vignette/apps/api/app/test_usage_report.py
2026-08-29 23:59:17 +09:00

158 lines
6 KiB
Python

from __future__ import annotations
import unittest
from .services.usage_report import REPORT_SCHEMA, build_model_cost_report
class UsageReportTest(unittest.TestCase):
def test_model_cost_report_derives_cost_efficiency_and_warnings(self) -> None:
report = build_model_cost_report(
{
"source": "database",
"durable": True,
"window_days": 7,
"total_turns": 5,
"metered_turns": 4,
"token_metered_turns": 3,
"token_unmetered_turns": 1,
"tokens_in": 300,
"tokens_out": 100,
"cost_usd": 0.4,
"budget": {
"status": "warn",
"limit_usd": 0.5,
"used_ratio": 0.8,
"remaining_usd": 0.1,
},
"evaluator_cache": {
"enabled": True,
"requests": 10,
"hits": 2,
"misses": 8,
"hit_rate": 0.2,
},
"by_provider": [
{
"provider": "cheap",
"model": "mini",
"turns": 3,
"token_metered_turns": 2,
"token_unmetered_turns": 1,
"tokens_in": 100,
"tokens_out": 50,
"cost_usd": 0.04,
},
{
"provider": "claude_cli",
"model": "opus",
"turns": 1,
"token_metered_turns": 1,
"token_unmetered_turns": 0,
"tokens_in": 200,
"tokens_out": 50,
"cost_usd": 0.36,
},
],
}
)
self.assertEqual(report["schema"], REPORT_SCHEMA)
self.assertEqual(report["summary"]["metered_coverage"], 0.8)
self.assertEqual(report["summary"]["token_metered_coverage"], 0.75)
self.assertEqual(report["summary"]["cost_per_1k_tokens_usd"], 1.0)
self.assertEqual(report["models"][0]["provider"], "claude_cli")
self.assertEqual(report["models"][0]["cost_share"], 0.9)
self.assertEqual(report["models"][0]["cost_per_1k_tokens_usd"], 1.44)
self.assertEqual(report["models"][1]["cost_per_turn_usd"], 0.013333)
self.assertIn("partial_metering", report["warnings"])
self.assertIn("partial_token_metering", report["warnings"])
self.assertIn("budget_warn", report["warnings"])
self.assertIn("low_evaluator_cache_hit_rate", report["warnings"])
self.assertIn("dominant_model_cost", report["warnings"])
def test_model_cost_report_handles_empty_usage(self) -> None:
report = build_model_cost_report({"budget": {}, "evaluator_cache": {}})
self.assertEqual(report["models"], [])
self.assertIsNone(report["top_cost_model"])
self.assertEqual(report["summary"]["metered_coverage"], 0.0)
def test_model_cost_report_warns_when_only_part_of_a_model_is_priced(self) -> None:
report = build_model_cost_report(
{
"cost_usd": 4.5,
"cost_basis": "partial",
"by_provider": [
{
"provider": "agy_cli",
"model": "gemini-3.7-flash-high",
"turns": 2,
"tokens_in": 2_000_000,
"tokens_out": 2_000_000,
"cost_usd": 4.5,
"cost_basis": "partial",
}
],
}
)
self.assertIn("partial_model_cost", report["warnings"])
self.assertIn("unavailable_model_cost", report["warnings"])
self.assertIsNone(report["summary"]["cost_per_turn_usd"])
self.assertIsNone(report["models"][0]["cost_share"])
def test_model_cost_report_warns_for_reference_upper_bound(self) -> None:
report = build_model_cost_report(
{
"cost_usd": 0.1,
"by_provider": [
{
"provider": "agy_cli",
"model": "gemini-3.7-flash-high",
"turns": 1,
"tokens_in": 100_000,
"tokens_out": 1_000,
"cost_usd": 0.1,
"cost_basis": "reference_upper_bound",
}
],
}
)
self.assertIn("reference_upper_bound_cost", report["warnings"])
def test_model_cost_report_preserves_partial_upper_bound_uncertainty(self) -> None:
report = build_model_cost_report(
{
"cost_usd": 4.5,
"cost_basis": "partial_upper_bound",
"budget": {
"status": "indeterminate",
"remaining_usd": None,
"cost_basis": "partial_upper_bound",
},
"by_provider": [
{
"provider": "agy_cli",
"model": "gemini-3.7-flash-high",
"turns": 2,
"tokens_in": 2_000_000,
"tokens_out": 2_000_000,
"cost_usd": 4.5,
"cost_basis": "partial_upper_bound",
}
],
}
)
self.assertEqual(report["summary"]["cost_basis"], "partial_upper_bound")
self.assertIsNone(report["summary"]["cost_per_turn_usd"])
self.assertIsNone(report["budget"]["remaining_usd"])
self.assertIn("partial_upper_bound_model_cost", report["warnings"])
self.assertIn("unavailable_model_cost", report["warnings"])
self.assertIn("budget_indeterminate", report["warnings"])
if __name__ == "__main__":
unittest.main()