← All Test Cases

SDR-003

semantic drift

high

Repetitions

Documents

Questions

Reasoning

DIRECT

semantic-search negation paraphrase-query

📖 In Plain English

What this category tests

Does the brain find documents when the query uses synonyms?

How the test works

A document uses one term ('authorized'); the query uses a synonym ('approved'). The brain must bridge the vocabulary gap and retrieve the document.

Why it matters

Users don't always know the exact phrasing in the brain. Synonym handling is essential for usability.

⚙️ How a single rep runs

① Generate

Model creates 1 synthetic document and 1 question with unique canary tokens

→ Fresh content per run prevents memorization and proves real retrieval

② Ingest (MCP)

Model calls brain_ingest to store the 1 document

→ Tests the brain's storage and indexing pipeline

③ Query (MCP)

Model answers the question using brain retrieval tools (search, fetch, context_pack, etc.)

→ Core test — does the brain return correct evidence and let the model build a faithful answer?

④ Evaluate

Model judges the answer against ground truth (the document it generated in phase 1)

→ Produces a score 0–100 with detailed sub-scores (retrieval, fidelity, reasoning, etc.)

This rep is run 5 times per test run. A pass requires score ≥ 85 and no critical failures.

🔬 Technical Instructions (raw prompts sent to AI)

🔧 ① Setup Instructions 1271 chars

Generate a document with a prohibition/restriction statement.
Generate a paraphrased yes/no question that asks about the opposite of the restriction.

Requirements:
- Document: "Access is forbidden after <time>." or "<Action> is prohibited during <condition>."
- Question: "Can users <action> past <time>?" or "Is <action> allowed during <condition>?" — phrased as a yes/no question using different vocabulary.
- No keyword overlap between document and question.
- source_id: KB-{{RUN}}-SDR-003-A-v1

Output ONLY this JSON:
{
  "run_id": "{{RUN}}",
  "test_id": "SDR-003",
  "repetition": {{REP}},
  "documents": [
    {"source_id": "KB-{{RUN}}-SDR-003-A-v1", "content": "<prohibition statement>", "title": "<title>", "version": 1}
  ],
  "questions": ["<paraphrased yes/no question about whether the action is allowed>"],
  "expected_answers": [{
    "question_index": 0,
    "correct_answer": "No — the document states this is forbidden/prohibited.",
    "required_source_ids": ["KB-{{RUN}}-SDR-003-A-v1"],
    "required_quote": null,
    "required_label": "DIRECT",
    "must_refuse": false,
    "notes": "Brain must: (1) find the document via semantic search, (2) return the prohibition clearly, (3) not silently answer 'yes' because the negation was missed."
  }]
}

▶ ② Execution Instructions 148 chars

Use ONLY MCP Knowledge Brain retrieval tools with semantic search enabled.
Question: {{QUESTION}}
Return ONLY this JSON:
{{REQUIRED_ANSWER_SCHEMA}}

⚖ ③ Judge Instructions 423 chars

Ground truth:
{{SETUP_CONTEXT_JSON}}

Tested AI answer:
{{EXEC_ANSWER_JSON}}

The brain must find the prohibition document via semantic search AND correctly convey the negation (no/forbidden).
Critical failures: answered "yes" (missed the negation), failed to find the document, hallucinated source_id.
Partial credit: found document but was ambiguous about the prohibition.

Return ONLY this JSON:
{{JUDGE_RESULT_SCHEMA}}

Critical Failure Conditions

answered_yes_negation_missed
semantic_search_failed_to_retrieve_document
hallucinated_source_id

Recent Run History

2 runs

When	Run ID	Pass Rate	Avg Score	Reps
2026-05-24 13:08	20260524T130808Z-kqze	100%	100.0	1/1	View →
2026-05-24 11:37	20260524T113756Z-kduj	100%	100.0	1/1	View →

📄 Raw YAML cases/semantic_drift/SDR-003.yaml

schema_version: "1.0"
test_id: "SDR-003"
category: "semantic_drift"
severity: "high"
repetitions: 5
reasoning_type: "DIRECT"
num_documents: 1
num_questions: 1
tags: ["semantic-search", "negation", "paraphrase-query"]

setup_instructions: |
  Generate a document with a prohibition/restriction statement.
  Generate a paraphrased yes/no question that asks about the opposite of the restriction.

  Requirements:
  - Document: "Access is forbidden after <time>." or "<Action> is prohibited during <condition>."
  - Question: "Can users <action> past <time>?" or "Is <action> allowed during <condition>?" — phrased as a yes/no question using different vocabulary.
  - No keyword overlap between document and question.
  - source_id: KB-{{RUN}}-SDR-003-A-v1

  Output ONLY this JSON:
  {
    "run_id": "{{RUN}}",
    "test_id": "SDR-003",
    "repetition": {{REP}},
    "documents": [
      {"source_id": "KB-{{RUN}}-SDR-003-A-v1", "content": "<prohibition statement>", "title": "<title>", "version": 1}
    ],
    "questions": ["<paraphrased yes/no question about whether the action is allowed>"],
    "expected_answers": [{
      "question_index": 0,
      "correct_answer": "No — the document states this is forbidden/prohibited.",
      "required_source_ids": ["KB-{{RUN}}-SDR-003-A-v1"],
      "required_quote": null,
      "required_label": "DIRECT",
      "must_refuse": false,
      "notes": "Brain must: (1) find the document via semantic search, (2) return the prohibition clearly, (3) not silently answer 'yes' because the negation was missed."
    }]
  }

execution_instructions: |
  Use ONLY MCP Knowledge Brain retrieval tools with semantic search enabled.
  Question: {{QUESTION}}
  Return ONLY this JSON:
  {{REQUIRED_ANSWER_SCHEMA}}

judge_instructions: |
  Ground truth:
  {{SETUP_CONTEXT_JSON}}

  Tested AI answer:
  {{EXEC_ANSWER_JSON}}

  The brain must find the prohibition document via semantic search AND correctly convey the negation (no/forbidden).
  Critical failures: answered "yes" (missed the negation), failed to find the document, hallucinated source_id.
  Partial credit: found document but was ambiguous about the prohibition.

  Return ONLY this JSON:
  {{JUDGE_RESULT_SCHEMA}}

critical_failures:
  - "answered_yes_negation_missed"
  - "semantic_search_failed_to_retrieve_document"
  - "hallucinated_source_id"