poi-plugin-item-improvement2-beta 1.0.20 → 1.0.22

package/REFACTOR_NOTES.md

@@ -1,52 +1,35 @@
-# Selector / assistant-name refactor
+# Data projection refactor
 
-## Responsibility boundary
-
-- Spider parses Wiki-specific ship expressions and emits ship IDs.
-- Poi joins those IDs with the local `$ships` master data.
-- UI/detail/TXT consumers use the same normalized assistant view model.
-- `shipWeek.text` is retained only as source/fallback text.
-
-## Normalized fields
-
-Each improvement now has `assistantList`. Each assistant contains:
+## Runtime files
 
-- `sourceText`: original Spider/Wiki text.
-- `shipIds`: normalized positive IDs.
-- `ships`: local master-data records (`canonicalName`, `displayName`).
-- `displayText`: the single presentation value consumed by UI/export code.
-- `week`, `days`, `fullWeek`: normalized schedule fields.
-- `identity`: stable deduplication key.
-- `idsComplete`: whether IDs are authoritative for presentation.
+The plugin now consumes exactly two backend projections:
 
-Each item has `assistantTextByDay`, including key `-1` for all days.
+- `assets/db/improvement-list.json`: compact all + seven-day list data.
+- `assets/db/improvement-detail.nedb`: full equipment detail records.
 
-## Legacy compatibility
+The old `arsenal_all.nedb`, `arsenal_weekday.nedb`, and `items.nedb` runtime split has
+been removed.
 
-The current Spider data often stores only one anchor ID for irregular text such as
-`夕張改二/特/丁`. To avoid losing information, a single ID replaces source text only when:
-
-- Spider explicitly marks it with `idsComplete: true`, `resolved: true`, or
-  `parseStatus: "resolved"`; or
-- the source text exactly matches the local canonical ship name; or
-- multiple IDs are present; or
-- no source text exists.
+## Responsibility boundary
 
-The legacy `{ id: [0], text: "" }` record remains a valid schedule rule meaning no
-specific assistant display name. It still participates in weekday availability.
+- Spider parses Wiki rules and owns list aggregation.
+- The list page reads backend-provided `[itemId, assistantTexts]` rows directly.
+- The detail page reads the detail NeDB records.
+- Spider already emits `shipIdList` and `assistantShipIdsByDay`, but this plugin version
+  deliberately does not consume them yet.
+- Detail presentation continues to use the legacy anchor `id` and source `text`, so an
+  irregular description such as `夕張改二/特/丁` is not reduced to one anchor ship.
 
-## Removed legacy path
+## Data source
 
-The unused remodel-family selectors were removed from `selectors.es`. They compressed
-ship families and were not suitable for exact normalized ID presentation.
+`views/improvement-data-source.es` loads, validates, downloads, and replaces the two
+files as one dataset. `selectors.es` only joins the detail records with Poi master data
+and attaches the backend list summary.
 
 ## Verification
 
-Run:
-
 ```bash
+npm ci --include=dev
 npm test
+npm pack --dry-run
 ```
-
-This transpiles the plugin, runs the assistant normalization tests, and removes generated
-JavaScript files afterward.