sigrank-standard (Python)

Portable five-metric core for AI operator token-processing efficiency. Conforms to the TTEOP v0.1-draft specification.

Note: The PyPI distribution name remains sigrank-standard for backward compatibility with existing installs. The canonical protocol name is TTEOP (Token Telemetry Evaluation Operator Protocol). See teps/TEP-0001.md.

Install

pip install sigrank-standard

Usage

Compute metrics

from sigrank_standard import compute_metrics, build_record

# Four token pillars: input, output, cache_write, cache_read
result = compute_metrics(
    input_tokens=1_251_211,
    output_tokens=11_296_121,
    cache_write=128_196_310,
    cache_read=2_555_179_769,
)
print(result["metrics"])
# {'yield': 18436.98, 'leverage': 2042.2, 'velocity': 9.028, 'output_fraction': 0.9003, 'log_leverage': 3.31}

Build a full record

from sigrank_standard import build_record
import json

record = build_record(
    input_tokens=1_251_211,
    output_tokens=11_296_121,
    cache_write=128_196_310,
    cache_read=2_555_179_769,
    provider="anthropic",
    model="claude-sonnet-4",
    tool="claude-code",
)
print(json.dumps(record, indent=2))

Run conformance suite

sigrank-conformance
# or
python -m sigrank_standard

Metrics

Legacy aliases (deprecated): SNR → output_fraction, 10xDEV → log_leverage.

Null semantics

• When a denominator is zero, the metric is None (not 0 or Infinity)
• When cache_read is None, Yield, Leverage, and log_leverage are None
• When cache_write is None, log_leverage is None

Spec

TTEOP v0.1-draft (Token Telemetry Evaluation Operator Protocol). See signalaf.com/standard.

License

Apache-2.0