Multi-agent audit of the money-moving economy modules surfaced three
confirmed correctness bugs. All three are fixed here with regression tests.
heist (core/economy/heist.py):
- `house = await get_user(house.HOUSE_ID)` shadowed the imported `house`
module for the whole function, so `house.HOUSE_ID` raised UnboundLocalError
on every successful heist (win payout was entirely dead) and on the
fail-path compensation branch. Rename the local to `house_rec`.
- Un-shadowing exposed a latent mint: the pot was floored at 300 but the
house was debited only min(total, balance), so a poor house paid out more
than it lost. Cap the pot at the balance and debit exactly what is paid
(house debit == sum of payouts). No mint, no leak.
- Add the missing `_is_jailed` import (do_heist_check referenced it unimported).
blackjack (commands/economy_games_commands.py):
- Button callbacks had no reentrancy guard; discord.py dispatches each click
as its own task, so double-clicking Stand within the dealer-reveal window
paid out twice (mint), and double-clicking Double/Split deducted the extra
bet twice. Add a synchronous `_busy` guard (matching the existing RpsGame
idiom) on all four callbacks plus a `_resolved` idempotency flag on
settlement, so a game can only pay out once.
bail (core/economy/jail.py, commands/economy_extra_commands.py):
- do_bail only checked balance, never jail state; a double-click or a stale
BailView from a re-run /jailbreak charged bail twice, destroying coins (bail
is a pure sink). Make do_bail a no-op when the user is not jailed, and add a
UI reentrancy guard + "already free" message.
Tests: 47 passed (4 new regression tests covering heist coin-conservation and
bail idempotency).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
