subtractive-debugging

What problem does it solve?

Subtractive debugging prevents regression hunts from turning into broad, noisy speculation by forcing you to identify the first cascaded mismatch between a known-good baseline and the current broken state.

Core Features & Use Cases

• Known-good anchoring: Requires defining one stable control plus exact property targets so the mechanism can be validated while shrinking the suspect space.
• Surface-specific checking: Narrows comparison to a single closure layer (loader, schema, import/display, runtime behavior, or artifact/render) to make differences actionable.
• Signature-based comparison: Compares entity signatures (IDs and required-field sections) instead of raw file churn to reduce noise from wide diffs.
• Interval subtraction then entity subtraction: Collapses the history interval first, then the changed-entity set, ensuring the introducing boundary is found with evidence.
• Causality-style mismatch naming: Produces a single first cascaded mismatch claim, then patches only that isolated boundary and reruns the same checker.

Quick Start

Ask the agent to isolate the introducing change for your regression by anchoring a known-good control, defining the exact property, running a surface-specific checker across progressively tighter baselines, and then patching only the first cascaded mismatch it identifies for your specified surface.