forked from hesabix/arc
26 lines
959 B
Python
26 lines
959 B
Python
"""Phase 7 — schema loading eval smoke + production K unchanged."""
|
|
from __future__ import annotations
|
|
|
|
from app.services.ai.ai_constants import (
|
|
MAX_TOOLS_AUTONOMOUS,
|
|
MAX_TOOLS_PER_REQUEST,
|
|
PROGRESSIVE_SCHEMA_LOADING,
|
|
)
|
|
from app.services.ai.ai_tool_discovery_gold import load_gold_queries
|
|
from app.services.ai.ai_tool_schema_eval import simulate_mode
|
|
|
|
|
|
def test_production_discovery_k_unchanged():
|
|
assert MAX_TOOLS_PER_REQUEST == 48
|
|
assert MAX_TOOLS_AUTONOMOUS == 128
|
|
assert PROGRESSIVE_SCHEMA_LOADING is False
|
|
|
|
|
|
def test_progressive_simulation_uses_fewer_wire_tokens_than_k48():
|
|
gold = load_gold_queries()[:40]
|
|
current = simulate_mode(k=48, progressive=False, rows=gold)
|
|
progressive = simulate_mode(k=10, progressive=True, rows=gold)
|
|
assert progressive["avg_wire_schema_tokens_per_request"] < current[
|
|
"avg_wire_schema_tokens_per_request"
|
|
]
|
|
assert progressive["recall"] <= current["recall"] + 1e-9
|