India Macro Intelligence

Runtime dataResearch workspace

Agent Evaluation

Reliability benchmark design for existing agents; no scores are fabricated.

Agent Reliability Benchmark

Benchmark dataset format supports research question, expected/actual tools, result, warnings, runtime, and model version.

NO BENCHMARK RUN

Inflation

Growth

External Sector

Forecasting

Policy

Missing Data

Wrong Units

Stale Observations

Base-Year Conflicts

Extreme Scenarios

Unsupported Causal Claims

Tool Failures

Target benchmark size: 50–100 research questions. No question set or evaluation run is present yet.

Evaluation dimensions

Tool Selection Accuracy

—

Numerical Faithfulness

—

Citation Accuracy

—

Schema Validity

—

Unsupported Claims

—

Failure Handling

—

Benchmark run records

Run ID, agent, question, expected tools, actual tools, result, warnings, runtime, and model version will be persisted when evaluation execution is implemented.

PLANNED

No validated output is available. This module will display data after its backend integration is ready.

Loading research workspace

Preparing the mock dashboard surface...