QiXCite Documentation

Citation governance for legal AI with managed cloud defaults, explicit treatment labeling, and defensibility-ready evidence bundles.

What is QiXCite?

QiXCite is a citation governance engine for legal AI, part of the Quorum Integrity eXchange (QIX) framework by OrchIntel. It prevents hallucinated citations, verifies authority existence, separates treatment lanes (Licensed Treatment vs Treatment Insight (Estimated β€’ Beta)), and produces verifiable evidence bundles for every verification.

QiXCite is purpose-built for law firms, legal departments, and compliance teams that use AI in their workflows. It runs as a managed service on IOA Cloud or on your own infrastructure with a local/on-premise deployment.

Documentation

Everything you need to understand, deploy, and use QiXCite.

API Endpoints

QiXCite exposes verification, monitoring, and OCG operations endpoints. Endpoints are available at http://localhost:8000 (local) or https://api.orchintel.com/api/qixcite (cloud).

Method Endpoint Description Auth
GET /healthz Health check with mode and version No
GET /v1/mode Deployment mode, auth requirements, available features No
GET /version Version and Git SHA No
POST /verify Main V1 pipeline: extract, validate, semantic support, quote-lock, evidence Yes
POST /v1/cite/validate Single citation validation with hallucination scoring Yes
POST /v1/cite/quote-lock SHA-256 cryptographic quote integrity hash Yes
POST /v1/documents/scan Full-document citation extraction and validation Yes
POST /v1/efiling/precheck E-filing compliance checks (PASS/WARN/FAIL grading) Yes
POST /v1/batch Submit async batch of citations for validation Yes
GET /v1/batch/{job_id} Poll for batch job status and results No
GET /qixcite/providers Available citator provider capabilities No
GET /v1/analytics Citation validation analytics No
GET /v1/cache/stats Validation cache statistics No
DEL /v1/cache Flush validation cache No
GET /v1/evidence/sample Download a sample evidence bundle (audit-grade JSON + SHA-256 signature) No
POST /v1/monitor Register citations for continuous monitoring (detect status changes over time) Yes
GET /v1/monitor List all active citation monitors Yes
GET /v1/monitor/{id} Get monitor details including status change history Yes
POST /v1/monitor/{id}/check Manually trigger a re-check on a monitored citation Yes
DEL /v1/monitor/{id} Remove a citation from continuous monitoring Yes
POST /v1/authority-alignment Authority Alignment (Beta) β€” Check claim-to-authority alignment for a single citation via LLM NLI Yes
GET /v1/authority-alignment/quota Authority Alignment (Beta) β€” Returns free-tier AAE check quota status for the authenticated tenant Yes

Core Capabilities

Citation Extraction

Extracts legal citations from documents using 50+ regex patterns across US, UK, AU, NZ, CA, EU jurisdictions and academic DOIs.

Existence Check

Verifies citations with a policy-gated source matrix. Cloud defaults prioritize managed providers; licensed and BYO providers can be enabled where permitted.

Quote-Lock

SHA-256 cryptographic hash locks quoted material to its source, preventing AI fabrication of quotes.

Semantic Support

Dual-signal embedding analysis verifies the citation actually supports the claim. Disagreement detection flags uncertain results.

E-Filing Precheck

Pre-check documents for e-filing compliance with PASS/WARN/FAIL grading, formatting checks, and duplicate detection.

Batch Processing

Submit multiple citations for async validation. Returns a job_id immediately. Poll for results as they complete.

Auto-Correction

Suggests corrections for failed citations: reporter typos, missing periods, section symbols, and formatting issues.

Evidence Bundles

Cryptographically signed JSON bundles with provenance, input, output, user context, and metadata for audit trails.

Continuous Monitoring

Register cited authorities for ongoing monitoring. QiXCite re-checks on schedule and alerts via webhook when a citation's status changes.

Outcome Visibility

Every citation gets a clear PASS/WARN/FAIL/BLOCK label with reason and recommended action. Document summary shows overall risk and top issues.

Authority Alignment BETA

LLM NLI verifies that cited authorities actually support the claims being made. Two-stage pipeline: Claude Haiku (fast triage) + Claude Sonnet (quality review). 10 free checks per tenant via IOA Cloud.

API Quick Examples

Copy-paste examples for core QiXCite operations. All endpoints available at http://localhost:8000 (local) or https://api.orchintel.com/api/qixcite (cloud).

