OpenAI Agents SDK Research Misconduct Audit System
Scientific misconduct investigations require cross-referencing multi-year email archives with peer-reviewed figure data. Build an OpenAI Agents SDK system with multi-turn tool calling to cross-examine digital correspondence against image manipulation reports. Achieve high precision in flagging discrepancies without false positives.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an autonomous investigative agent system using OpenAI Agents SDK to cross-reference scientific correspondence with image manipulation logs.
How work is evaluated
Evaluates temporal discrepancy identification, evidence mapping accuracy, and agent tool execution efficiency.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
discrepancy_detected
Verifies that true timeline contradictions are flagged.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
forensic_precision
Precision of identified timeline contradictions. • target: 0.94 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Implement OpenAI Agents SDK multi-turn agent with custom tool bindings
Cross-examine timeline events between email metadata and image duplication logs
Build dynamic graph visualization tools for institutional communication logs
Output structured forensic investigation timelines with evidence confidence scores
Reference links and supporting material
Synthetic investigation archives containing simulated email logs, committee meeting minutes, and image analysis reports.
Challenge setup
Evaluates temporal discrepancy identification, evidence mapping accuracy, and agent tool execution efficiency.
Challenge input
JSON containing emails array and figure_reports array
OpenAI Agents SDK
Required framework for building multi-turn autonomous tool agents.
GPT-4o
Provides tool execution and reasoning capabilities for digital forensics.
Evaluated output
JSON with identified_discrepancies array and overall_verdict
- Verifies that true timeline contradictions are flagged.
- Precision of identified timeline contradictions. • target: 0.94 • range: 0-1
- Benchmark: ResearchIntegrity-Bench
- Forensic Precision target: 0.94
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- OpenAI · Required
- Exa · Optional
- Oracle OCI Generative AI · Optional
Evaluation contract
- discrepancy_detected · Weight 1
- forensic_precision · Weight 1
Metrics
- forensic_precision · Target 0.94
Benchmarks
- ResearchIntegrity-Bench
- biotech
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing emails array and figure_reports array
Output: JSON with identified_discrepancies array and overall_verdict
Execution harness: Python
Evaluation tools
- OpenAI Agents SDK · Agent core framework
- GPT-4o · LLM Provider
Recipe freeze marker
No recipe freeze time is recorded.