write-patterns

name	write-patterns
description	Invoke for patterns.json format reference. Defines the schema for tracking unexplained data patterns.

patterns.json Format

Write to /workspace/run/patterns.json. This file tracks all discovered patterns — anomalies, structures, and associations found during EDA and experiments. Updated dynamically throughout exploration.

Schema

Root object with a single patterns array.

Pattern fields

Field	Type	Required	Description
id	string	yes	Pattern identifier (P1, P2, ...)
observation	string	yes	Specific, concrete description of the pattern. Must reference actual data values or statistics.
source	string	yes	Experiment ID that revealed this pattern (e.g. "base002", "exp003") or "eda"
status	string	yes	One of: unexplained, partially_explained, explained
investigation_id	string or null	yes	ID of the investigation addressing this pattern, or null if unattached
related_variables	array of strings	yes	Column names from the dataset involved in this pattern
priority	number	yes	0 to 1. Based on anomaly strength, effect size, or scientific interest. Higher = more worth investigating.

Status transitions

• unexplained → partially_explained: some aspects addressed by an investigation but uncertainties remain
• partially_explained → explained: investigation has resolved the pattern with sufficient evidence
• unexplained → explained: directly resolved in one cycle

When to update

• Create patterns: after EDA + base experiments (Phase 0), and after each experiment cycle when residuals or new anomalies emerge
• Update status: after judge evaluates an investigation — mark patterns as explained if the investigation's adjudication resolves them
• Add new patterns: from experiment residuals, unexpected findings, or anomalies in results