What this site measures, and what it leaves alone.

What is tracked, and the question each event answers

Seven events, no more. Each exists to answer one product question; an event without a question is noise that costs quota.

• Page Viewed

  Which pages are reached at all, and from where, so the landing page can be judged against the pages behind it. Carries the path, the title, and the referring domain.

• Product Card Clicked

  Which build draws interest, and whether that is the build or the slot: if position one always wins, the grid order is doing the work, not the card. Carries the product and its position, one to six.

• Case Study Opened

  Whether readers arrive through the grid or land directly from a shared link, which changes what the page's opening has to do. Carries the product and the entry point.

• Case Study Depth Reached

  Whether case studies are read or skimmed. A card click is interest and a full scroll is attention, and those should not be conflated. Fires at 25, 50, 75, and 100 percent, once each per page load, four at most.

• Resume Downloaded

  Which page turns a reader into someone who wants the résumé, the closest thing this site has to a conversion. Carries the page it was clicked from.

• Outbound Link Clicked

  Whether readers go on to the code or the live product, which is the difference between reading a claim and checking it. Carries the destination and whether it is GitHub, a live demo, LinkedIn, or other.

• Contact Initiated

  Whether anyone reaches out, and by which channel, so the contact options are shaped by use rather than habit. Carries the method: email or LinkedIn.

What is deliberately not tracked

This list is the part that took judgment. Every item was available with a checkbox and was turned off on purpose.

• No session recording. It would capture the contact form as a visitor types it, and a portfolio has no use for a replay of someone reading.
• No form field capture. Nothing on a form here is analytically useful and everything on it is potentially personal.
• No click-everything autocapture. Recording every click and its text would swamp the seven named events without answering a question they do not already answer.
• No user identification. Anonymous device IDs only. No user ID, no email matching, no enrichment from any other source. There is no logged-in user on a portfolio, and knowing who is reading is unnecessary and hostile.
• No cookies. Identity is held in localStorage, so the site sets no cookies and needs no consent banner under most readings.
• No time on page. It cannot distinguish a reader from an abandoned tab, and reporting it as engagement would be a number that is wrong more often than right.
• No location finer than country. Country answers whether anyone in the markets I am applying in is reading. City would not answer anything more.
• No data from 2 to 7 September 2026. The instrumentation was live before the author exclusions were, so that window carries my own traffic and is excluded from any count published here.

What the numbers cannot answer

• Who is reading. A recruiter, a hiring manager, a peer, and the author checking a deploy all look the same. The referring domain hints at the channel, not the reader.
• Whether a read led to anything. A phone screen, an interview, or an offer cannot be attributed to a session. The only downstream signal is a contact click, which measures a click, not a conversation.
• What a depth milestone means. Full depth on a long case study is strong evidence of attention. On a short page it is the default.
• Repeat readers. The same person on a phone and a laptop counts twice, and a cleared browser counts as new.
• How much is missing. Content blockers and fast navigations both undercount, and the size of the undercount is not measurable from inside the data.

The counts

The full taxonomy, including property names and the implementation notes behind each decision, is kept in the site's repository alongside the code that fires the events.