← All Test Cases

NUM-003

numerical

high

Repetitions

Documents

Questions

Reasoning

INFERRED

threshold-reasoning numerical-inference inferred

📖 In Plain English

What this category tests

Can the brain compare and aggregate numbers across documents?

How the test works

Documents contain quantitative facts. The test asks for comparisons ('which is largest?', 'what's the total?') requiring numeric reasoning across multiple sources.

Why it matters

Many real questions involve numbers — figures, dates, counts, percentages.

⚙️ How a single rep runs

① Generate

Model creates 2 synthetic documents 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 2 documents

→ 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 1359 chars

Generate two documents — one stating a threshold and one stating a value. Ask if the value exceeds the threshold.

Requirements:
- Doc A: "The maximum allowed <metric> is <Threshold>." (e.g., 100)
- Doc B: "The current <metric> reading is <Value>." (e.g., 143)
- Value must clearly exceed OR be under the threshold (choose one, make it obvious).
- source_ids: KB-{{RUN}}-NUM-003-A-v1 and KB-{{RUN}}-NUM-003-B-v1.
- Question: "Is the current <metric> within the allowed limit?"

Output ONLY this JSON:
{
  "run_id": "{{RUN}}",
  "test_id": "NUM-003",
  "repetition": {{REP}},
  "documents": [
    {"source_id": "KB-{{RUN}}-NUM-003-A-v1", "content": "The maximum allowed <metric> is <Threshold>.", "title": "<title>", "version": 1},
    {"source_id": "KB-{{RUN}}-NUM-003-B-v1", "content": "The current <metric> reading is <Value>.", "title": "<title>", "version": 1}
  ],
  "questions": ["Is the current <metric> within the allowed limit?"],
  "expected_answers": [{
    "question_index": 0,
    "correct_answer": "<Yes/No — based on Value vs Threshold>",
    "required_source_ids": ["KB-{{RUN}}-NUM-003-A-v1", "KB-{{RUN}}-NUM-003-B-v1"],
    "required_quote": null,
    "required_label": "INFERRED",
    "must_refuse": false,
    "notes": "Requires comparing Value against Threshold across two docs. INFERRED because comparison is not directly stated."
  }]
}

▶ ② Execution Instructions 144 chars

Use ONLY MCP Knowledge Brain retrieval tools with domain="brain-test".
Question: {{QUESTION}}
Return ONLY this JSON:
{{REQUIRED_ANSWER_SCHEMA}}

⚖ ③ Judge Instructions 360 chars

Ground truth:
{{SETUP_CONTEXT_JSON}}

Tested AI answer:
{{EXEC_ANSWER_JSON}}

Check: correct Yes/No based on arithmetic, both source_ids cited, INFERRED label (comparison not stated directly).
Critical failures: wrong answer, missing source, DIRECT label (comparison requires inference), hallucinated source_id.

Return ONLY this JSON:
{{JUDGE_RESULT_SCHEMA}}

Critical Failure Conditions

wrong_threshold_comparison
missing_required_source_id
hallucinated_source_id
wrong_label_direct_instead_of_inferred

Recent Run History

2 runs

When	Run ID	Pass Rate	Avg Score	Reps
2026-05-24 13:08	20260524T130808Z-kqze	100%	95.0	1/1	View →
2026-05-24 11:37	20260524T113756Z-kduj	100%	97.0	1/1	View →

📄 Raw YAML cases/numerical/NUM-003.yaml

schema_version: "1.0"
test_id: "NUM-003"
category: "numerical"
severity: "high"
repetitions: 3
reasoning_type: "INFERRED"
num_documents: 2
num_questions: 1
tags: ["threshold-reasoning", "numerical-inference", "inferred"]

setup_instructions: |
  Generate two documents — one stating a threshold and one stating a value. Ask if the value exceeds the threshold.

  Requirements:
  - Doc A: "The maximum allowed <metric> is <Threshold>." (e.g., 100)
  - Doc B: "The current <metric> reading is <Value>." (e.g., 143)
  - Value must clearly exceed OR be under the threshold (choose one, make it obvious).
  - source_ids: KB-{{RUN}}-NUM-003-A-v1 and KB-{{RUN}}-NUM-003-B-v1.
  - Question: "Is the current <metric> within the allowed limit?"

  Output ONLY this JSON:
  {
    "run_id": "{{RUN}}",
    "test_id": "NUM-003",
    "repetition": {{REP}},
    "documents": [
      {"source_id": "KB-{{RUN}}-NUM-003-A-v1", "content": "The maximum allowed <metric> is <Threshold>.", "title": "<title>", "version": 1},
      {"source_id": "KB-{{RUN}}-NUM-003-B-v1", "content": "The current <metric> reading is <Value>.", "title": "<title>", "version": 1}
    ],
    "questions": ["Is the current <metric> within the allowed limit?"],
    "expected_answers": [{
      "question_index": 0,
      "correct_answer": "<Yes/No — based on Value vs Threshold>",
      "required_source_ids": ["KB-{{RUN}}-NUM-003-A-v1", "KB-{{RUN}}-NUM-003-B-v1"],
      "required_quote": null,
      "required_label": "INFERRED",
      "must_refuse": false,
      "notes": "Requires comparing Value against Threshold across two docs. INFERRED because comparison is not directly stated."
    }]
  }

execution_instructions: |
  Use ONLY MCP Knowledge Brain retrieval tools with domain="brain-test".
  Question: {{QUESTION}}
  Return ONLY this JSON:
  {{REQUIRED_ANSWER_SCHEMA}}

judge_instructions: |
  Ground truth:
  {{SETUP_CONTEXT_JSON}}

  Tested AI answer:
  {{EXEC_ANSWER_JSON}}

  Check: correct Yes/No based on arithmetic, both source_ids cited, INFERRED label (comparison not stated directly).
  Critical failures: wrong answer, missing source, DIRECT label (comparison requires inference), hallucinated source_id.

  Return ONLY this JSON:
  {{JUDGE_RESULT_SCHEMA}}

critical_failures:
  - "wrong_threshold_comparison"
  - "missing_required_source_id"
  - "hallucinated_source_id"
  - "wrong_label_direct_instead_of_inferred"