/explain¶
Get a plain-language explanation of recent experiment results and their implications.
Usage¶
What It Does¶
- Reviews recent history -- Reads the latest entries in
experiment_log.tsvand any recent research or debug artifacts. - Synthesizes context -- Connects results across multiple experiments to identify patterns, trends, and key takeaways.
- Generates explanation -- Produces a clear, non-technical summary of what happened, why it matters, and what to try next.
- Updates insights -- Appends the most important findings to
.researchpad/experiments/insights.md.
Example¶
Output might include:
Over the last 5 experiments, reducing the learning rate from 1e-2 to 3e-4 steadily improved validation accuracy from 0.82 to 0.89. The biggest jump came from adding gradient clipping (exp-043), which eliminated the NaN issues seen in exp-042. The model appears to be approaching a plateau -- consider trying a different optimizer or augmenting the training data.
Artifact Format¶
The /explain command appends to:
.researchpad/experiments/insights.md-- New insights are added as sections at the end of the file
Unlike /research and /debug, this command does not create a new file. It updates the existing insights document, which the Dashboard displays in the Insights panel.
Cumulative by design
Each /explain invocation adds to the insights file rather than replacing it. Over time, this builds a comprehensive record of your project's learnings.