Document Scan POST /v1/documents/scan
curl -X POST http://localhost:8000/v1/documents/scan \
  -H "Content-Type: application/json" \
  -d '{"text": "As held in Miranda v. Arizona, 384 U.S. 436 (1966)..."}'
Validate Citation POST /v1/cite/validate
curl -X POST http://localhost:8000/v1/cite/validate \
  -H "Content-Type: application/json" \
  -d '{"citation": "384 U.S. 436", "jurisdiction": "US-FED", "audit_id": "test-1"}'
Quote-Lock POST /v1/cite/quote-lock
curl -X POST http://localhost:8000/v1/cite/quote-lock \
  -H "Content-Type: application/json" \
  -d '{"quote": "the prosecution may not use statements...", "source_url": "384 U.S. 436", "audit_id": "test-1"}'
Full Verify Pipeline POST /verify
curl -X POST http://localhost:8000/verify \
  -H "Content-Type: application/json" \
  -d '{"text": "...", "jurisdiction": "US-FED", "require_corroboration": true}'
E-Filing Precheck POST /v1/efiling/precheck
curl -X POST http://localhost:8000/v1/efiling/precheck \
  -H "Content-Type: application/json" \
  -d '{"text": "...", "jurisdiction": "US-FED", "court": "district"}'
Batch Processing POST /v1/batch
curl -X POST http://localhost:8000/v1/batch \
  -H "Content-Type: application/json" \
  -d '{"citations": ["384 U.S. 436", "347 U.S. 483"], "jurisdiction": "US-FED"}'
Continuous Monitoring POST /v1/monitor
curl -X POST http://localhost:8000/v1/monitor \
  -H "Content-Type: application/json" \
  -d '{"citations": ["384 U.S. 436"], "jurisdiction": "US-FED", "check_interval": "daily"}'
Evidence Bundle Sample GET /v1/evidence/sample
curl http://localhost:8000/v1/evidence/sample
Authority Alignment Beta POST /v1/authority-alignment
curl -X POST http://localhost:8000/v1/authority-alignment \
  -H "Content-Type: application/json" \
  -H "X-Tenant-ID: my-firm" \
  -d '{
    "citation": "Miranda v. Arizona, 384 U.S. 436 (1966)",
    "context_text": "Police must give Miranda warnings before any interrogation.",
    "audit_id": "audit-aae-demo",
    "user_llm_api_key": null
  }'

# Response includes verdict (aligned/misaligned/partial), alignment_score, rationale,
# authority_excerpt, authority_excerpt_hash (SHA-256), and quota_status.

Supported Jurisdictions

QiXCite supports citations from multiple jurisdictions and legal systems.

Region Jurisdictions Citation Types
United States US-FED, US-CA, US-NY, US-TX, US-IL, US-MA Case law (U.S., F.2d/3d/4th, S. Ct., L. Ed.), statutes (U.S.C.), regulations (C.F.R.)
Australia AU, AU-NSW, AU-QLD, AU-VIC, AU-WA, AU-SA, AU-TAS Official legislation/title existence checks in cloud default path; additional treatment depends on licensed/private sources
United Kingdom UK, UK-ENG, UK-SCT, UK-NIR Conditional by licensing/policy gate (managed cloud may restrict computational analysis paths)
Canada CA, CA-FED, CA-ON, CA-BC, CA-AB, CA-SK, CA-MB, CA-QC, CA-NB, CA-NS, CA-NL, CA-PE Conditional by licensing/policy gate (managed cloud may restrict CanLII automation paths)
European Union EU, EU-CJ, EU-GC Case citations, GDPR articles, regulations, directives
New Zealand NZ Official-source existence checks (courts + legislation). Treatment remains source/coverage dependent.
Other ZA, IN, SG, HK Various national formats (extraction only; provider adapters planned)

Governance Runtime Modes

QiXCite integrates with IOA Proxy for three governance runtime modes. Choose how strictly to govern AI-generated content.

SHADOW

Pilot programs, internal research

Observe AI-generated content without blocking attorneys. QiXCite logs every verdict silentlyβ€”see what would have been flagged without disrupting workflow.

CONSENSUS

Legal research, brief drafting

Require corroboration across providers before accepting authority. Returns 409 Conflict with repair suggestions if verification is uncertainβ€”never silent failures.

ENFORCE

Court filings, regulatory submissions

Block submissions when hallucinations or invalid authority are detected. Production-grade protection for documents where bad citations mean sanctions.

Anti-Hallucination Design

QiXCite uses multiple layers to prevent and detect citation hallucinations.

1

Existence Verification

Citations are checked against jurisdiction-aware sources. Each result exposes treatment provenance as user-visible labels: Licensed Treatment, Treatment Insight (Estimated β€’ Beta), or Treatment Not Available.

