Testing judgment: confidence per cost

The checkout's basket and four of its tests. Each test starts with const b = new Basket(); b.add('mug', 1200, 2); b.add('tea', 450, 1);.

// basket.test.mjs
it('a', () => { expect(b.total('TEN')).toBe(2565); });
it('b', () => { expect(b.lines).toEqual([{ sku: 'mug', pricePence: 1200, qty: 2 }, { sku: 'tea', pricePence: 450, qty: 1 }]); });
it('c', () => { const spy = spyOn(pricing, 'discount'); b.total('TEN'); expect(spy.calls).toEqual([[2850, 'TEN']]); });
it('d', () => { expect(b.lines.length).toBe(2); expect(b.total('TEN')).toBe(2565); });

All four pass today. A refactor then changes how Basket is built, and changes no total it returns:

// basket.mjs (after the refactor; pricing unchanged)
export const pricing = {
  lineTotal: (line) => line.pricePence * line.qty,
  discount: (pence, code) => (code === 'TEN' ? Math.round(pence / 10) : 0),
};
export class Basket {
  constructor() { this.qty = new Map(); this.price = new Map(); }
  add(sku, pricePence, qty) { this.price.set(sku, pricePence); this.qty.set(sku, (this.qty.get(sku) ?? 0) + qty); }
  total(code) {
    let sub = 0;
    for (const [sku, n] of this.qty) sub += this.price.get(sku) * n;
    return code === 'TEN' ? sub - Math.round(sub / 10) : sub;
  }
}

The checkout's gift-card service calls one repository method, which runs this on Postgres:

UPDATE gift_cards SET balance_pence = balance_pence - $1
WHERE code = $2 AND balance_pence >= $1
RETURNING balance_pence;

Both gift-card bugs that reached customers last quarter were in that SQL: one let a card go below zero, and one charged a card that had been cancelled. The team has a day for gift-card tests.

Last week a three-item return refunded VAT on the first item only. An MR fixes refundVat() and adds refund.test.mjs, and CI is green on the MR.

An MR adds three parts to the checkout: countryName(code), which returns COUNTRIES[code] from a table generated from the ISO 3166 list; giftCardRounding(pence), copied unchanged from the old checkout, where it ran for a year with no reported bug; and the branch that refunds the card when the payment provider times out, which runs about once a week. The MR adds three tests of countryName(), one of giftCardRounding() and none of the timeout refund.

The basket API's reply and the web client's total label:

Two MRs are coming. MR 1 changes the reply to { total: (basket.totalPence / 100).toFixed(2) } and does not touch the client. MR 2 changes it to { totalPence: basket.totalPence, currency: 'GBP' }. The team adds one test now, before either MR, and wants it to fail on MR 1 and pass on MR 2.

it('A', () => { expect(totalLabel({ totalPence: 4599 })).toBe('£45.99'); });
it('B', () => { expect(totalLabel(basketBody({ totalPence: 4599 }))).toBe('£45.99'); });
it('C', () => { expect(basketBody({ totalPence: 4599 })).toEqual({ totalPence: 4599 }); });