Citation Lexical Coverage

knowledge.citation_lexical_coverage

ProductionBounded custom (Class B)Deterministicbest-effortfastv1.0.0

No network

Overview

This measures wording overlap only; it never reports semantic support, truth or entailment. Send an answer (we segment it into claims) or the claims you segmented yourself, plus the sources they cite, and for each claim we resolve its citation markers, find the best matching window inside each cited source, and report how much of the claim's vocabulary that window covers. A citation aimed at the wrong paragraph or the wrong document shows up clearly. A fabrication that borrows the source's own words does not, and no lexical check can catch it. Markers are matched by a closed grammar rather than a pattern you supply, and a marker naming a source you did not send comes back as unresolved instead of being ignored. Neither your claims nor your sources are returned: you get ids, offsets and numbers, so the response stays small enough to store.

A citation is the part of a generated answer a reader is most likely to trust and least likely to check. Confirming that the cited text at least contains the claim's own vocabulary catches the common failure — a marker pointing at the wrong place — cheaply, deterministically, and without sending anything to a model.

Reliability

BEST_EFFORT because lexical overlap is not semantic evidence. meetsThreshold means only that the cited window contains the configured share of the claim's words. It cannot catch a plausible-sounding invention built from the source's vocabulary. The measurement itself is fully deterministic — the same claims, sources and settings always give the same verdicts and the same spans, and nothing here reads a clock or a random number. Terms are case-folded but never re-normalised, so every offset points into the text you submitted. Failures, none of which get better on their own: CITATION_INPUT_INVALID means you sent neither answer nor claims, the answer was empty or segmented into nothing, a claims entry was malformed or repeated an id, or an entry carried a property we do not accept; CITATION_SOURCE_INVALID means sources was empty or an entry was missing its id or text; CITATION_SOURCE_ID_DUPLICATE means two sources share an id, and the error says which position repeated it; CITATION_LIMIT_EXCEEDED means more than 500 sources or claims, an answer segmenting into more than 500 claims, a text over its character limit, more than 20000 claim-against-source comparisons, or a scan reaching 2000000 matched positions — check the answer in parts; CITATION_MARKER_GRAMMAR_UNSUPPORTED means markerGrammar or claimSegmentation named something that does not exist, and the accepted values come back with the error; CITATION_COVERAGE_PARAMETER_INVALID means a coverage setting was outside its range or named an unknown method.

Context footprint

The response carries ids, offsets, coverage scores and reason codes only, never your claims or sources. At most 500 claim entries come back, each with at most 64 citation entries. A successful call costs 1 execution credit.

When to use it

After a retrieval-augmented answer is produced and before it reaches a reader, or in review, when you want the evidence span for each claim.

When you need factual verification, when citations have to be discovered in free prose, or when the sources are not in your hands.

Powered by Squillify's own closed marker grammar and windowed lexical-overlap measurement. No model, no reranker and no third-party citation parser is involved.

Scope

Note

Resolve each claim's citation markers against the supplied sources and report how much of the claim's wording the best window inside a cited source covers.

Non-goals

Does not judge whether a claim is true, only whether the cited text covers it
Does not discover citations in prose — no URLs, DOIs or author-year forms
Does not fetch a source or follow a link
Does not compile a citation pattern you supply
Does not return your claim text or your source text

Properties

IdempotentRetry safeStreaming supportedFile input supportedRequires external service

Capabilities

text-processingvalidation

Cost & billing

Flat rate per invocation

MetercallsBillable onsuccessCost per call1 credit

File handling

Accepts file_refProduces file_refMax input2 MB

How we handle your data

Sensitive input, safe output

What this means

This skill may receive sensitive content, but it does not target sensitive data and returns only a structured non-sensitive result.

