dcl-semantic-drift-guard

name

dcl-semantic-drift-guard

description

Use this skill to detect semantic hallucinations and context drift in LLM outputs. Triggers when an agent or pipeline needs to verify that a generated response is faithfully grounded in a source document or knowledge base — and has not fabricated, contradicted, or materially distorted any claims. Use whenever you need to validate AI output against a contract, policy document, RAG-retrieved context, or any authoritative source of truth. Returns a tamper-evident DCL audit record with verdict IN_COMMIT or HALLUCINATION_DRIFT. Part of the Fronesis Labs / Leibniz Layer™ verification suite alongside DCL Policy Enforcer and DCL Sentinel Trace.

DCL Semantic Drift Guard

Publisher: @daririnch · Fronesis Labs
Version: 1.0.0
Part of: Leibniz Layer™ Verification Suite

---

What this skill does

Semantic Drift Guard compares an LLM-generated response against a trusted source of truth and detects:

• Hallucinated facts — claims not present in the source
• Logical contradictions — statements that directly conflict with the source
• Omission drift — critical information from the source that was silently dropped
• Fabricated specifics — invented numbers, dates, names, clauses, or identifiers

It supports two source modes:

• context mode — inline document or contract passed directly in the request
• kb_query mode — knowledge base lookup via RAG endpoint

Every verification produces a cryptographic audit record compatible with the DCL Evaluator tamper-evident chain.

---

Verdicts

Verdict	Meaning
IN_COMMIT	Response is faithfully grounded in the source. No hallucinations detected. Safe to proceed.
HALLUCINATION_DRIFT	Response contains fabricated, contradicted, or unsupported claims. Do not commit. Review drift_items.

---

Input schema

{
  "source_mode": "context" | "kb_query",

  // For source_mode = "context":
  "source_document": "<full text of the authoritative document>",

  // For source_mode = "kb_query":
  "kb_endpoint": "<RAG endpoint URL>",
  "kb_query": "<query string to retrieve relevant chunks>",

  // Always required:
  "llm_output": "<the LLM-generated response to verify>",
  "strictness": "strict" | "balanced" | "lenient",  // default: "balanced"
  "policy": "eu_ai_act" | "gdpr" | "fstek" | "internal" | "none"  // optional
}

Strictness levels

• strict — any unverifiable claim triggers HALLUCINATION_DRIFT. Use for contracts, medical, legal, financial outputs.
• balanced — minor paraphrasing and reasonable inferences are tolerated. Use for customer support, summaries.
• lenient — only direct factual contradictions trigger HALLUCINATION_DRIFT. Use for creative or exploratory outputs.

---

Output schema

{
  "status": "success" | "error",
  "data": {
    "verdict": "IN_COMMIT" | "HALLUCINATION_DRIFT",
    "confidence": 0.0–1.0,
    "source_mode": "context" | "kb_query",
    "strictness": "strict" | "balanced" | "lenient",
    "policy": "eu_ai_act" | "none" | "...",
    "drift_items": [
      {
        "type": "hallucination" | "contradiction" | "omission" | "fabricated_specific",
        "claim": "<the problematic claim in the LLM output>",
        "source_reference": "<relevant excerpt from source, or null if absent>",
          |  | 
      
    
     
     
     
  

drift_items is an empty array [] when verdict is IN_COMMIT.

---

Verification workflow

When this skill is invoked, follow these steps:

Step 1 — Retrieve source of truth

If source_mode = "context":
Use source_document directly. Chunk it into logical sections for comparison.

If source_mode = "kb_query":
Query the kb_endpoint with kb_query. Retrieve top-k relevant chunks. Treat the union of retrieved chunks as the authoritative source. If the endpoint is unreachable, return status: "error" with reason: "kb_unavailable".

Step 2 — Decompose LLM output into claims

Parse the llm_output into atomic, verifiable claims:

• Factual assertions ("The contract states X")
• Numerical values ("The penalty is €10,000")
• Named entities ("The responsible party is Company A")
• Temporal claims ("The deadline is March 15")
• Logical conclusions ("Therefore, clause 4.2 applies")

Step 3 — Cross-reference each claim against source

For each claim, determine:

Finding	Classification
Claim is explicitly supported by source	✅ Grounded
Claim is a reasonable paraphrase (strictness: lenient/balanced)	✅ Grounded
Claim introduces information absent from source	⚠️ hallucination
Claim directly contradicts source	🚨 contradiction
Critical source information was omitted from output	⚠️ omission
Specific value (number, date, name) was invented	🚨 fabricated_specific

Step 4 — Apply strictness filter

• strict: any ⚠️ or 🚨 → HALLUCINATION_DRIFT
• balanced: any 🚨, or multiple ⚠️ → HALLUCINATION_DRIFT
• lenient: only 🚨 contradiction or fabricated_specific → HALLUCINATION_DRIFT

Step 5 — Compute audit record

Generate:

tx_hash = SHA-256(source_fingerprint + llm_output + verdict + timestamp)
audit_chain_id = Merkle leaf position in DCL Evaluator chain

Return the full output schema.

