FE-012C Repeatability Run 001

Purpose

This folder defines a repeatability run for FE-012C.

It reuses the same packet instrument from the original FE-012C manual replication run and is intended to test whether fresh model conversations produce similar results across time.

This is not a new theory experiment.

This is not a redesign exercise.

This is a repeatability test.

Run Requirements

• Use fresh GPT-5, Claude, and Gemini conversations.
• Do not show prior model outputs to any model.
• Do not compare outputs until all responses are locked.
• Do not modify the packet text during execution.

Directory Structure

• pre-registration.md
• repeatability-protocol.md
• run-instructions.md
• packet-index.md
• comparison-plan.md
• interpretation-rules.md
• packets/
• responses/gpt/
• responses/claude/
• responses/gemini/
• comparison/
• scripts/
• package.json

Packet Instrument Status

The packet contents in packets/ were copied from:

research/experiments/FE-012C-manual-replication/packets/

No packet content changes were introduced during copy.

If any packet text is corrected later to address a broken placeholder or transcription defect, document the change in this file and in repeatability-protocol.md before the run is executed.

Dataset Comparison

Run dataset comparison from this experiment directory with:

npm run compare

This generates reproducible comparison outputs in comparison/generated/.

The comparison script does not modify raw response JSON files.

Current repeatability responses include noncanonical filenames and at least some JSON files that require quote normalization for parsing. The comparison script uses packet_id from the payload when available and records parse normalization in the generated raw comparison output.

Run semantic comparison beside the literal comparator with:

npm run semantic-compare

The literal comparator remains the syntactic baseline.

The semantic comparator adds provisional normalization rules without hiding literal differences.