Open Research Memory

Open Research Memory is a shared repository for reusable AI R&D experience records.

The goal is to accumulate evidence-backed memory that other contributors can apply safely across projects.

This repository is intended to be used as a read-only shared retrieval source during active runs and as a reviewed export target after task completion.

Scope

This repository accepts:

• verified episodes
• active procedures
• active insights

This repository does not accept:

• run-local working state
• unverified noise logs
• secrets or organization-specific sensitive data

Repository Structure

episodes/
  environment/
  training/
  evaluation/
  reproduction/
procedures/
  setup/
  debug/
  reproduce/
  experiment/
insights/
  planning/
  debugging/
  evaluation/
templates/
schemas/
scripts/

Record Contract

Each record must be a Markdown file with YAML frontmatter.

Required frontmatter fields:

• id
• type (episode|procedure|insight)
• status
• title
• tags
• created_at
• updated_at
• confidence
• human_verified
• source_run_id
• schema_version

Recommended statuses:

• draft
• reviewed
• verified
• trusted
• deprecated
• conflicted
• active

Minimum Content Requirements

Every shared record must contain these sections:

1. Context
2. Reproduce
3. Evidence
4. Failure Boundary

Use templates from templates/.

Contribution Workflow

1. Export candidate records from your local workspace.
2. Place records in the correct folder (episodes/, procedures/, insights/).
3. Run validation:
   • python3 scripts/validate_records.py
4. Open a pull request.
5. Merge only after checks and review pass.

Do not push directly to main.

Validation

Validation script:

• checks frontmatter required fields
• checks status/type consistency
• checks minimum required sections
• checks confidence and schema version format

Run locally:

python3 scripts/validate_records.py

Review Standard

A record is mergeable only when it is:

• reproducible by another contributor
• evidence-backed (logs, metrics, commits, artifacts)
• explicit about applicability and failure boundary
• free of sensitive data

Schema

See schemas/memory-record.schema.json for the machine-readable schema.