Turn platform snapshots into tables: products, ratings and source checks

01Start with a source ledger for all four inputs

Domains, products, reviews and posts are fictional. source-manifest.json gives the tool, Source URL and expected count for every file. Use pasted source or file import rather than trying to collect these example domains online.

1. 1
   Process store-a-products.json and store-b-products.json separately in platform product collection, with https://shop-a.example/catalog and https://shop-b.example/catalog respectively.
2. 2
   Use review collection for reviews.json with https://reviews.example/tea, and social collection for social.json with https://posts.example/team.
3. 3
   Choose pasted source or import, load the complete file and select Parse content. Source text is sent to the KitNelo parsing service; this is not offline browser processing.
4. 4
   For each run retain JSON plus CSV or Excel, linking input and output filenames. Do not combine counts from these different tasks into one business unit.

02Choose the operation for this step

• Collect product snapshots: Preserve SKU, variant, currency and collection time; a displayed price is not necessarily the final checkout price or a live feed.
• Organize review records: Keep the original rating and its minimum and maximum scale, plus source and date; do not compare 4/5 directly with 4/10.
• Extract article records: Verify author, publication time and body completeness against the source; static HTML may omit dynamically loaded material.
• Inspect video metadata: Treat extracted duration, cover and public URL as metadata, and do not assume the tool downloads video or verifies playback statistics.
• Organize social posts: Keep author, post time and source-provided engagement fields; missing counts should remain missing rather than becoming zero.

03Define each file’s answer before accepting the preview

Seeing a table is insufficient. Store A has two product rows and store B one. Three reviews have two supplied ratings and one missing rating. Of two posts, one explicitly has zero likes and one supplies no count. Entirely empty fields may disappear from preview but still need export review.

04Complete product inputs: the same SKU can describe separate observations

Process store A’s variants separately from store B. TEA-100 repeats across stores and is not a global identifier; neither SKU nor Demo tea alone justifies merging. Keep original currencies rather than ranking USD 12 and EUR 11 as the same monetary unit.

{
  "@context": "https://schema.org",
  "@type": "ProductGroup",
  "name": "Demo tea",
  "hasVariant": [
    {
      "@type": "Product",
      "name": "Demo tea 100g",
      "sku": "TEA-100",
      "offers": {
        "@type": "Offer",
        "price": "12.00",
        "priceCurrency": "USD",
        "url": "/products/tea-100"
      }
    },
    {
      "@type": "Product",
      "name": "Demo tea 200g",
      "sku": "TEA-200",
      "offers": {
        "@type": "Offer",
        "price": "20.00",
        "priceCurrency": "USD",
        "url": "/products/tea-200"
      }
    }
  ]
}

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Demo tea 100g",
  "sku": "TEA-100",
  "offers": {
    "@type": "Offer",
    "price": "11.00",
    "priceCurrency": "EUR",
    "url": "/products/tea-100"
  }
}

Schema.org’s ProductGroup reference describes groups and variants. This exercise reconciles supplied snapshots, not checkout prices, shipping or live inventory.

05Three reviews supply two ratings: retain their scales first

Process the complete reviews.json in review collection. Expect Review A/B/C, with rating 4, 6 and blank; worstRating 1, 0 and blank; bestRating 5, 10 and blank. C supplies no rating. Do not fill zero or include it among known-rating observations.

Value divided by maximum gives A 80% and B 60%. Position within the stated interval gives A (4−1)/(5−1)=75% and B (6−0)/(10−0)=60%. These metrics differ; raw 4 and 6 also cannot be averaged into a common star scale. Known-rating coverage is 2/3 of this input, not coverage of all platform reviews.

[
  {
    "@type": "Review",
    "name": "Review A",
    "reviewBody": "Fictional review A.",
    "reviewRating": {
      "ratingValue": 4,
      "worstRating": 1,
      "bestRating": 5
    }
  },
  {
    "@type": "Review",
    "name": "Review B",
    "reviewBody": "Fictional review B.",
    "reviewRating": {
      "ratingValue": 6,
      "worstRating": 0,
      "bestRating": 10
    }
  },
  {
    "@type": "Review",
    "name": "Review C",
    "reviewBody": "Fictional review C has no rating."
  }
]

Schema.org Rating distinguishes ratingValue, bestRating and worstRating. The two calculations are explicitly stated editorial comparison methods; the collector does not create a trustworthy cross-platform average for you.

06Keep the JSON envelope and separate parsing from observation time

Process social.json next: the first record explicitly has zero likes; the second omits likes. Expect 0 and blank. Preview may omit entirely empty columns such as comments, but CSV/Excel and JSON preserve those fields. Visible column count is not completeness evidence.

The JSON download contains sourceUrl, source, mode, collectedAt and result, with rows at result.items. CSV/Excel primarily contain rows. A product row’s url is often its product page, not the imported catalog URL. Keep the envelope or a separate filename/source ledger.

collectedAt records this parsing operation. October 1/2 in the ledger are fictional observation labels, not verified historical prices. Parsing today does not establish that history; importing an old file does not refresh live platform data. A source value of paste does not establish an online collection.

[
  {
    "@type": "SocialMediaPosting",
    "articleBody": "Fictional post A",
    "likes": 0
  },
  {
    "@type": "SocialMediaPosting",
    "articleBody": "Fictional post B"
  }
]

Use the source-boundary guide for evidence scope and missing-versus-zero checks for absent fields rather than inventing counts.

07Reconcile each file and retain unresolved provenance

Keep four original JSON inputs, their ledger, four parsed JSON envelopes and the required row tables. Correct a wrong Source URL or input by parsing a new version rather than editing only a display label in the output.