Product Name Cleanup — Phase 3 Batch

Administrative exchange for unstructured Bucket B rows in pd_name_cleanup_staging (status = ai_ready AND bucket = bucket_b). Export writes those rows, with the live brand, model, and MPN, into JSON files a Grok session can read. Import reads the completed files, applies a 70-character title guard, and writes the proposal back onto the same product_id + language_id row. Export files are grok_batch_001.json and count upward, with at most 250 records each. Import reads only grok_batch_*_done.json. A file that is not valid JSON is logged and skipped. A file that commits is moved into cleanup_scratchpad/archive/. The import sets status to ai_completed. Titles of 70 characters or fewer stay unflagged. Longer titles are stored with is_flagged = 1 and flag_reason = AI_TITLE_EXCEEDS_70_CHARS. Only ai_ready and ai_completed rows can change. original_name and original_description are never assigned. This phase does not change oc_product or oc_product_description.

Operator Workflow

Work through these five steps in order. You do not need to edit code or open the database.

Technical reference

• A browser visit shows the counts and the scratchpad file list. Generate, upload, and import start from the controls below. POST action=export_batches, action=upload_done, and action=import_done run the same engines. CLI still uses --action=export or --action=import.
• CLI: php productNameCleanup_phase3_batch.php --action=export or --action=import from the fsadmin directory. With no --action, the CLI asks for one. JSON clients send format=json or the XMLHttpRequest header. Those callers skip navigation.
• Scratchpad: /home/opencart/public_html/fsadmin/cleanup_scratchpad/. The directory is created at mode 0755 when it is missing. Files written by this page, and files received through the upload form, belong to the web server user and are stored at mode 0644. Apache is denied direct access. Download links stream the files through this script.
• Export selects status = ai_ready and bucket = bucket_b, joins oc_product for model and MPN, and left-joins oc_manufacturer for the brand. A missing manufacturer does not drop the row.
• A second export is refused while unfinished grok_batch_###.json files are still in the scratchpad. Move them aside, or finish them and save the replies as grok_batch_###_done.json, before exporting again.
• Import matches product_id and language_id. Rows at passed_audit, committed, or any status other than ai_ready or ai_completed are not overwritten.
• Writes commit per file, or in chunks of 2,500 rows when a done file is larger than that. A failed chunk rolls back that chunk only. The done file stays in place so it can be imported again.
• A proposed name longer than 255 characters is clamped to the staging column. char_count_proposed keeps the real length, up to 65,535, and the row is flagged because it is also over 70 characters.