| name | nematode-logbook |
| description | Create or update an experiment logbook with artifacts, supporting data, and roadmap updates. Use when the user wants to document evaluation results permanently. |
| metadata | {"author":"nematode","version":"1.0"} |
Create or update an experiment logbook with artifacts and documentation.
Input: Specify whether creating a new logbook or updating an existing one. Optionally provide a logbook number, title, or angle/hypothesis.
Steps
-
Determine logbook number and scope
If creating new:
- Check existing logbooks:
ls docs/experiments/logbooks/*.md
- Assign the next number (e.g., if 009 exists, use 010)
- Discuss angle/hypothesis with user if not provided
If updating existing:
- Identify which logbook to update
- Determine what new data to add
-
Collect experiment artifacts
For each experiment config to include:
a. Find all sessions matching the config in exports/:
b. Create artifact directory: artifacts/logbooks/{NNN}/{brain}_{environment}/
c. Copy per session:
- Experiment JSON from
experiments/{SESSION_ID}/{SESSION_ID}.json
- Config YAML from
experiments/{SESSION_ID}/*.yml
d. Copy best seed weights:
- Find the session with highest success rate
- Copy
exports/{SESSION_ID}/weights/final.pt to artifacts/logbooks/{NNN}/{dir}/weights/final.pt
Each artifact directory should contain: N session JSONs + 1 config YAML + 1 weights/final.pt
-
Create/update the main logbook
File: docs/experiments/logbooks/{NNN}-{title}.md
Follow the template structure from existing logbooks (see docs/experiments/templates/experiment.md):
- Objective: What question is being answered
- Background: Context and prior work
- Hypothesis: Testable predictions with expected ranges
- Method: Architecture, environments, key configuration, code changes
- Results: Summary tables (L100 as primary metric for consistency with logbooks 007-009), overall success rates, key findings
- Analysis: Hypothesis outcomes, root cause explanations, comparisons
- Conclusions: Numbered key takeaways
- Next Steps: Actionable items
- Data References: Artifact locations, config files, link to supporting doc
-
Create/update the supporting appendix
File: docs/experiments/logbooks/supporting/{NNN}/{title}-details.md
Include:
- Per-seed results tables (overall and L100)
- Learning curve analysis
- Ablation comparisons
- Hyperparameter tables
- Any data too detailed for the main logbook
-
Update the experiment index
Add entry to docs/experiments/README.md in the Active Experiments table.
-
Update the roadmap (if findings affect project direction)
Check docs/roadmap.md for:
- Phase status updates
- Exit criteria that can be checked off
- Quantum checkpoint assessments
- Go/no-go decisions
-
Verify
- All artifact directories have sessions + config + weights
- All markdown links resolve
- Numbers match between logbook summary and appendix detail
- L100 metrics are included for consistency with prior logbooks
Guardrails
- Always include L100 (last 100 episodes) as the primary convergence metric
- Always copy best-seed weights alongside session data
- Artifact configs are historical records — don't modify them after copying
- Keep the main logbook concise — detailed per-seed data goes in the appendix
- Reference existing logbooks (007, 008) for style consistency