{ "version": 1, "provenance": "Invented teaching corpus authored with both extractors on 2026-10-31. Split labels demonstrate a procedure; this is not a blind independent evaluation. Expected records were explicitly authored, not derived from extractor output.", "cases": [ { "id": "dev-single", "split": "development", "reason": "One direct product record", "html": "

Desk lamp

", "expected": [ { "sku": "D1", "title": "Desk lamp" } ] }, { "id": "dev-multiple", "split": "development", "reason": "Two distinct direct product records", "html": "

Table

Chair

", "expected": [ { "sku": "D2", "title": "Table" }, { "sku": "D3", "title": "Chair" } ] }, { "id": "nested-recommendation", "split": "heldout", "reason": "Recommendation inside a parent is not a catalog record", "html": "

Lamp

", "expected": [ { "sku": "H1", "title": "Lamp" } ] }, { "id": "entities-and-space", "split": "heldout", "reason": "Decode entities and collapse display whitespace", "html": "

Tea &\n coffee

", "expected": [ { "sku": "H2", "title": "Tea & coffee" } ] }, { "id": "empty-catalog", "split": "heldout", "reason": "Labeled genuine empty catalog, not a request failure", "html": "

No products available

", "expected": [] }, { "id": "outside-promotion", "split": "heldout", "reason": "Outside promotion must not join catalog records", "html": "

Book

", "expected": [ { "sku": "H3", "title": "Book" } ] }, { "id": "title-wrapper", "split": "heldout", "reason": "A field wrapper breaks a direct-child assumption", "html": "

Notebook

", "expected": [ { "sku": "H4", "title": "Notebook" } ] }, { "id": "late-renamed-record", "split": "heldout", "reason": "Sixth record uses a different root class; first five still look correct", "html": "

Item 1

Item 2

Item 3

Item 4

Item 5

Item 6

", "expected": [ { "sku": "L1", "title": "Item 1" }, { "sku": "L2", "title": "Item 2" }, { "sku": "L3", "title": "Item 3" }, { "sku": "L4", "title": "Item 4" }, { "sku": "L5", "title": "Item 5" }, { "sku": "L6", "title": "Item 6" } ] } ] }