Squillify guarantees

Runs inside Squillify
This skill runs inside Squillify.
Payloads are not sent to an external service by this skill.
Squillify does not permanently store submitted payloads.
Telemetry records metadata only, not submitted content.
Submitted payloads are not used for training.
Submitted payloads are not reviewed by humans.
This skill does not target sensitive data.
The response does not contain submitted content.

Your responsibility

Squillify does not determine whether submitted content is sensitive. Your application is responsible for deciding what data it is allowed to submit. Your application is responsible for securing the response.

Read the Microskill Data Handling Policy

Input schema

Field	Type	Required	Description
answer	string	No	Required unless claims is provided. Answer of 1 to 200000 characters to segment and scan for markers.
claims	array	No	Required unless answer is provided. From 1 through 500 closed claims; when both arrive, claims wins.
sources	array	Yes	From 1 through 500 closed sources {id, text} that citations resolve against.
markerGrammar	string	No	How a citation looks; omit it for BRACKET_NUMERIC.
claimSegmentation	string	No	How an answer is split; omit it for SENTENCE, and it is ignored when you send claims.
coverage	object	No	Closed coverage measurement; omit it for TOKEN_OVERLAP at threshold 0.35, shingleSize 3 and spanWindowChars 2000.
requireCitation	boolean	No	Whether an uncited claim is an error; omit it for true.

Output schema

Field	Type	Required	Description
passesPolicy	boolean	Yes	True when no claim raised an error-severity issue.
claims	array	Yes	One closed claim report per checked claim in order, with ids, citations, scores and reasons rather than claim or source text.
coverageIssues	array	Yes	Claims that did not meet the coverage policy, each a closed {id, reasonCode} object.
summary	object	Yes	Closed counts and settings receipt with claimCount, meetsThresholdCount, coverageIssueCount, uncitedCount, unresolvedCitationCount, method, threshold, markerGrammar, claimSegmentation, spanWindowChars and truncated.

API contract

POST /v1/execute

curl

curl -sS -X POST "$BASE/v1/execute" -H "Authorization: Bearer $KEY" -H 'Content-Type: application/json' -d '{"skillKey":"knowledge.citation_lexical_coverage","input":{"answer":"Refunds take 30 days. [1]","sources":[{"id":"policy","text":"Refunds take 30 days from the charge."}],"coverage":{"method":"TOKEN_OVERLAP","threshold":0.35}}}'

Integrations

TypeScript SDK

import { Squillify } from "@squillify/sdk";

const sq = new Squillify({ apiKey: process.env.SQUILLIFY_API_KEY });

const { output } = await sq.execute("knowledge.citation_lexical_coverage", {"answer":"We refund an unused subscription within 30 days of the charge. [1] Every refund is paid in the original currency. [2]","sources":[{"id":"policy","text":"Refunds. We refund an unused subscription within 30 days of the charge, provided the original order id is supplied."},{"id":"shipping","text":"Shipping. Parcels leave the warehouse within two working days."}],"coverage":{"threshold":0.5}});

Errors

Code	HTTP	Retryable	Description
VALIDATION_ERROR	400	No	The request payload failed schema validation.
UNAUTHORIZED	401	No	Missing or invalid API key.
SKILL_NOT_ACCESSIBLE	403	No	The API key is not granted access to this skill.
SKILL_NOT_AVAILABLE_FOR_PLAN	403	No	This skill is not included in your current plan.
PLAN_ELIGIBILITY_UNAVAILABLE	503	Yes	Plan eligibility could not be decided. Retry after backoff.
SKILL_NOT_FOUND	404	No	The requested skill or version does not exist.
TENANT_RATE_LIMITED	429	Yes	Request rate limit exceeded. Retry after backoff.
QUOTA_EXCEEDED	429	Yes	Plan quota exhausted. Retryable once the quota period resets.
EXECUTION_TIMEOUT	504	Yes	Execution exceeded the allowed time budget.
CITATION_INPUT_INVALID	400	No	The request payload failed this skill's input validation.
CITATION_SOURCE_INVALID	400	No	The request payload failed this skill's input validation.
CITATION_SOURCE_ID_DUPLICATE	400	No	The request payload failed this skill's input validation.
CITATION_LIMIT_EXCEEDED	429	No	A size, quota, rate, or time limit was exceeded.
CITATION_MARKER_GRAMMAR_UNSUPPORTED	400	No	The skill does not support the requested variant.
CITATION_COVERAGE_PARAMETER_INVALID	400	No	The request payload failed this skill's input validation.

