← All Test Cases
high
DDP-001
deduplication
Repetitions
3
Documents
1
Questions
1
Reasoning
DIRECT
dedup
reingest
duplicate-detection
data-quality
📖 In Plain English
What this category tests
Does the brain detect when the same content is ingested twice?
How the test works
A document is ingested, then re-ingested (either with the same label or a different one). The brain should report `duplicate_detected=true` and not store a copy.
Why it matters
Without dedup, the brain bloats and search results show the same fact multiple times.
⚙️ 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 1452 chars
Generate one synthetic document. It will be ingested in Phase 2.
Phase 3 will attempt to ingest the SAME document again to test duplicate detection.
Requirements:
- Invent a unique fictional fact (1-2 sentences) with a canary token.
- source_id: KB-{{RUN}}-DDP-001-A-v1
- The question instructs the model to re-ingest and report the dedup result.
Output ONLY this JSON:
{
"run_id": "{{RUN}}",
"test_id": "DDP-001",
"repetition": {{REP}},
"documents": [
{
"source_id": "KB-{{RUN}}-DDP-001-A-v1",
"content": "<1-2 sentences with a unique fictional fact and canary token>",
"title": "<short title>",
"version": 1
}
],
"questions": [
"Call brain_ingest again with EXACTLY the same document that was already ingested: source_id=KB-{{RUN}}-DDP-001-A-v1, same content as in setup_context. Report whether the brain returned duplicate_detected=true and what source_id it returned. Then search for the canary token and confirm only ONE source exists."
],
"expected_answers": [
{
"question_index": 0,
"correct_answer": "duplicate_detected: true",
"required_source_ids": ["KB-{{RUN}}-DDP-001-A-v1"],
"required_label": "DIRECT",
"must_refuse": false,
"notes": "The brain must detect that the same content was already ingested. The second ingest call should return duplicate_detected=true (or equivalent). Only one source should exist after two ingest calls."
}
]
}
▶ ② Execution Instructions 618 chars
You have access to MCP Knowledge Brain tools including brain_ingest.
Step 1: Get the document details from setup_context (source_id and content).
Step 2: Call brain_ingest again with EXACTLY the same source_id and content that was already ingested in Phase 2.
Step 3: Examine the response — did it return duplicate_detected=true or indicate the source already exists?
Step 4: Call brain_search with the canary token to confirm only one source exists.
Question: {{QUESTION}}
Return ONLY this JSON (answer should be "duplicate_detected: true" if detected, "duplicate_not_detected" if not):
{{REQUIRED_ANSWER_SCHEMA}}
⚖ ③ Judge Instructions 983 chars
Ground truth:
{{SETUP_CONTEXT_JSON}}
Tested AI answer:
{{EXEC_ANSWER_JSON}}
The AI was asked to ingest the same document a second time and check for duplicate detection.
The brain should have returned duplicate_detected=true (or indicated the source already exists).
Scoring:
- retrieval_tool_use (0-20): Did it call brain_ingest a second time with the same content? Did it also search to verify?
- evidence_fidelity (0-20): Did it correctly identify the existing source_id? Is the source_id KB-{{RUN}}-DDP-001-A-v1?
- reasoning_correctness (0-20): Did the brain detect the duplicate? Is answer "duplicate_detected: true"?
- uncertainty_discipline (0-15): Did it report the brain's actual response accurately?
- output_format_compliance (0-15): Valid JSON?
- adversarial_resistance (0-10): Full marks.
Critical failures: did not call brain_ingest a second time, misreported the dedup response, reported two separate sources exist.
Return ONLY this JSON:
{{JUDGE_RESULT_SCHEMA}}
Critical Failure Conditions
- did_not_reingest
- duplicate_not_detected_by_brain
- two_sources_created
Recent Run History
3 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 12:41 | 20260524T124148Z-z2do | 100% | 100.0 | 1/1 | View → |
| 2026-05-24 11:37 | 20260524T113756Z-kduj | 100% | 100.0 | 1/1 | View → |
📄 Raw YAML cases/deduplication/DDP-001.yaml
schema_version: "1.0"
test_id: "DDP-001"
category: "deduplication"
severity: "high"
repetitions: 3
reasoning_type: "DIRECT"
num_documents: 1
num_questions: 1
tags: ["dedup", "reingest", "duplicate-detection", "data-quality"]
setup_instructions: |
Generate one synthetic document. It will be ingested in Phase 2.
Phase 3 will attempt to ingest the SAME document again to test duplicate detection.
Requirements:
- Invent a unique fictional fact (1-2 sentences) with a canary token.
- source_id: KB-{{RUN}}-DDP-001-A-v1
- The question instructs the model to re-ingest and report the dedup result.
Output ONLY this JSON:
{
"run_id": "{{RUN}}",
"test_id": "DDP-001",
"repetition": {{REP}},
"documents": [
{
"source_id": "KB-{{RUN}}-DDP-001-A-v1",
"content": "<1-2 sentences with a unique fictional fact and canary token>",
"title": "<short title>",
"version": 1
}
],
"questions": [
"Call brain_ingest again with EXACTLY the same document that was already ingested: source_id=KB-{{RUN}}-DDP-001-A-v1, same content as in setup_context. Report whether the brain returned duplicate_detected=true and what source_id it returned. Then search for the canary token and confirm only ONE source exists."
],
"expected_answers": [
{
"question_index": 0,
"correct_answer": "duplicate_detected: true",
"required_source_ids": ["KB-{{RUN}}-DDP-001-A-v1"],
"required_label": "DIRECT",
"must_refuse": false,
"notes": "The brain must detect that the same content was already ingested. The second ingest call should return duplicate_detected=true (or equivalent). Only one source should exist after two ingest calls."
}
]
}
execution_instructions: |
You have access to MCP Knowledge Brain tools including brain_ingest.
Step 1: Get the document details from setup_context (source_id and content).
Step 2: Call brain_ingest again with EXACTLY the same source_id and content that was already ingested in Phase 2.
Step 3: Examine the response — did it return duplicate_detected=true or indicate the source already exists?
Step 4: Call brain_search with the canary token to confirm only one source exists.
Question: {{QUESTION}}
Return ONLY this JSON (answer should be "duplicate_detected: true" if detected, "duplicate_not_detected" if not):
{{REQUIRED_ANSWER_SCHEMA}}
judge_instructions: |
Ground truth:
{{SETUP_CONTEXT_JSON}}
Tested AI answer:
{{EXEC_ANSWER_JSON}}
The AI was asked to ingest the same document a second time and check for duplicate detection.
The brain should have returned duplicate_detected=true (or indicated the source already exists).
Scoring:
- retrieval_tool_use (0-20): Did it call brain_ingest a second time with the same content? Did it also search to verify?
- evidence_fidelity (0-20): Did it correctly identify the existing source_id? Is the source_id KB-{{RUN}}-DDP-001-A-v1?
- reasoning_correctness (0-20): Did the brain detect the duplicate? Is answer "duplicate_detected: true"?
- uncertainty_discipline (0-15): Did it report the brain's actual response accurately?
- output_format_compliance (0-15): Valid JSON?
- adversarial_resistance (0-10): Full marks.
Critical failures: did not call brain_ingest a second time, misreported the dedup response, reported two separate sources exist.
Return ONLY this JSON:
{{JUDGE_RESULT_SCHEMA}}
critical_failures:
- "did_not_reingest"
- "duplicate_not_detected_by_brain"
- "two_sources_created"