SUPPLIED CONTENT

Read it however it needs to be read: parse the file (a quoted comma inside a description is common), open the PDF, fetch the page, look at the images. Then map, don’t author:

• Every entry in the source becomes an entry in the plan. Names, titles, descriptions, prices, dates verbatim; never rename, reprice, reword, reorder or add.
• Column and field names vary; map by meaning to the fields the vertical’s SEED.md plan shape defines. Anything the plan has no field for is dropped — say which columns you dropped.
• Their images only: an image URL column → the plan’s image URL field, verified with curl -sI → 200; a local file → the plan’s image path field. An entry with no image is seeded without one and listed in your summary. Never an imagePrompt beside supplied content — these are their things, not a mood board.
• Currency, time zone, address: only when the source or the brief states them. A price written with its unit (“9 dollars”, “$9”, “€20”) states the currency: it goes in the plan’s currency. When the source states none, the site’s own currency applies: name it in the closing message (“prices are in ILS, the site’s default; say the word and I switch it”) rather than quoting prices in a currency you assumed.
• Show the user the count you read next to the count you seeded, and the mapping you applied.

The vertical’s SEED.md says what an entry is for that vertical and which of its plan fields the usual columns land in.