. This evaluates supplied annotations and boundaries, not the truth or meaning of their labels."
]
},
"provenance": {
"inputSha256": "17fd96b5568ac2fb16fcbdb52e9741bfa919234807e289f37c7ea47b30710f28",
"outputSha256": "69f28072bd7f33501f9d2566743476fd9024f8437868687e435fa8156ff5c4f1",
"deterministic": true,
"externalRequests": 0
}
}
Bounds and precision
JavaScript IEEE-754 numbers; use strings for large integer IDs/exact decimals where the schema accepts strings. No lossless numeric parsing.
{
"global": {
"requestBytes": 131072,
"responseBytes": 524288,
"jsonDepth": 32,
"jsonNodes": 20000,
"requestsPerMinute": 60,
"paidAttemptsPerMinute": 20,
"idempotencyHours": 24
},
"operation": {
"inputBytes": 100000,
"outputBytes": 400000,
"jsonNodes": 12000,
"depth": 20,
"evidence": 200,
"keyWorkBytes": 3000000,
"matchingSteps": 2000000,
"textUtf16Units": 50000,
"references": 100,
"predictions": 100
}
}
Complete schemas, descriptions and cross-field validation may impose additional limits.
Proposed price and protocol definitions
{
"unit": "one successful operation call",
"proposedNominalUsd": "0.01",
"sixDecimalTokenBaseUnits": "10000",
"subscription": false,
"includesPayerWalletOrNetworkFees": false,
"liveQuoteVerified": false,
"condition": "Actual SDK challenge is authoritative only within the caller's explicit authorization; configured six-decimal token peg is an operator assertion, not a conversion guarantee."
}
Protocol definitions: x402, mpp. MPP uses Tempo charge. Paid MCP execution is unsupported. All runtime readiness is not evaluated in this build.
API path templates, not endpoints on this documentation host
{
"x402": "/v1/x402/span-evaluate",
"mpp": "/v1/mpp/span-evaluate"
}
Required headers
{
"Content-Type": "application/json",
"Idempotency-Key": "random 16–128 character operation identifier"
}
Actual SDK challenge amount, asset, network, recipient and wallet costs must pass independent authorization. Preserve identical key, body, protocol and credential on retries; on PAYMENT_UNCERTAIN stop and reconcile.
Execution profile and provider conditions
{
"deterministic": true,
"externalRequests": 0,
"maxExternalRequests": 0,
"resultSnapshotPersisted": false,
"fixedExampleIsIllustrativeSnapshot": false,
"requiresPayment": true,
"supportsMcpExecution": false
}
Deterministic supplied-input operation with no external requests or stored request/result bodies. Payment infrastructure retains payment metadata and hashes.
Failure handling
- HTTP 400: Malformed JSON, missing/invalid idempotency key, or payment identifier mismatch Correct the request before payment
- HTTP 402: Payment challenge or rejected payment Use official protocol SDK; inspect payment outcome before another payment
- HTTP 409: Idempotency conflict, duplicate proof, or PAYMENT_UNCERTAIN Keep original key, body, and proof; reconcile uncertainty with operator; never blindly repay
- HTTP 413: Input or generated output too large Reduce input; no payment attempted for validation failure
- HTTP 415: Unsupported media type or compression Send uncompressed application/json
- HTTP 422: Schema or service-specific semantic validation failure Correct input using returned error code; no payment attempted
- HTTP 429: Request/payment-attempt rate exceeded Wait for rate limit window; preserve existing payment identity
- HTTP 503: Payment configuration/provider/state unavailable, or live DNS preparation failed before settlement Check readiness; DNS preparation failures may retry the identical key/body/credential only; uncertainty requires reconciliation
Declared requirements
Before any paid call, refresh the live operation contract and POST the complete bounded budgeted plan to the separate API's /preflight. Unknown requirements block selection; compatible preflight is not permission to spend.