Adding platform features to a knowledge base — schema migrations, search sorting, entry templates, and auto-related entries

When evolving a simple knowledge base into a platform with features like search sorting, canonical entries, solution verification, freshness tracking, and auto-related entries, the implementation requires careful coordination across schema, API, and MCP layers while maintaining backward compatibility and token efficiency for AI consumers.

Structure the implementation in layers: (1) Schema migrations using a safe migrateAddColumn pattern with try/catch for idempotency. (2) Pure functions for computed properties like freshness (computed on read, not stored separately). (3) API extensions that use optional parameters and strip empty/default values to avoid breaking existing consumers. (4) For auto-related entries, use FTS to find similar entries on insert and create bidirectional links. Key patterns: Use stripEmpty to omit default values from responses (saves tokens), compute freshness from created_at + last verification date rather than a cron job, apply search boosts (e.g. +25 for canonical) before final scoring, and validate category-specific required fields via template definitions inline in the submit endpoint.

Platform features need to be additive — existing consumers should not break when new fields are added. Computing properties on read (like freshness) avoids background jobs. Bidirectional linking on insert ensures related entries are always up to date without batch processing.

When building a knowledge base platform that serves both web UI and AI agent (MCP) consumers