Analyze Token Waste

Use check to measure a payload before choosing a transformation.

Compare estimates with the default pipeline

Python

from axm_smelt import check, smelt

data = '{"name": "Alice", "notes": null}'
report = check(data)
for strategy, savings in report.strategy_estimates.items():
    print(f"{strategy}: {savings:.2f}% in isolation")
print(f"Safe pipeline: {report.savings_pct:.2f}%")
assert report.compacted == data
assert report.savings_pct == smelt(data).savings_pct

Each estimate applies one strategy to the original input. Positive reductions alone are included and rounded to two decimals. They overlap, so do not add them. A structural strategy can also save whitespace by serializing JSON. An absent estimate can mean inapplicable, unchanged, equal-token, or larger output.

report.savings_pct measures the chained safe preset; it does not estimate moderate or aggressive. To evaluate those, call smelt(data, preset=...) and inspect the returned text. Neither function mutates your source object.

From CLI or MCP

Bash

printf '{"name": "Alice", "notes": null}\n' | axm smelt_check --json-output
axm smelt_check --input-path ./payload.json

The tool returns format, tokens, and strategy_estimates. Cumulative savings is available in Python's report only. To obtain actual pipeline metrics through MCP/CLI, call smelt with the chosen preset. The text “no waste detected” means no registered strategy produced a positive isolated reduction; it does not mean the content is minimal or semantically safe.

Difference from smelt

Report behavior	check	smelt
compacted	Input unchanged	Accepted output text
compacted_tokens	Input count	Output count
strategy_estimates	Positive isolated estimates	Empty
savings_pct	Projected safe pipeline gain	Actual selected pipeline gain
strategies_applied	Empty	Accepted transforms

Continue with presets and report contracts.