Use Strategies

Pass registered names as a Python list, or as one JSON-encoded CLI argument. Read the catalog for fidelity and format limits.

Select a small pipeline

Bash

printf '{"a": 1, "b": null, "c": ""}\n' | axm smelt --strategies '["minify", "drop_nulls"]'

Python

from axm_smelt import smelt

report = smelt(
    '{"a": 1, "b": null, "c": ""}',
    strategies=["minify", "drop_nulls"],
)
assert report.compacted == '{"a":1}'
print(report.strategies_applied)

The list's order matters. The report lists only candidates accepted by the token/length guard, so selecting minify does not guarantee it appears.

JSON transformations

The examples below are independent inputs in one script:

Python

from axm_smelt import smelt

cases = [
    ('{"a": 1, "b": 2}', ["minify"]),
    ('{"a": {"b": 1}}', ["flatten"]),
    ('[{"name":"Alice","age":30},{"name":"Bob","age":25}]', ["tabular"]),
    ('{"name": "Alice"}', ["strip_quotes"]),
    ('{"x": 3.14159265}', ["round_numbers"]),
]
for data, strategies in cases:
    report = smelt(data, strategies=strategies)
    print(strategies, report.compacted, report.strategies_applied)

Flattening can collide with existing dotted keys; tabular output loses cell types; stripped keys are not standard JSON. Float precision defaults to two decimal places. The public smelt function accepts names, not strategy instances or per-strategy configuration.

Repeated strings

Give aliasing enough repetition to compensate for its envelope overhead:

Python

from axm_smelt import smelt

value = "A long repeated description of this record and its context. " * 3
payload = {f"item_{i}": value for i in range(8)}
report = smelt(parsed=payload, strategies=["dedup_values_with_refs"])
print(report.compacted)
assert "dedup_values_with_refs" in report.strategies_applied

The output contains _refs and _data, changing its schema. An existing top-level _refs or _data causes aliasing to be skipped. The alias contract explains literal collisions and reconstruction.

Prose and Markdown

Python

from axm_smelt import smelt

text = "# Report\n\n| Name | Value |\n| --- | --- |\n| A | 1 |\n\n<!-- draft note -->\n"
report = smelt(text, strategies=["compact_tables", "strip_html_comments"])
print(report.compacted)

compact_tables requires a Markdown classification. Comment removal also applies to plain text. Review the backtick-fence limitations before processing code examples.

Combine with a file input

Bash

axm smelt --input-path ./payload.json --strategies '["minify", "drop_nulls", "flatten", "tabular"]' --json-output

Once a transform emits a table or relaxed syntax, later JSON transforms may become ineffective. The pipeline preserves the input's format label rather than re-detecting each intermediate result.