Every production release adds a dated entry here: what changed, what was measured and what was left alone on purpose. Newest first.
An almost empty plinth. The notes fill it in. Conceptual illustration. Not a customer result.
These notes describe shipped work and measured checks. They are not claims about traffic, rankings, customers or revenue.
FAQ, About and Glossary pages, and an llms.txt
pagesstructured-data
Three pages now answer the questions a careful buyer asks before paying, in the same words the report uses. The FAQ carries twelve questions and answers. The first six are the pairs the homepage already showed, byte for byte, and the page emits FAQPage structured data that a test compares with the rendered text.
The About page says what the service is, how it is operated, what it does not do and how to reach us. Its Organization data carries the PO Box as a postal address and the support contact. Founder, legal entity, founding date and operator city stay empty until the owner confirms them in writing, and an empty field renders nothing.
The glossary defines eighteen terms once each, with the wording from the methodology and the sample report: mention, citation, discovery and branded questions, coverage gap, retained evidence, finding and the rest. The methodology and every field guide link to those anchors.
A plain-text llms.txt summarises the offer for answer engines: the engine sampled today, the refund sentences, the contact, and links to published pages only. Where these pages name the price or the engine, the value comes from the catalog or from configuration at render time, so an unfilled placeholder fails the page instead of shipping, and a test holds the llms.txt engine line to the same configured label.
Fourteen illustrations from one miniature world
designimagery
Every illustration on this site is conceptual, and the new set of fourteen keeps to one world: warm ivory buildings and objects on low limestone plinths, matte forest green, one luminous chartreuse accent, spherical topiary, no people and no words. The homepage neighborhood was the reference. Fourteen new scenes were rendered against it with Azure's GPT Image 2.5 deployment, using image edits rather than fresh generations so the material and the light stay consistent.
The set, rendered and reviewed on October 13: page heroes for the sample report, the methodology, the resource desk, the checklist, the FAQ, the About page, the learn section and the question pages; three field guide covers; a pricing ladder; an order signpost; and a nearly empty plinth. Each master is 1536 by 1024 pixels and was reviewed one by one, then derived to AVIF and WebP at 640, 960 and 1440 pixels wide with the saturation eased to 0.91 on the web copies.
All fourteen are placed. The heroes open the sample report, the methodology, the resource desk, the checklist, the learn index and its articles, the FAQ, the About page, the question bank with its category pages and the free question planner; the three covers sit on the guide pages, the desk cards and the home shelf; the pricing ladder stands beside the price cards on the home page; the signpost waits on the order page and the checkout confirmation; and the empty plinth opens this notes page. The 404 page keeps its plain text until a later craft pass. Every placement reserves its box with explicit dimensions, keeps the whole 3:2 frame and ends its caption with the same words: conceptual illustration, not a customer result.
Every public page also got its own social card, rendered from the same artwork with the site's own type and the page title; only the home and sample cards name the price, and it comes from the catalog. A provenance file in the repository records the deployment, the prompt hash and the hash of every master and derivative, and a second one records every card. None of these is a photograph, a screenshot of an answer or a customer result, and their alt text calls them conceptual models.
Checkout now carries the site's identity
checkout
Card payment runs on Stripe Checkout, on a Stripe account shared with another product. Without its own settings, that account's dashboard branding, a purple page with the other product's name, would front every Top of Answers payment.
Each Checkout session now sets its own branding: the name Top of Answers, the paper background and forest-green button from the site's palette, rounded borders and the site icon. Stripe fetches the icon from a URL, so it is attached only when the site runs on a public https origin; the local purchase harness, which runs on a loopback address, gets the colours and the name without it.
The legal business name still appears on receipts and in Stripe's terms. Prices, question plans and the webhook did not change, and a contract test asserts the branding on both payment and subscription session parameters.
Both search consoles verified, one sitemap each
search
Google Search Console confirms ownership of the domain property. The verification record was already public in DNS when the property was reopened, and both authoritative nameservers return the same value. The sitemap was submitted once and read back as Success with 11 discovered pages.
Bing Webmaster Tools imported the site through its Google import, with only this site selected and an unrelated property unchecked. Its sitemap list was empty, so the canonical sitemap was submitted once. The read-back at 13:32 UTC showed Success, 11 URLs, one sitemap, zero errors and zero warnings.
Discovery is not indexing, ranking, traffic or revenue, and both performance reports were still processing. One loose end stays open: Google's settings view reported no robots file while the live file answers 200 with the intended exclusions and the sitemap declared. We left the working file alone and will recheck when crawl data arrives.
The hero came back at full size
designrelease
A redesign that went live at 01:06 UTC on October 13 had shrunk the homepage illustration from 704 to 180 pixels wide on desktop, grown the visible homepage copy from 735 to 1,118 words and stretched the mobile page from 7,642 to 10,315 pixels. We corrected it the same morning. The earlier homepage was rebuilt from its own commit rather than read from an old screenshot folder, and 15 routes and states were captured at 1440 and 390 pixels for the comparison.
The release restores the original ivory composition at about 719 pixels wide, brings visible homepage copy down to 750 words and mobile height to 8,922 pixels, and moves the full fictional specimen into a native details disclosure beside the report section. The sample PDF and the blank worksheet are shown as direct renders of their real first pages, not generated images, and drawn diagrams organise the process and the guides. Paid links consistently read Preview my audit. Form markup, payment handlers, prices and configuration did not change.
Before release: typecheck, a build with the outbound-link gate, 768 unit tests, 42 operator tests and 77 browser tests passed, with two live-only cases intentionally skipped; 24 axe page and viewport combinations reported zero violations; a local mobile Lighthouse run scored 100 in all four categories with LCP 1.5 seconds and CLS 0. After deployment at 06:23 UTC, 13 page routes answered 200, five asset hashes matched, and one live mobile Lighthouse run read 99, 100, 100 and 100 with LCP 1.9 seconds. Lab runs, not field data.
Our own automated screenshot reviews scored the hero 9 of 10 on desktop and mobile, the sample report opening 9, the resource desk 8.5 and the methodology 8. Those are internal design judgments, not customer research or evidence of sales.
How this site is built
Top of Answers is an Astro site served by one Cloudflare Worker. The same Worker serves the pages and the API routes, consumes the audit queue and runs the drain every five minutes. Most pages are prerendered at build time; the free check, checkout and order pages render on request.
The type is self-hosted: Onest for text and Instrument Serif italic for the emphasised line in headings, both under the SIL Open Font License 1.1, subset to Latin and converted to woff2 from a pinned commit of the Google Fonts repository. No font loads from a third-party host. Cited sources appear as plain text, and the only outside link on the site is our own Bluesky profile.
The illustrations are conceptual, generated from written prompts with Azure's GPT Image 2.5 deployment in one miniature world of ivory buildings, forest green and a single chartreuse accent, and their alt text calls them conceptual models. They are never photographs, screenshots of answers or customer results. The sample PDF and worksheet previews are direct renders of the real documents.
The engine named across the site, OpenAI GPT with web search, is the one configured today. The FAQ, the About page and checkout read that name from configuration at render time, and the price comes from the product catalog, so the pages, checkout and every report describe the same source and the same price. Checkout confirms the configured sources before payment.
Every change runs the same local gates before it ships: a TypeScript typecheck, the unit suite, the operator suite, a production build that fails on any outbound link, and browser tests at 320, 390, 768 and 1440 pixels with and without JavaScript. One more test refuses em and en dashes in page copy.
Know what appeared. Choose what comes next.
One business. One US city. A $149 one-time PDF audit.