Use JSON or JSONL

JSON formats preserve native types (numbers, booleans) — no quoting or delimiter issues.

JSON (.json)

Array of objects:

[
  { "Source Text": "Budi", "Opacity": 100 },
  { "Source Text": "Sari", "Opacity": 60 },
  { "Source Text": "Andi", "Opacity": 25 }
]

Object with rows key:

{
  "rows": [
    { "Source Text": "Budi", "Opacity": 100 },
    { "Source Text": "Sari", "Opacity": 60 }
  ]
}

Type preservation

JSON value	Becomes in EffectCraft
"Budi" (string)	text control → "Budi"
100 (number)	slider control → 100
true (boolean)	checkbox control → true
75.5 (float)	slider control → 75.5

No coercion needed — types flow directly.

JSONL / NDJSON (.jsonl, .ndjson)

One JSON object per line. Ideal for large datasets (streaming, no memory pressure).

{"Source Text": "Budi", "Opacity": 100}
{"Source Text": "Sari", "Opacity": 60}
{"Source Text": "Andi", "Opacity": 25}

• Empty lines ignored
• Each line must be valid JSON
• Line numbers in errors = JSONL line number

Validation

Same validation as CSV: every object key must match a control (or be mapped). Extra keys = error.

lakreet-swarm check --template template.ectemplate --data data.json
lakreet-swarm check --template template.ectemplate --data data.jsonl

When to use which

Format	Best for
CSV	Spreadsheets, Excel export, human editing
JSON	Programmatic generation, native types, small–medium data
JSONL	Large datasets, streaming, log pipelines, 10k+ rows

Mixed types in one file

All objects must have the same keys (or missing keys = empty). Missing keys for slider/checkbox = validation error unless --empty-as-default.

Next steps

• Prepare a CSV — if you prefer spreadsheets
• Map columns — remap keys to control names
• Data formats reference — full specification