Docs: Update documentation for model loader changes

- Updated MODELLOADER.md to reflect hash-based change detection
- Clarified that formatting/whitespace changes do not trigger reloads
- Updated CHANGELOG.md with changes from #29

Author: sveneisenschmidt

CHANGELOG.md

@@ -7,6 +7,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Fixed
+- Model loader callback consistency - N8nApiModelLoader and JsonFileModelLoader now both only fire callbacks when models actually change (Closes #29)
+- JsonFileModelLoader no longer triggers reloads on formatting/whitespace changes
+
+### Changed
+- Added `getModelsHash()` method to base ModelLoader class for consistent hash-based change detection
+- Both loaders now use unified `lastHash` property name
+
 ## [0.0.9] - 2025-10-22
 
 ### Added

docs/MODELLOADER.md

@@ -28,10 +28,10 @@ MODELS_POLL_INTERVAL=1              # Polling interval in seconds (default: 1)
 
 **Behavior:**
 - Startup: Reads file synchronously, throws if not found or invalid JSON
-- Hot-reload: Polls file content (hash-based), reloads on changes
+- Hot-reload: Polls file and compares model hash, reloads only when models change
 - Polling interval: Configurable via `MODELS_POLL_INTERVAL` (default: 1s)
 - Invalid models: Filtered out with warnings, server continues
-- Hash comparison: Only reloads when file content actually changed
+- Hash comparison: Only reloads when model content actually changed (not formatting/whitespace)
 
 **Validation:**
 - Model ID: Non-empty string
@@ -82,6 +82,7 @@ AUTO_DISCOVERY_POLL_INTERVAL=300
 
 **Polling:**
 - Runs at startup, then at configured interval
+- Hash comparison: Only fires callbacks when models actually change
 - On failure: Logs error, keeps existing models, retries later
 - Disabled when `AUTO_DISCOVERY_POLL_INTERVAL=0`