Transaction Fee Calculator

Phone-screen-style string-processing round. Given a CSV of transactions and a fee-rule table, compute the fee per row based on payment provider, payment type, and status. Requirements Input: a single CSV string. The first row is the header; remaining rows are transaction records. Canonical columns: id, reference, amount, currency, date, merchant_id, buyer_country, transaction_type, payment_provider, status. amount is in cents (e.g. 1000 = $10.00). status values include payment_completed, payment_failed, payment_pending. payment_provider ∈ {card, klarna, bank_transfer}. Output: a CSV string with header id,transaction_type,payment_provider,fee, one line per input transaction. The fee per row comes from a rules table keyed by (buyer_country, payment_provider, status); each entry is a percentage rate plus a fixed cent fee. Always floor the final fee to an integer (use int() truncation). The problem is delivered in three escalating parts: Part 1 — basic per-provider fee Each provider charges amount × rate + fixed, floored: provider rate fixed card 2.9% 30 klarna 3.5% 50 bank_transfer 0.8% 0 (flat) def calculate_fees(csv_data: str) -> str: ... # Parse header → {col_name: index}; skip header row when iterating. # fee = int(amount * rate + fixed) # floor via int() # Emit "id,transaction_type,payment_provider,fee" per row. Part 2 — status gate + regional rates Only charge a fee when status == "payment_completed"; for failed/pending the fee is 0. Ireland (buyer_country == "ie") uses cheaper rates; all other countries fall back to the Part-1 standard rates. ie provider rate fixed card 1.9% 20 klarna 2.5% 40 def get_fee_config(country: str, provider: str) -> tuple[float, int]: ... # Return regional override if (country, provider) present, else the default rate. # Caller: if status != "payment_completed": fee = 0 Part 3 — per-merchant volume discount Count successful (payment_completed) transactions per merchant_id. Look up the discount using the merchant's count before the current transaction (i.e. on the (count+1)-th completed tx), then increment the count only for completed transactions. Process rows in input order — ordering is significant. Fees are now sourced from a passed-in country_fees dict (fall back to the "default" key when a country is absent; within a country fall back to default for a missing provider): country_fees = { "ie": {"card": (0.019, 20), "klarna": (0.025, 40), "bank_transfer": (0.006, 0)}, "de": {"card": (0.025, 25), "klarna": (0.030, 45), "bank_transfer": (0.007, 0)}, "fr": {"card": (0.027, 28), "klarna": (0.032, 48), "bank_transfer": (0.008, 0)}, "default": {"card": (0.029, 30), "klarna": (0.035, 50), "bank_transfer": (0.008, 0)}, } Discount tiers (by cumulative completed-transaction count): count so far discount 1–10 0% 11–50 10% 51–100 15% 101+ 20% def calculate_fees(csv_data: str, country_fees: dict) -> str: ... # base = amount * rate + fixed # fee = int(base * (1 - discount)) # discount applied, then floored Example output (Part-1 rates, floor each fee): id,transaction_type,payment_provider,fee py_1,payment,card,59 py_2,payment,card,102 py_3,payment,klarna,169 py_4,payment,bank_transfer,40 py_1: 1000 × 0.029 + 30 = 59. py_2: 2500 × 0.029 + 30 = 102.5 → 102. py_3: 3400 × 0.035 + 50 = 169. py_4: 5000 × 0.008 = 40. Minority variant: some candidates report a rules table also keyed on status (different rate per status) and smaller fee figures (e.g. card → 51, klarna → 101) — confirm the exact rate table and rounding direction with the interviewer before coding. Examples Part 2 — same four rows plus a pending py_5 (ie, card, 2000): id,transaction_type,payment_provider,fee py_1,payment,card,39 py_2,payment,card,0 py_3,payment,klarna,125 py_4,payment,bank_transfer,40 py_5,payment,card,0 py_1: Ireland rate → 1000 × 0.019 + 20 = 39. py_2: failed → 0. py_3: Ireland klarna → 3400 × 0.025 + 40 = 125. py_4: France falls back to standard bank_transfer → 40. py_5: pending → 0. Part 3 — discount applied, then floored. For a completed ie/card tx of 1500 that is a merchant's 11th completed tx (10% tier): base = 1500 × 0.019 + 20 = 48.5; discounted = 48.5 × 0.9 = 43.65; floored → 43. Notes One report (Java) was told the interviewer explicitly evaluates Java candidates more leniently on time. The core skill is fast string handling — read CSV by hand, split, look up in a dict, emit lines. Multiple reports note clean output formatting matters more than algorithmic optimization. Both Part 1 and Part 3 run in O(n) (single pass); Part 3 also keeps O(m) merchant counters. Questions to clarify with the interviewer (Part 3) Is the discount applied before or after rounding? If a merchant sells across multiple countries, do all sales count toward one shared volume total? Does the volume count ever reset (e.g. monthly)? Edge cases to guard Empty input string. A payment_provider not present in the fee table. Malformed / short CSV rows. Preparation Drill clean CSV parsing without pandas (Stripe rounds typically want stdlib). Pre-write a rule-table evaluator: (provider, type, status) → rate. Practice writing the expected output to stdout exactly, including trailing-newline handling.