Pricing and Underwriting

# Generated from ch13_underwriting_agent/02_pricing_reconciliation_tool.py for in-browser execution.
# Agno agent wiring is removed so the tool runs as a plain function;
# the full version is on GitHub. Do not edit: regenerate with
# scripts/build_demos.py.
# Pricing reconciliation tool — GLM technical premium vs comparable median.
# Book reference: Chapter 13, "The Three-Agent Workflow"


def compare_to_pricing_model(
    emblem_radar_premium_usd: float,
    comparable_median_premium_usd: float,
    target_loss_ratio: float,
) -> dict:
    """Reconcile the GLM technical premium against the comparable-account median.

    A reconciliation gap above the firm's 10% threshold flags the submission
    for senior underwriter review before the recommendation enters Sophie
    Laurent's queue. Below threshold, the agent produces a draft with both
    anchors visible.

    Args:
        emblem_radar_premium_usd: technical premium from the GLM (USD).
        comparable_median_premium_usd: median of three comparable accounts (USD).
        target_loss_ratio: portfolio target for the renewal cycle.

    Returns:
        Dict with status, reconciliation_gap_pct, recommended_premium_usd,
        provenance, and note fields.
    """
    # Compute the percentage gap between the two technical-premium anchors
    gap_pct = abs(emblem_radar_premium_usd - comparable_median_premium_usd) \
              / comparable_median_premium_usd * 100

    # Threshold encodes the firm's reconciliation tolerance
    if gap_pct > 10.0:
        return {
            "status": "out_of_range",
            "reconciliation_gap_pct": round(gap_pct, 2),
            "recommended_premium_usd": None,
            "provenance": {
                "glm_method": "emblem_radar_v4.2",
                "comparable_method": "marketview_p_2025q3",
            },
            "note": "gap above 10% threshold; route to senior underwriter",
        }

    # Within tolerance — return midpoint as the draft anchor
    midpoint_premium = (emblem_radar_premium_usd + comparable_median_premium_usd) / 2

    return {
        "status": "ok",
        "reconciliation_gap_pct": round(gap_pct, 2),
        "recommended_premium_usd": round(midpoint_premium, 2),
        "provenance": {
            "glm_method": "emblem_radar_v4.2",
            "comparable_method": "marketview_p_2025q3",
            "target_loss_ratio": target_loss_ratio,
        },
        "note": None,
    }


if __name__ == "__main__":
    print(compare_to_pricing_model(86_000.0, 82_500.0, 0.62))   # within tolerance
    print(compare_to_pricing_model(86_000.0, 71_000.0, 0.62))   # out_of_range path

# Three-agent underwriting workflow — fixed path.
# Book reference: Chapter 13, "The Three-Agent Workflow"
# Repo note: tools referenced "from the firm's library" live in
# support.py and the sibling listing modules.
import importlib

from agno.agent import Agent
import os
import sys
sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), "..")))  # for common/
from common.config import get_model
from agno.workflow import Workflow, Step

from support import (
    fetch_emblem_radar_premium,
    parse_loss_summary,
    query_internal_loss_db,
    query_marketview_aggregator,
)

# Numeric module names can't be imported with a plain import statement.
extract_cope_attributes = importlib.import_module(
    "01_cope_extraction_tool").extract_cope_attributes
compare_to_pricing_model = importlib.import_module(
    "02_pricing_reconciliation_tool").compare_to_pricing_model

# Three specialised agents — each with a narrow, named scope
submission_agent = Agent(
    model=get_model(),
    tools=[extract_cope_attributes, parse_loss_summary],
    tool_call_limit=10,  # bound the agent loop; case-study default
    instructions="Extract COPE attributes and prior loss history.",
    markdown=True,
)

market_data_agent = Agent(
    model=get_model(),
    tools=[query_internal_loss_db, query_marketview_aggregator],
    tool_call_limit=10,
    instructions="Retrieve comparable internal claims and market benchmarks.",
    markdown=True,
)

pricing_comparison_agent = Agent(
    model=get_model(),
    tools=[fetch_emblem_radar_premium, compare_to_pricing_model],
    tool_call_limit=10,
    instructions="Compare GLM technical premium against comparable-account median.",
    markdown=True,
)

# Fixed-path workflow — the sequence is decided in advance, not at runtime
underwriting_workflow = Workflow(
    name="commercial_property_underwriting",
    steps=[
        Step(name="extraction", agent=submission_agent),
        Step(name="market_data", agent=market_data_agent),
        Step(name="pricing_comparison", agent=pricing_comparison_agent),
    ],
)

if __name__ == "__main__":
    # Run the workflow on a single submission. The path is resolved from
    # this file, not the working directory, so the script runs the same
    # from the repo root as from the chapter folder.
    submission_pdf = os.path.join(
        os.path.dirname(os.path.abspath(__file__)),
        "..", "data", "submissions", "MR-CHI-2025-Q3-018.pdf",
    )
    underwriting_workflow.print_response(
        f"Process submission {submission_pdf} "
        "(reference MR-CHI-2025-Q3-018) and produce a draft recommendation.",
        markdown=True,
        stream=True,
    )

Runs the unmodified chapter script on the server and streams the agent's tool calls and reasoning here.