I build and fix WordPress and WooCommerce sites.

I wrote the plugin behind it: how tickets are given out against an order, how the winner is picked, and the reports on who entered.

Each competition keeps a range of ticket numbers and a list of the ones already issued. Numbers can be allocated randomly or in sequence, set per competition, and a competition can be switched from one to the other part-way through: the plugin works out what has been issued so far and carries on. Allocation runs inside a per-product database lock, so two orders arriving together queue up instead of colliding.

A wrong answer to the skill question earns no tickets and doesn't count as a sale. If an order is cancelled, fails or is refunded, its numbers go back into the pool, and the competition page is purged from the LiteSpeed and Cloudflare caches so the tickets-left count stays right. Per-order and per-customer limits are enforced when someone adds to the cart and checked again when they log in. The entrant reports pull orders, quiz answers, coupons and ticket numbers into one list.

By the time the site was sold in 2025 it was running competitions of up to 10'000 tickets.

The page also shows each visitor the dates that suit them. A Cloudflare Worker reads the visitor's location at the edge, before the page reaches the browser. It tags the page with their continent and sets a cookie to remember it. People in the UK, Europe, Africa and Asia see the UK dates, and everyone else sees the US dates. Both lists are in the page, and a little CSS shows the right one. The same switch picks the contact numbers in the site header: the UK and European numbers for those regions, a North American number for everyone else. It fails safe: if anything goes wrong, the visitor just gets the normal page.

The site currently holds about 4'200 generated dates, 1'250 of them still to come, and the same dates feed the course data Google reads. I also added testimonial and review post types, and sent Contact Form 7 leads into Salesforce.

One model researches and drafts, and has to cite pages it actually opened. A script checks the draft for stock phrases, broken links, spelling and citations. A second model checks every claim against the pages the writer opened. If the reviewer isn't satisfied, the writer sends back exact edits instead of a new draft, and the script applies them so the rest of the text stays identical. If the reviewer is not satisfied after three rounds, the edits are marked for human review.

When a page does publish, the script re-reads the live page and restores the original from a snapshot if it doesn't match. Every change is logged, any page can be rolled back with one command, and the whole process pauses itself if too many pages fail, with a notification to the site administrator.

In week one it had published 5 pages, and the checks had refused 6, mostly over sources that didn't check out. That's the safeguards working. The one real fault was the live-page check treating straight quotes in a draft as different from the curly ones WordPress displays, so it rolled back a good edit that should have passed. The mistake was caught, the script adjusted, and the page published.

To comply with privacy laws relating to children, it doesn't accept anything that identifies a child beyond a first name and logs breaches by category of breach and person who breached, notifies the person of the breach, and reports the third and further breaches to the manager. A child mentioned is searched in a database to determine pronouns and age-band, ensuring a response is appropriate (speaking about a girl if it's a girl, or a boy if it's a boy, and not mentioning anything about - for example - nappies when speaking about a teenager). It gives no legal, medical or disciplinary advice. Each conversation resets at 02:00, and history is deleted after 30 days, as the Children's Home's records policy requires.

Every night WordPress sites are automatically backed up and copied to Google Drive. An automated script restores those backups into a temporary location and checks that the result confirms the backup, reporting any deviances.