docs: update documentation for normalized config and improve command

- Add improve command to all command tables and usage sections
- Add normalized_config field to BestCarRecord struct documentation
- Update CarCounts.by_rid type from String to (String, String)
- Update CarCounts.guid_to_rid type to HashMap<String, HashSet<(String, String)>>
- Add owned_configs field to MyCarsData and CarCounts docs
- Add normalize_to_final_build() to API docs
- Update CSV format examples to include FinalBuild column
- Update my_cars.json structure with engineMajor/weightMajor/chassisMajor
- Update timing file example with v2c_ version prefix
- Add version prefix stripping to data transformation docs
- Update CHANGELOG.md with Unreleased section
- Fix compare_records description to include normalized config sorting

docs/ARCHITECTURE.md

@@ -14,13 +14,13 @@ The Top Drives Records Analyzer is a Rust CLI tool for analyzing car performance
 
 | Module | File | Key Exports |
 |--------|------|-------------|
-| `types` | `src/types.rs` | `AppError`, `BestCarRecord`, `Car`, `TripletKey` |
+| `types` | `src/types.rs` | `AppError`, `BestCarRecord`, `Car`, `TripletKey`, `normalize_to_final_build` |
 | `context` | `src/context.rs` | `AnalysisContext`, `CarCounts`, `compute_car_counts` |
 | `user_data` | `src/user_data.rs` | `MyCarsData`, `load_my_cars_data` |
 | `translate` | `src/translate.rs` | `translate_track_name`, `translate_track_type` |
 | `http` | `src/http.rs` | `get_file` |
 | `csv_util` | `src/csv_util.rs` | `compare_cars`, `escape_csv_field` |
-| `test_utils` | `src/test_utils.rs` | `make_record`, `make_car` (test helpers) |
+| `test_utils` | `src/test_utils.rs` | `make_record`, `make_car`, `empty_my_data` (test helpers) |
 
 #### Error Handling
 All public functions return `Result<T, AppError>` where `AppError` is a unified error type:
@@ -58,6 +58,7 @@ All public functions return `Result<T, AppError>` where `AppError` is a unified
 | `wishlist.rs` | `wishlist` | Generates `wishlist.csv`, `add.csv`, `remove.csv` |
 | `details.rs` | `details` | Generates `details.csv` with all timing data |
 | `track.rs` | `track` | Prints best car for a specific track/condition/RQ |
+| `improve.rs` | `improve` | Generates `improve.csv` (owned cars not yet at final build) |
 | `all.rs` | `all` | Generates all output files in a single analysis pass |
 
 #### Shared Helpers (`src/commands/mod.rs`)
@@ -86,14 +87,16 @@ All public functions return `Result<T, AppError>` where `AppError` is a unified
 
 **CSV output utilities** and sorting helpers.
 
-- `write_sorted_csv()` — Generic sorted CSV writer (index-based sorting to avoid cloning)
+- `atomic_write()` — Writes to a temporary file first, then renames on success (no partial files)
+- `write_sorted_csv()` — Generic sorted CSV writer (index-based sorting to avoid cloning, uses `atomic_write`)
 - `write_best_csv()`, `write_very_best_csv()` — Best car CSV output
 - `write_trash_csv()` — Trash car CSV output
 - `write_details_csv()` — Comprehensive timing data CSV output
 - `write_wishlist_files()` — Wishlist, add, and remove CSV output
+- `write_improve_csv()` — Improve car CSV output (owned cars not yet at final build)
 - `print_best_car()` — Stdout output for track-specific query
 - `compute_ideal_wishlist()` — Ideal wishlist computation
-- `compare_records()` — Sorting comparator for `BestCarRecord`
+- `compare_records()` — Sorting comparator for `BestCarRecord` (by RQ, normalized config, brand, model)
 
 #### Data Flow
 
@@ -146,16 +149,17 @@ struct Car {
 #### BestCarRecord
 ```rust
 struct BestCarRecord {
-    car_rid: String,   // Car identifier
-    brand: String,     // Manufacturer
-    only_name: String, // Model name
-    year: u16,         // Year
-    rq: u8,            // RQ level
-    config: String,    // Tune config (111, 233, etc.)
-    time: f64,         // Track time (seconds)
-    owned: bool,       // User ownership status
-    prize: bool,       // Prize car status
-    guid: String,      // Game identifier
+    car_rid: String,           // Car identifier
+    brand: String,             // Manufacturer
+    only_name: String,         // Model name
+    year: u16,                 // Year
+    rq: u8,                    // RQ level
+    config: String,            // Original tune config from JSON (111, 221, 233, etc.)
+    normalized_config: String, // Normalized final build (323, 233, or 332)
+    time: f64,                 // Track time (seconds)
+    owned: bool,               // User ownership status
+    prize: bool,               // Prize car status
+    guid: String,              // Game identifier
 }
 ```
 
@@ -175,15 +179,17 @@ struct AnalysisContext {
 ### CarCounts
 ```rust
 struct CarCounts {
-    by_rid: HashMap<String, (BestCarRecord, usize)>,
-    guid_to_rid: HashMap<String, String>,
+    by_rid: HashMap<(String, String), (BestCarRecord, usize)>,
+    guid_to_rid: HashMap<String, HashSet<(String, String)>>,
     car_map: HashMap<String, Car>,
     car_list: Vec<Car>,
+    owned_configs: HashMap<String, String>,
 }
 ```
 
 **Purpose**: Aggregated per-car win counts computed once from `AnalysisContext`
 **Usage**: Shared across all output functions to avoid redundant computation
+**Key grouping**: `by_rid` is keyed by `(car_rid, normalized_config)` so different tune configs of the same car are tracked separately.
 
 ## Algorithm Details
 
@@ -241,7 +247,8 @@ Multiple validation layers ensure data quality:
 ├── trash.csv             # Owned cars safe to remove
 ├── wishlist.csv          # Ideal wishlist (very_best unowned, capped)
 ├── add.csv               # Cars to add to wishlist
-└── remove.csv            # Cars to remove from wishlist
+├── remove.csv            # Cars to remove from wishlist
+└── improve.csv           # Owned cars not yet at final build config
 ```
 
 ## Performance Characteristics