How We Launched a Lab-Supply Storefront in a Day on the Tacavar SEO/GEO Stack

On October 12, 2026 we stood up a new storefront, Research Water Supply, and had it taking real checkout traffic the same day. It is a distribution brand of NextGen Biologics USA that sells one product: Research Water, sterile water with 0.9% benzyl alcohol preservative, supplied in sealed glass and plastic vials for research use only, with a certificate of analysis (COA) for each lot and shipping from the USA. This post is the working timeline of how the day went, written for operators who run their own sites on agent tooling. It is not a launch announcement and it contains no traffic or revenue numbers, because a day-one site has none worth quoting.

The point of the exercise was to test whether the stack we already run for our own properties could absorb a brand-new site without custom work. Most of it could. The parts that needed attention are the interesting ones, so we have called those out.

Morning: plan from keyword data, not from taste

We did not start with a design. We started with the keyword-opportunity scorer that already runs weekly across our sites, pointed it at the product category, and let the data say which pages were worth building first. The output was a short list of page types: a product page, a bulk-quote page, a COA page, and a handful of supporting articles. That list became the sitemap. Anything the data did not support stayed off the first release.

One decision came from the product rather than the keywords. The store sells a laboratory consumable, so the copy has to stay inside what the product is: a preserved sterile water for research use. We wrote a short list of banned phrasings before drafting a single page, and the drafting agents were given that list as a hard constraint. Cheap to do on day one, expensive to retrofit.

Midday: clone the pattern instead of inventing one

The storefront is a Next.js static export hosted on Cloudflare Pages, the same shape as our other sites. Rather than scaffold from scratch, we cloned an existing site pattern and swapped in the brand, the product schema, and the page set. That gave us the header and footer components, the metadata helpers, the JSON-LD patterns, the sitemap generator, and the Pages configuration for free. The saving is not typing time. It is that every one of those pieces had already survived contact with our publishing checks, so the new site did not rediscover old bugs.

Checkout is the one dynamic part. We used Stripe through its plain REST API from Cloudflare Pages Functions, with no SDK in the bundle. A Pages Function creates the checkout session server-side and the browser is redirected to the hosted page. Keeping it to a couple of small functions meant the whole payment path was readable in one sitting, and the site stays a static export everywhere else. Bulk buyers are routed elsewhere on purpose: the bulk quote page collects the request so a person can answer it, rather than pushing large orders through a cart.

Afternoon: publish through the gate

Our sites no longer deploy with a bare build-and-upload. Every deploy goes through a gated wrapper, tacavar-publish, which builds from a clean worktree, runs content and citation gates, smoke-tests a preview, and writes an audit entry. If a gate objects, the deploy reports a hold with a reason and waits for a human to approve it. We do not force holds. For a new site this is exactly the behavior you want, because the first release is the one most likely to contain a careless sentence. Registering the new site with the publish tool was a configuration entry rather than a project, and the same entry is what enrolls it in the canary and digest checks.

Registration did a second job. Adding the site to our site registry is what places it into the scheduled SEO, GEO, and content jobs: the weekly keyword run, the generative-search probes, and the article pipeline all read that registry. We did not wire any of those up by hand for this site. We added one record and the crons picked it up on their next pass.

Late afternoon: measurement and mail through APIs

Search Console and GA4 were set up through service-account API calls rather than clicking through consoles: verify the property, submit the sitemap, create the analytics property and tag. The value of doing it through the API is repeatability. The next site gets the same sequence from the same script, and the steps are written down because they are code. Email followed the same logic. Store addresses are routed with Cloudflare Email Routing configured through its API, so the domain could receive mail the same day without a mailbox server.

Throughout: delegation to a cheaper model

The mechanical work was delegated. A lead session kept the plan and made the judgment calls; Sonnet-class subagents did the bounded jobs, such as drafting page copy against the banned-phrase list, writing the supporting posts, and filling in the config. Each subagent got a narrow brief, a list of files it could touch, and explicit paths it must not touch. Two rules made this work. First, a subagent reports what it changed and the lead verifies with grep and curl rather than trusting the summary. Second, anything that touches money or compliance wording stays with the lead. Delegation saved time, but it is the verification step that made the saved time safe to use.

What we would flag if you copy this

Four things are worth carrying over. Start from the keyword data so the first release is small and defensible. Clone a proven site pattern so you inherit its gates. Put a gated publisher in front of production before the first deploy, not after the first incident. And write down the language the product may not use before any agent drafts a page. Products in regulated-adjacent categories punish loose copy, and an agent will happily write loose copy at scale if nobody tells it not to.

What we do not claim: that a day-one site ranks, converts, or earns. It does not, and anyone selling that on day one is guessing. What a day buys you is a correct foundation: indexable pages, a working checkout, measurement, mail, and enrollment in the recurring jobs that do the compounding work afterward.

See the result

If you want to see what came out of it, the Research Water product page describes the item and its specifications, the COA page explains how lot documentation is handled, and the bulk quote form is where larger orders start. The brand home page is at researchwatersupply.com. For the distributor behind it, see NextGen Biologics USA.

If you want the same SEO/GEO operations layer running your own properties, the Tacavar Growth Starter is the self-serve way in. For the orchestration thinking behind the delegation pattern, our teardown of coordinated Claude Code sessions covers the same checkpoint idea from another angle.