JSON Analyzer

What gets measured

The structural pass walks every node once and reports size in bytes, total node count, maximum nesting depth, the number of objects, arrays, strings, numbers, booleans and nulls, how many keys exist and how many of them are distinct.

Depth and distinct-key count are the two worth reading first. Depth tells you how painful this document will be to query and render — anything past ten is usually an API that grew rather than an API that was designed. A distinct-key count far below the total key count means you are looking at repeated records, which is the shape everything else here works best on.

Ten checks, on every document

The findings are heuristics with explicit thresholds rather than opinions. Here is every check, and exactly what triggers it.

Findings are sorted by severity and then by how many values they affect, so the check that touches the most data appears first within its band. Nothing is hidden — an empty findings list means all ten checks passed.

A worked example

Three user records, each slightly wrong in a different and entirely realistic way:

{ "users": [
  { "id": 1, "name": "A", "signed_up": "2024-01-02", "score": 10,   "note": null, "tags": [] },
  { "id": 2, "name": "B", "signed_up": "03/04/2024", "score": "20", "note": null },
  { "id": 2, "name": "C", "signed_up": "2024-01-05", "score": 30,   "note": null, "nickname": "" }
] }

warning  Consistency   $.users[].score has inconsistent types
                       66.7% number, 33.3% string
warning  Consistency   $.users[].signed_up uses multiple date formats
                       ISO 8601 date (2), MM/DD/YYYY (1)
warning  Consistency   $.users[].id contains 1 duplicate value
                       First repeated value: "2"
info     Data quality  3 null values detected — 13.6% of all values
info     Completeness  $.users[].nickname is missing from 2 of 3 objects
info     Completeness  $.users[].tags is missing from 2 of 3 objects
info     Data quality  1 empty string detected
info     Data quality  1 empty array detected

Every one of those is a real problem in waiting. The string "20" sorts and sums wrong. Two date formats mean whatever parses them is guessing. The repeated id breaks any code that builds a lookup keyed on it — including the diff tool, which needs unique identifiers to match array items properly.

Field profiles

Alongside the findings, every field is profiled: how many times it was seen, the breakdown of types it held, how many of its values were null or empty, its dominant type, and up to three real sample values. Fields are listed most-frequent first.

This is the fastest way to answer "what is actually in this response?" for an API whose documentation you do not trust. The sample values matter more than they sound — seeing three real values of a field usually tells you in a second what a type name never could.

From a finding to a fix

Behaviour on very large documents

Every traversal is iterative with an explicit stack, so depth cannot overflow the call stack no matter how nested the input is. Per-field data is stored as counters rather than value lists, which keeps memory bounded as row counts grow.

• Up to 5,000 distinct field paths are profiled; past that the shape is established and new paths are not tracked.
• Up to three sample values are kept per field.
• Duplicate-identifier scanning covers up to 200 identifier fields.
• Analysis runs in a Web Worker, so the page stays responsive while a large document is processed.

The analyzer never modifies your document — it only reads. And like everything else here, it runs in your browser: no upload, no stored copy, nothing transmitted.