---

Interpreting results

IN_COMMIT — safe to proceed

{
  "status": "success",
  "data": {
    "verdict": "IN_COMMIT",
    "confidence": 0.97,
    "drift_items": [],
    "tx_hash": "0xa3f1...c72e",
    "timestamp": "2026-04-09T14:22:00Z",
    "audit_chain_id": "dcl-leaf-0047"
  }
}

The LLM output is faithfully grounded in the source. Log tx_hash to your audit trail.

HALLUCINATION_DRIFT — do not commit

{
  "status": "success",
  "data": {
    "verdict": "HALLUCINATION_DRIFT",
    "confidence": 0.89,
    "drift_items": [
      {
        "type": "fabricated_specific",
        "claim": "The penalty for breach is €50,000.",
        "source_reference": "Section 8.3: The penalty shall not exceed €10,000.",
        "severity": "critical"
      },
      {
        "type": "hallucination",
        "claim": "The agreement includes a 90-day cooling-off period.",
        "source_reference": null,
        "severity": "major"
      }
    
     
     
     
  

Block the output. Surface drift_items to the human reviewer or trigger a re-generation loop.

---

Integration patterns

With DCL Policy Enforcer (recommended pipeline)

Run Policy Enforcer first (jailbreak / compliance check), then Semantic Drift Guard (factual grounding):

LLM Output
    │
    ▼
DCL Policy Enforcer ──► REJECT? → Block immediately
    │ COMMIT
    ▼
DCL Semantic Drift Guard ──► HALLUCINATION_DRIFT? → Block / re-generate
    │ IN_COMMIT
    ▼
Safe to deliver

Both tx_hash values are logged to the same DCL Evaluator audit chain, giving end-to-end verifiability.

With DCL Sentinel Trace (full Leibniz Layer™ stack)

Sentinel Trace → strip PII before source reaches LLM
Policy Enforcer → compliance check on output
Semantic Drift Guard → factual grounding check

Standalone (quick RAG validation)

result = dcl_semantic_drift_guard(
    source_mode="kb_query",
    kb_endpoint="https://kb.yourapp.com/query",
    kb_query="penalty clauses breach of contract",
    llm_output=agent_response,
    strictness="strict",
    policy="eu_ai_act"
)

if result["data"]["verdict"] == "HALLUCINATION_DRIFT":
    raise ValueError(f"Drift detected: {result['data']['drift_items']}")

---

Use cases

Domain	Source mode	Strictness	Why
Legal contract summarization	context	strict	Fabricated clauses = liability
RAG-based customer support	kb_query	balanced	Prevent wrong product info
Medical documentation	context	strict	Patient safety
Financial report generation	context	strict	Regulatory compliance
EU AI Act compliance auditing	kb_query	strict	FSTEK / AI Act article mapping
Internal knowledge assistant	kb_query	lenient	Lower stakes, exploratory

---

Compliance notes

• Audit records are compatible with EU AI Act Article 12 (logging requirements for high-risk AI systems)
• tx_hash chain is admissible as tamper-evident evidence under GDPR Article 5(2) accountability principle
• All source documents processed in context mode are never stored — only their fingerprint is hashed
• Compatible with FSTEK audit trail requirements for AI systems in Russian regulated industries

---

Privacy & Data Policy

This skill is operated by Fronesis Labs under a strict no-retention data policy.

What is processed: Only the text submitted for evaluation. No user identity, no API keys, no metadata beyond what is required to run the verification.

Retention: Evaluations are processed in-memory only. No text is written to disk, no logs are retained, no data is shared with third parties. The only persistent record is the cryptographic tx_hash and chain_hash — these contain no personal data.

Source documents: Content passed via source_document (context mode) is never stored or logged. Only a cryptographic fingerprint is included in the audit hash.

Infrastructure: Webhook hosted on a private VPS operated solely by Fronesis Labs. No cloud analytics, no third-party processors.

Full policy: https://fronesislabs.com/#privacy · Questions: [e-mail]

---

Related skills

• dcl-policy-enforcer — Compliance and jailbreak detection (run before Drift Guard)
• dcl-sentinel-trace — PII redaction and identity exposure detection (run before source reaches LLM)

Leibniz Layer™ · Fronesis Labs · fronesislabs.com