Other limits

Limit	Value
Recommended inline bytes	32 KiB

Limitations

Does not judge whether a claim is true, only whether the cited text covers it
Does not discover citations in prose — no URLs, DOIs or author-year forms
Does not fetch a source or follow a link
Does not compile a citation pattern you supply
Does not return your claim text or your source text
Fact-checking a claim against the world rather than against a source
Extracting citations from free prose, which text.extract_citations handles
Detecting a fabrication written in the source's own vocabulary

How to call

POST /v1/execute (requires authentication)

Request bodyjson

{
  "skillKey": "knowledge.citation_lexical_coverage",
  "input": {
    "answer": "We refund an unused subscription within 30 days of the charge. [1] Every refund is paid in the original currency. [2]",
    "sources": [
      {
        "id": "policy",
        "text": "Refunds. We refund an unused subscription within 30 days of the charge, provided the original order id is supplied."
      },
      {
        "id": "shipping",
        "text": "Shipping. Parcels leave the warehouse within two working days."
      }
    ],
    "coverage": {
      "threshold": 0.5
    }
  }
}

Examples

An answer segmented into sentences, one claim above threshold and one below

Inputjson

{
  "answer": "We refund an unused subscription within 30 days of the charge. [1] Every refund is paid in the original currency. [2]",
  "sources": [
    {
      "id": "policy",
      "text": "Refunds. We refund an unused subscription within 30 days of the charge, provided the original order id is supplied."
    },
    {
      "id": "shipping",
      "text": "Shipping. Parcels leave the warehouse within two working days."
    }
  ],
  "coverage": {
    "threshold": 0.5
  }
}

Expected outputjson

{
  "summary": {
    "claimCount": 2,
    "meetsThresholdCount": 1,
    "coverageIssueCount": 1,
    "uncitedCount": 0,
    "unresolvedCitationCount": 0,
    "method": "TOKEN_OVERLAP",
    "threshold": 0.5,
    "markerGrammar": "BRACKET_NUMERIC",
    "claimSegmentation": "SENTENCE",
    "spanWindowChars": 2000,
    "truncated": false
  },
  "claims": [
    {
      "id": "claim-1",
      "citations": [
        {
          "reference": "1",
          "resolved": true,
          "sourceId": "policy"
        }
      ],
      "meetsThreshold": true,
      "bestSourceId": "policy",
      "bestSpan": {
        "start": 9,
        "end": 70
      },
      "coverageScore": 0.9167,
      "reasonCode": "MEETS_THRESHOLD",
      "severity": "NONE"
    },
    {
      "id": "claim-2",
      "citations": [
        {
          "reference": "2",
          "resolved": true,
          "sourceId": "shipping"
        }
      ],
      "meetsThreshold": false,
      "bestSourceId": "shipping",
      "bestSpan": {
        "start": 24,
        "end": 27
      },
      "coverageScore": 0.1111,
      "reasonCode": "BELOW_THRESHOLD",
      "severity": "ERROR"
    }
  ],
  "passesPolicy": false,
  "coverageIssues": [
    {
      "id": "claim-2",
      "reasonCode": "BELOW_THRESHOLD"
    }
  ]
}

Tip

The first sentence cites source 1 and repeats its wording almost exactly, so it meets the threshold and comes back with the span that matched. The second cites source 2, which is about shipping, so it falls below the threshold and passesPolicy turns false.

Claims you segmented yourself, with one citation that resolves to nothing