ST: Research Configuration

Configure the API key and models used across experiments E2–E6. These settings define the shared technical environment for research. Changes affect future runs and are recorded for reproducibility.

Research Control Configuration changes apply only to future experiment runs. Every saved run records the exact models, currency rule and configuration version used.
Experiment Configuration
Model and Input Configuration — define the models, connection and input handling used in experiments.
Model used for E1 OCR and document understanding.
Model used for policy reasoning and verdicts in E2–E6.
Force a specific currency for verdict prompts, or "Any" to use the currency detected on the receipt.
Persist configuration for future runs.
Validate API key and model access.
Experiment Result Management
Export or delete saved experiment results.
Export Experiment Results

Export all saved experiment run records and associated research metadata as JSON.

Includes:
  • Experiment and run identifiers
  • Date and time
  • Receipt / source reference
  • Extracted fields (E1)
  • Policy variant and project phase
  • Model names and versions
  • Model outputs (raw and parsed)
  • Verdicts, reasons and confidence
  • Ambiguity / hallucination indicators
  • Configuration metadata used
Delete All Data

Permanently delete all saved experiment outputs for E1–E6 and ground-truth annotations.

This will delete:
  • Experiment run history (E1–E6)
  • Extracted results and cached extractions
  • Model outputs and raw responses
  • Verdicts, reasons and confidence
  • Ground-truth annotations (ground_truth.json)
This will NOT delete:
  • Policies and policy versions
  • API credentials
  • Configuration settings
  • System and application data
This action cannot be undone. A confirmation step with run count (0 saved runs) will be required before deletion.
Note:For audit and reproducibility, do not change configuration during an active evaluation cycle. Record any changes in the research log.