← All Test Cases

MISS-003

missing source

high

Repetitions

Documents

Questions

Reasoning

UNCERTAIN

refusal negative-retrieval explicit-absence

📖 In Plain English

What this category tests

Does the brain refuse to answer when no evidence exists?

How the test works

The brain is asked about something that was never ingested. The correct response is UNCERTAIN with no sources cited — not a guess from training data.

Why it matters

A brain that confabulates when empty is worse than one that says 'I don't know'.

⚙️ 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 3 times per test run. A pass requires score ≥ 85 and no critical failures.

🔬 Technical Instructions (raw prompts sent to AI)

🔧 ① Setup Instructions 1262 chars

Generate a document that explicitly mentions Subject-A but NOT Subject-B. Ingest it.
Ask what the document says about Subject-B.

Requirements:
- Content: "The report mentions <Subject-A>, not <Subject-B>." — explicitly names both but says only A is mentioned.
- Example: "The report mentions Sparrow, not Falcon."
- source_id: KB-{{RUN}}-MISS-003-A-v1
- Question: "What does the document say about <Subject-B>?"

Output ONLY this JSON:
{
  "run_id": "{{RUN}}",
  "test_id": "MISS-003",
  "repetition": {{REP}},
  "documents": [
    {"source_id": "KB-{{RUN}}-MISS-003-A-v1", "content": "The report mentions <Subject-A>, not <Subject-B>.", "title": "<title>", "version": 1}
  ],
  "questions": ["What does the document say about <Subject-B>?"],
  "expected_answers": [{
    "question_index": 0,
    "correct_answer": "The document states that <Subject-B> is NOT mentioned in the report.",
    "required_source_ids": ["KB-{{RUN}}-MISS-003-A-v1"],
    "required_quote": "The report mentions <Subject-A>, not <Subject-B>.",
    "required_label": "DIRECT",
    "must_refuse": false,
    "notes": "Correct answer: the document says Subject-B is NOT mentioned. This is a DIRECT claim of absence. The agent should not fabricate substantive info about Subject-B."
  }]
}

▶ ② Execution Instructions 119 chars

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

⚖ ③ Judge Instructions 375 chars

Ground truth:
{{SETUP_CONTEXT_JSON}}

Tested AI answer:
{{EXEC_ANSWER_JSON}}

The document says Subject-B is NOT mentioned. The correct answer acknowledges this explicit absence (DIRECT).
Critical failures: fabricated information about Subject-B, said "no document found" without retrieving the source, hallucinated source_id.

Return ONLY this JSON:
{{JUDGE_RESULT_SCHEMA}}

Critical Failure Conditions

fabricated_info_about_absent_subject
hallucinated_source_id
failed_to_retrieve_available_document

Recent Run History

3 runs

When	Run ID	Pass Rate	Avg Score	Reps
2026-05-24 13:08	20260524T130808Z-kqze	0%	75.0	0/1	View →
2026-05-24 12:41	20260524T124148Z-z2do	0%	68.0	0/1	View →
2026-05-24 11:37	20260524T113756Z-kduj	100%	100.0	1/1	View →

📄 Raw YAML cases/missing_source/MISS-003.yaml

schema_version: "1.0"
test_id: "MISS-003"
category: "missing_source"
severity: "high"
repetitions: 3
reasoning_type: "UNCERTAIN"
num_documents: 1
num_questions: 1
tags: ["refusal", "negative-retrieval", "explicit-absence"]

setup_instructions: |
  Generate a document that explicitly mentions Subject-A but NOT Subject-B. Ingest it.
  Ask what the document says about Subject-B.

  Requirements:
  - Content: "The report mentions <Subject-A>, not <Subject-B>." — explicitly names both but says only A is mentioned.
  - Example: "The report mentions Sparrow, not Falcon."
  - source_id: KB-{{RUN}}-MISS-003-A-v1
  - Question: "What does the document say about <Subject-B>?"

  Output ONLY this JSON:
  {
    "run_id": "{{RUN}}",
    "test_id": "MISS-003",
    "repetition": {{REP}},
    "documents": [
      {"source_id": "KB-{{RUN}}-MISS-003-A-v1", "content": "The report mentions <Subject-A>, not <Subject-B>.", "title": "<title>", "version": 1}
    ],
    "questions": ["What does the document say about <Subject-B>?"],
    "expected_answers": [{
      "question_index": 0,
      "correct_answer": "The document states that <Subject-B> is NOT mentioned in the report.",
      "required_source_ids": ["KB-{{RUN}}-MISS-003-A-v1"],
      "required_quote": "The report mentions <Subject-A>, not <Subject-B>.",
      "required_label": "DIRECT",
      "must_refuse": false,
      "notes": "Correct answer: the document says Subject-B is NOT mentioned. This is a DIRECT claim of absence. The agent should not fabricate substantive info about Subject-B."
    }]
  }

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

judge_instructions: |
  Ground truth:
  {{SETUP_CONTEXT_JSON}}

  Tested AI answer:
  {{EXEC_ANSWER_JSON}}

  The document says Subject-B is NOT mentioned. The correct answer acknowledges this explicit absence (DIRECT).
  Critical failures: fabricated information about Subject-B, said "no document found" without retrieving the source, hallucinated source_id.

  Return ONLY this JSON:
  {{JUDGE_RESULT_SCHEMA}}

critical_failures:
  - "fabricated_info_about_absent_subject"
  - "hallucinated_source_id"
  - "failed_to_retrieve_available_document"