Exercise 2: Deliberately Breaking the Shared-Core Guarantee — Possible Solution
====================================================================

THE BROKEN WIRING
------------------------------
    quote_engine = PricingEngine(RealInventoryAdapter({'PROD-1': 50}), LoggingNotificationAdapter())  # sees fresh stock
    order_engine = PricingEngine(RealInventoryAdapter({'PROD-1': 5}), LoggingNotificationAdapter())    # sees stale, low stock

    order_service = OrderService(repository, order_engine, event_bus)  # uses the STALE engine

    quoted_price = quote_price_via_api(quote_engine, 'PROD-1', 100, is_loyalty_member=False)
    actual_price = order_service.place_order('ORD-3', 'PROD-1', 100, 'USER-3', is_loyalty_member=False)

Two SEPARATE PricingEngine instances are constructed, each with its own
RealInventoryAdapter reading different stock data - simulating a real
scenario where the quote API's own view of inventory (50, fresh) has
drifted out of sync with the order-processing flow's own view (5,
stale), because they were never actually sharing one source of truth
in the first place.

VERIFYING THE DISCREPANCY REPRODUCES
------------------------------
    quoted price (via quote_engine, stock=50): 100
    actual price (via order_engine, stock=5): 115.0
    match: False
    discrepancy: 15.0

The customer was quoted $100 (no scarcity pricing, since quote_engine
saw plenty of stock) but was actually charged $115.0 (scarcity pricing
applied, since order_engine saw only 5 units) - a genuine, verified
$15.00 discrepancy for the identical product and the identical base
price.

WHY THIS IS THE SAME BUG AS CHAPTER 8'S, ONE LEVEL DEEPER
------------------------------
Chapter 8 verified two independently-implemented CLIENT-side pricing
functions disagreeing by $0.50. This exercise reproduces the identical
failure mode entirely on the SERVER side: two technically-identical
PricingEngine classes, running the exact same code, still disagree -
not because the logic differs, but because the DATA each one sees
differs, and nothing enforces that both engines stay in sync. This
confirms this chapter's own ADR-007 was addressing a real risk, not a
hypothetical one - "two separate implementations" and "two separate
instances with drifted data" are both instances of the same underlying
problem: no single source of truth.

WHY THIS WORKS AS AN ANSWER
------------------------------
The broken version is constructed by deliberately reversing this
chapter's own "one shared instance" decision - using two separate
PricingEngine objects instead of one - and the resulting discrepancy is
verified directly with a specific dollar amount, rather than only
asserted to exist.