2

Dual-Signal Semantic Support

Two embedding models independently assess whether the source text supports the citation context. If models disagree (disagreement > 0.15), the verdict is "Uncertain" regardless of individual scores.

3

Corroboration Requirement

When require_corroboration: true (default), both the adapter validation AND semantic support must pass. A citation that exists but isn't semantically supported is flagged.

4

Quote-Lock Integrity

SHA-256 hashes lock quoted text to its source. Any modification to the quote or its attribution is detectable.

5

Auto-Correction Suggestions

When a citation fails, QiXCite suggests corrections (typos, formatting, missing components) with confidence scores, reducing false positives from formatting issues.

6

Authority Alignment Beta

Even when a citation exists and passes existence/semantic checks, it may be misused β€” cited for a proposition it does not support. Authority Alignment uses LLM-based natural language inference (Claude Haiku + Sonnet) to verify that the cited authority actually supports the claim in context. Catches citation misappropriation, over-broad readings, wrong-jurisdiction errors, and cherry-picking. Includes immutable SHA-256 excerpt hash for audit verification.

Try It: Sample Citations

Test QiXCite with these sample citations in the Console Playground or via the API.

Citation Expected Result
Miranda v. Arizona, 384 U.S. 436 (1966) Valid — US Supreme Court landmark
Brown v. Board of Education, 347 U.S. 483 (1954) Valid — Desegregation ruling
Obergefell v. Hodges, 576 U.S. 644 (2015) Valid — Marriage equality
999 U.S. 999 Unknown — Citation not found
1 Umbrella 100 Unknown — Invalid reporter

Continuous Monitoring

QiXCite can continuously monitor citations in your filed documents. If a cited authority is overruled, questioned, or retracted after filing, QiXCite detects the change and notifies you via webhook.

1

Register Citations

POST /v1/monitor with your citations, jurisdiction, and optional webhook URL. QiXCite begins tracking at your chosen interval (daily, weekly, monthly).

2

Automatic Re-Checks

QiXCite re-validates each monitored citation on schedule. Any status change (e.g., "valid" to "overruled") is recorded and triggers an alert.

3

Status Change Alerts

When a citation's status changes, QiXCite sends a webhook notification with the old status, new status, timestamp, and full validation details.

How QiXCite Compares

QiXCite is the only citation governance tool that combines anti-hallucination verification, cryptographic evidence bundles, and continuous monitoring in a single platform.

Feature QiXCite CiteCheck AI Pelaikan
Anti-hallucination verification 6-layer pipeline (existence + semantic + corroboration + quote-lock + evidence + authority alignment) Existence check only Existence check only
Dual-signal semantic support Primary + fallback embedding models with disagreement detection No No
Cryptographic evidence bundles SHA-256 signed JSON bundles with provenance and audit trails No No
Continuous monitoring Register citations, auto re-check, webhook alerts on status changes No No
International coverage Policy-gated matrix (US/EU managed defaults; AU/NZ existence-first; licensed enterprise paths) US only UK only
Quote integrity lock SHA-256 cryptographic hash No No
Auto-correction Suggests fixes for formatting, reporter typos, missing periods Limited No
E-filing precheck PASS/WARN/FAIL grading with court-specific rules No No
Authority alignment (Beta) LLM NLI claim-to-authority check; misalignment taxonomy; 6 misalignment types detected No No
Word Add-in Full sidebar with scan, validate, e-file, risk assessment, and authority alignment Basic No
Deployment Cloud (IOA) or Local/On-Premise Cloud only Cloud only
Batch processing Async batch with job polling No No
Free tier Local mode is fully free; Cloud has free tier Limited free Paid only

Academic Citation Governance

The same governance engine that validates legal citations now validates academic and scientific references. Same PASS/WARN/FAIL/BLOCK outcomes, same evidence bundles, same APIβ€”different authority providers.

Academic capabilities include:

  • DOI resolution via CrossRef and DataCite
  • PubMed / PMC article existence and metadata verification
  • Retraction Watch integration for retracted paper detection
  • arXiv paper verification and metadata
  • Citation formatting validation (APA, Bluebook, OSCOLA, AGLC)

Set jurisdiction: "ACADEMIC" in your API calls to use academic providers. The same endpoint that validates case law validates research citations.

Authority Alignment Beta

Authority Alignment answers the question existence/semantic checks cannot: does the cited authority actually support the claim? A citation can exist, pass configured treatment checks, and be semantically similar β€” yet still be misused. Schenck v. United States cited to argue all speech is protected is a classic example.

