How we collect and compare prices

A price comparison is only worth as much as the judgement behind it. This page explains exactly how a number gets onto this site, what we do when shops disagree, and where the method breaks down. It is longer than a page like this usually is, deliberately — the interesting decisions are the ones most comparison sites do not show you.

1. Collection

Every hour, an automated job reads the public product pages of the Sri Lankan retailers we track — currently 15 shops across roughly 100 individual product listings. Nothing is entered by hand, and no shop sends us a feed. We read the same pages you would.

Shops are built on different platforms and publish prices in different places, so there is no single way to read them. We maintain eight separate readers: Shopify's product JSON, WooCommerce's embedded variation data, schema.org structured markup in two different shapes, React server-component payloads, in-page product objects, cart forms, and OpenGraph meta tags. Which reader applies to which shop is recorded per listing rather than guessed at runtime.

A price is stored with a timestamp every time it changes, which is what makes the price history charts possible. Note that this history began in August 2026, so for now it is shallow. It deepens on its own.

2. Working out which machine is which

This is the hard part, and it is most of the value. Shops describe identical hardware in incompatible ways, so before anything can be compared it has to be identified. Some real examples of what has to be handled:

• Some shops label a variant by colour alone — “Midnight”, “Starlight” — with the RAM and storage fixed by which page you are on and stated nowhere in the label.
• One shop labels its 15-inch variants “16/512gb” with no colour at all, so the colour has to come from elsewhere on the page.
• One page serves both the 13-inch and 15-inch machines from a single URL.
• Two shops publish an identical product slug, an identical title — including the same double space in the middle — and an identical base price. They are almost certainly one storefront under two names, so we label them as related rather than counting them as two independent quotes.
• Two shops' Mac mini URLs are swapped: the page at /mac-mini-m4 actually sells the M4 Pro. We map those by the verified product title, not by the URL.

Every quirk of this kind is recorded explicitly against the listing it affects. None of it is inferred by pattern-matching at runtime, because a pattern that silently mis-fires produces a confident wrong answer, which is worse than no answer.

3. Stock: what we will not guess

Shops report availability three different ways: as a real boolean, as a schema.org token that is sometimes an empty string, or not at all. For one retailer, an “Available colours” heading is the only hint the page gives.

When a shop publishes nothing, we record “unknown” and show a question mark. We never fall back to a cheerful default. Assuming in-stock would make the site look more decisive and would send people to shops that cannot sell them the machine, which is the worst error a site like this can make.

Where a shop's own signals contradict each other, we take the stricter one. One retailer's WooCommerce data reports a product as in stock while its structured markup on the same page correctly declares it a preorder. Taken at face value, that listing would have headlined as the cheapest available machine on the site — for something nobody could buy yet. We read its stock status from the markup instead.

That override is applied per-signal, not per-shop, and for a specific reason: a different retailer's structured markup is wrong about price while being right about stock. So for that listing we take price from one source and availability from the other. There is no single trustworthy source; there is only knowing which part of each page to believe.

4. Which price we show

• One row per shop. Where a shop lists the same configuration more than once, we show its lowest price rather than flooding the table.
• Sale prices are shown as sale prices. Where a shop publishes both a regular and a discounted price, the original is struck through so you can see the discount is real.
• Cash and card prices are labelled. Some retailers quote different amounts depending on payment method. Where we detect this, we say so rather than quietly showing the lower one.
• Delivery is not included in any price shown here.
• Shops that publish only a starting price are not ranked. Some retailers advertise one “from” figure covering every configuration. Ranking that against exact per-configuration prices would make it look artificially cheap, so those listings appear in a separate section, clearly marked, and are excluded from the comparison.

5. Colour changes the answer

Finish is not cosmetic in this market. A shop may stock only one colour of a machine, and its price for that colour can undercut everyone else. So the shop table filters by finish, and picking a different colour genuinely changes which retailer is cheapest.

Shops that publish a single price covering every finish are shown under all of them and tagged “all colours”, so you can tell a colour-specific quote from a general one.

6. What counts as the base model

Each machine is ranked on one entry configuration, which is what the “cheapest” comparison is awarded on. For both the MacBook Air M5 and the MacBook Pro M5 that is 16GB / 512GB.

This one is worth showing our working on, because we got it wrong first. Apple's configure-to-order flow for the 14-inch Pro starts at 1TB, which made 1TB look like the base build. But Apple's own technical specifications still list 512GB under standard storage, and 512GB is the configuration Sri Lankan retailers actually lead with. The 1TB build is still tracked and shown — it is simply not the one the cheapest badge is awarded on.

7. Known limits

Stated plainly, because a methodology page that only lists strengths is marketing:

• Prices lag. We check hourly. A shop can change a price or sell out minutes after we look. Always confirm on the retailer's page before paying.
• Not every shop can be read. One retailer blocks automated requests outright and has been removed rather than guessed at. Another intermittently times out, which is why a sync sometimes reports as partial.
• Some distinctions cannot be made. The M5 Pro ships in different CPU and GPU core counts at the same RAM and storage, at different prices — but most shops never publish which bin they are selling. Where that is the case we keep the lower price and cannot tell you which chip it is.
• We do not verify retailers. Inclusion is not endorsement. We do not check whether a shop is an authorised reseller, what warranty it offers, or whether it is reputable.
• Coverage is new hardware only, from Sri Lankan online retailers with public pricing. Refurbished, used and grey-market listings are out of scope.

Corrections

If a price, specification or stock status here is wrong, tell us at hello@macpricelk.top and it will be fixed. See About for how the site stays independent, and Terms for the formal position on accuracy.