highest-cost-scenario-identification

name

Highest_Cost_Scenario_Identification

description

Identifies the highest-cost scenario in payment processing fee analysis for the dabstep dataset. Use this skill when asked to find the most expensive MCC (Merchant Category Code) for a given transaction amount, or the most expensive ACI (Authorization Characteristics Indicator) for a given card scheme and transaction amount. Triggers on questions like "what is the most expensive MCC", "which ACI leads to highest fees", "find the costliest transaction scenario", or any query asking to identify maximum-fee scenarios in payment processing.

Highest Cost Scenario Identification

Dataset Overview

The dabstep dataset models payment processing fees with these key files:

• fees.json: 1000 fee rules, each specifying conditions and fee parameters
• merchant_category_codes.csv: MCC descriptions (not exhaustive — fee rules may reference MCCs not in this file)
• manual.md: Domain definitions including ACI codes (A–G) and fee formula

Fee formula: fee = fixed_amount + rate * transaction_amount / 10000

Two Question Types

Type 1: Most Expensive MCC

"What is the most expensive MCC for a transaction of X euros, in general?"

Find which MCC(s) can incur the highest fee across all fee rules.

import json

with open('fees.json') as f:
    fees = json.load(f)

amount = 50  # transaction amount in euros

mcc_max_fee = {}
for rule in fees:
    mcc_list = rule['merchant_category_code']
    if not mcc_list:  # empty list means rule applies to all MCCs, skip for MCC ranking
        continue
    fee = rule['fixed_amount'] + rule['rate'] * amount / 10000
    for mcc in mcc_list:
        mcc_max_fee[mcc] = max(mcc_max_fee.get(mcc, 0), fee)

max_fee = max(mcc_max_fee.values())
result_mccs = sorted([mcc for mcc, f in mcc_max_fee.items() if f == max_fee])
# Output as comma-separated sorted list
print(', '.join(str(m) for m in result_mccs))

Critical: Do NOT filter MCCs against merchant_category_codes.csv. Some MCCs in fee rules (e.g., 3003, 7231) are absent from the description file but are valid and must be included in results.

Type 2: Most Expensive ACI

"For a credit transaction of X euros on [CardScheme], what would be the most expensive ACI?"

Find which ACI letter leads to the highest possible fee for the specified card scheme and credit type. Only filter by the explicitly given conditions (card_scheme, is_credit); take the maximum fee across all other conditions.

import json

with open('fees.json') as f:
    fees = json.load(f)

card_scheme = 'GlobalCard'   # exact match required
is_credit = True             # set False for debit
amount = 1                   # transaction amount in euros
ALL_ACIS = ['A', 'B', 'C', 'D', 'E', 'F', 'G']

aci_max_fee = {}
for rule in fees:
    # Filter by card scheme
    if rule['card_scheme'] != card_scheme:
        continue
    # Filter by credit status: null means applies to both credit and debit
    if rule['is_credit'] is not None and rule['is_credit'] != is_credit:
        continue

    fee = rule['fixed_amount'] + rule['rate'] * amount / 10000
    # Empty aci list means rule applies to ALL ACI values
    acis = rule['aci'] if rule['aci'] else ALL_ACIS
    for aci in acis:
        aci_max_fee[aci] = max(aci_max_fee.get(aci, 0), fee)

max_fee = max(aci_max_fee.values())
best_acis = sorted([a for a, f  aci_max_fee.items()  f == max_fee])
result = [best_acis[]]   
(result)             

The final <answer> must look like: <answer>['C']</answer> — a Python list containing the single ACI letter.

Key Rules and Pitfalls

Null/Empty field semantics (from manual.md)

Field	Null/Empty means
is_credit: null	Applies to both credit and debit
aci: []	Applies to all ACI values (A–G)
merchant_category_code: []	Applies to all MCCs (skip for MCC ranking)
account_type: []	Applies to all account types
Other null fields	Apply to all values of that field

Common errors to avoid

1. Outputting fee amount instead of ACI letter — the answer is always a letter (C, B, etc.) wrapped in a list (['C']), never a fee value like 0.23 or 0.1494. The example <answer>0.23</answer> in task instructions is a generic format example only.
2. Outputting bare letter without list — always wrap in a list: ['C'] not just C, even if the question says "Answer must be just one letter". The question also says "Provide the response in a list even if there is only one value" — the list format takes precedence.
3. Filtering MCCs by merchant_category_codes.csv — wrong. Always use MCCs from fees.json directly.
4. Ignoring null is_credit rules for ACI questions — rules with is_credit: null apply to credit transactions and must be included.
5. Forgetting empty aci list — an empty aci list in a rule means it covers ALL ACIs; don't skip these rules.

"In general" interpretation

"Most expensive in general" means the worst-case maximum fee across all possible rule configurations. Only apply the filters explicitly given in the question (card_scheme, is_credit, transaction amount). All other rule fields (capture_delay, monthly_fraud_level, monthly_volume, etc.) are ignored — take the maximum fee achievable under any applicable rule.

Tie-breaking

When multiple ACIs or MCCs share the same maximum fee, return the one with the lowest alphabetical/numerical order. For ACI questions, sort the tied ACIs alphabetically and take the first: sorted(tied_list)[0].

Output Format

• MCC questions: Comma-separated sorted list of MCC integers, e.g., 3000, 3001, 3002
• ACI questions: Single letter wrapped in a Python list, e.g., ['C'] — the answer is the LETTER, not the fee amount

ACI answer examples:

• <answer>['C']</answer> ✓
• <answer>['B']</answer> ✓
• <answer>C</answer> ✗ (missing list)
• <answer>0.23</answer> ✗ (fee value, not letter)

ACI Reference (from manual.md)

ACI	Description
A	Card Present - Non-authenticated
B	Card Present - Authenticated
C	Tokenized card with mobile device
D	Card Not Present - Card On File
E	Card Not Present - Recurring Bill Payment
F	Card Not Present - 3-D Secure
G	Card Not Present - Non-3-D Secure

Card Schemes in Dataset

GlobalCard, NexPay, SwiftCharge, TransactPlus