1. ClaimExtractor

Sentence-boundary detection finds the proposition associated with each citation in the document text.

2. AuthorityRetriever

Fetches authority text from CourtListener (free, 18M+ opinions). Results cached 24 hours to minimise API calls.

3. AlignmentChecker

Stage 1: Claude Haiku (fast/cheap). Stage 2: Claude Sonnet escalated for partial or uncertain results (score < 0.7).

4. Evidence Bundle

Verdict, score, rationale, authority excerpt, SHA-256 excerpt hash, and misalignment type β€” all recorded in the evidence bundle.

Verdict Types

Aligned

Cited authority directly supports the claim. Score ≥ 0.7.

Partial

Authority partially supports the claim; limiting language applies.

Misaligned

Authority does not support or contradicts the claim. One of 6 misalignment types identified.

Insufficient

Retrieved text too short or unavailable to assess alignment.

Retrieval Failed

CourtListener did not return a result for this citation.

Quota Exceeded

Plan quota reached. Sign in to continue in managed cloud mode, or configure BYO model keys only if your deployment requires that option.

Misalignment Taxonomy

TypeDescriptionExample
over_broadClaim states more than authority supportsCiting Schenck for absolute free speech
wrong_jurisdictionAuthority from a different jurisdictionUK Court of Appeal cited as US precedent
temporalAuthority overruled or supersededCiting pre-Lawrence cases on sodomy law
factualFactual claim contradicted by authorityMisquoting statutory damages amount
legal_standardWrong legal test appliedStrict scrutiny where rational basis is correct
cherry_pickedKey limiting language in authority ignoredOmitting the "unless" clause from a holding

Free Tier & Cost

Each tenant receives 10 complimentary Authority Alignment checks on IOA Cloud LLMs. Beyond that, use cloud quota or optionally configure your own Anthropic/OpenAI key in Settings.

ModelStageCost per citation
Claude Haiku 4.5Fast triage (most checks)~$0.004
Claude Sonnet 4.5Quality review (uncertain cases)~$0.016
Blended averageβ€”~$0.008

Word Add-in β€” Interactive Preview

The QiXCite sidebar integrates directly into Microsoft Word. Click the tabs below to explore the interface β€” including the Authority Alignment panel.

Brief_v3.docx — Microsoft Word

I. INTRODUCTION

The defendant's Fifth Amendment rights were violated when officers proceeded with interrogation without first providing the required warnings. As established by Miranda v. Arizona, 384 U.S. 436 (1966), suspects must be informed of their rights prior to custodial interrogation.

Evidence obtained in violation of the exclusionary rule must be suppressed. Mapp v. Ohio, 367 U.S. 643 (1961) extended this protection to state proceedings.

Q
QiXCite
Managed by IOA Cloud
CLOUD
2 citations found 2 PASS
Miranda v. Arizona, 384 U.S. 436 (1966) PASS
CourtListener • US Supreme Court
Treatment: Licensed Treatment
AUTHORITY ALIGNMENT (BETA) Aligned 94%
Authority directly establishes the obligation to provide warnings prior to custodial interrogation.
"Prior to any questioning, the person must be warned that he has a right to remain silent..."
Mapp v. Ohio, 367 U.S. 643 (1961) PASS
CourtListener • US Supreme Court
Treatment: Treatment Insight (Estimated β€’ Beta)
AUTHORITY ALIGNMENT (BETA) Aligned 91%
Authority correctly extends the exclusionary rule to state courts via the Fourteenth Amendment.
"the exclusionary rule is an essential part of both the Fourth and Fourteenth Amendments..."
Authority Alignment checks used: 2 / 10 free — Open Settings
PASS
E-Filing Ready
2 citations verified • No formatting issues
Citation existence verified
No overruled authorities
Quote integrity verified
Authority Alignment: Aligned (2/2)
Formatting: Bluebook compliant
Evidence Bundle
SHA-256 signed • 2 citations • Authority Alignment included
Excerpt hash (Miranda):
e3b0c44298fc1c...

Authority Alignment Verdict Chips:

Aligned Authority supports the claim
Partial Authority partially supports; limiting language applies
Misaligned Authority contradicts or doesn't support the claim

The Word add-in installs from Microsoft AppSource or via sideloading. Run the web Playground at console.orchintel.com/qixcite.

Support & Resources

Getting Started

Full Getting Started Guide

IOA Console

console.orchintel.com

Sales Inquiries

orchintel.com/contact