instinct-system

Track recurring codebase patterns by recording them as confidence-scored hypotheses in a dedicated instincts file, requiring active confirmation across multiple files before promoting them to permanent project conventions.

Confidence scoring for hypothesis tracking works by assigning a 0.3 to 0.9 range to observed patterns, controlling whether the system notes, suggests, follows, or promotes them based on certainty levels.

Safely confirm project conventions during codebase navigation by checking 2 to 3 related files to verify or disconfirm recurring style, architecture, naming, testing, or EF Core patterns before increasing confidence.

Yes, you can use instinct tracking for dotnet patterns and EF Core conventions by loading project-scoped instincts to learn per-repository conventions without incorrectly transferring rules across different codebases.

A confirmed instinct gets promoted to permanent memory rules when its confidence score reaches 0.9, offering evidence-backed upgrades to permanent rules or keeping instincts capped if the user declines.

The best way to manage project-scoped memory for repository-specific conventions is maintaining a dedicated instincts file with an observe-hypothesize-confirm lifecycle, ensuring patterns are validated before becoming acting conventions.