Hourly SEO run for six sites: the proposal

This is a working proposal for running the SEO kit on a timer for six sites: Study Shot, Learn AI Alignment, Tank City Reboot and Two Towers on vlvd.net, plus Cheap Recorder and whitehatstoic. It sets out what exists today, what an hourly run would need, the risks, and the decisions still open.

The request, 9 October 2026: "make a cron job and every 1 hour, pump 3 to 5 high quality blogs on each of the projects that we have under vlvd.net, and both cheaprecorder and whitehatstoic.com."

Nothing is scheduled yet. A loop that runs by itself on a timer changes how the whole system behaves, so it is proposed first and built only after the open decisions below are made.

What exists today

Sites: all six sites and Electrified answer normally. These are the six sites the kit knows; another site needs its own settings, facts, seeds and SEO document first.
Who writes: the kit has no model key, so it cannot write articles by itself. Every article so far was written in a Claude Code session.
Topics: the topic queue held 4 topics in all. Without a model key or Search Console, the only source of new topics is each site's seed list of 8 to 12 titles, many already written.
Posts: 80 posts are published on Electrified. The 51 that were drafts were published on 9 October 2026 and each was checked in a browser.
Results: nothing is measured yet. Without PostHog and Search Console there is no way to tell which posts bring visitors.
Pace: the last run, 50 articles, took most of a day with a session working full time: opening sources, writing, making pictures, running the checks and reading each article end to end.
Shared parts: all sites share one topic queue, one account and one domain, electrified.vlvd.net. There is no lock, so two runs at once could write the same topic.

An hourly run as asked

Volume: 18 to 30 articles an hour, 432 to 720 a day, about 3,000 to 5,000 a week.
Writer: needs a model key with a budget, or a headless Claude Code run started by the timer. Either way nobody reads each article before it goes out.
Topics: the queue runs dry in the first hour; after that the job must invent its own titles.
Merging: 24 kit pull requests a day, which today only merge when a person starts the run.

Failure modes of an hourly run

Search ranking: Google's spam policies name scaled content abuse, many pages made mainly to rank, whether by people or AI. Every post lives on electrified.vlvd.net, so a penalty there would reach all six sites' posts at once.
Quality: the fact gate checks product claims, but nothing replaces the read-through that caught two wrong claims in the last run.
Overlap: an hour is shorter than a run. Without a lock, the next run starts while the last is still writing.
Other sites: each article opens research sources. Hundreds a day would put a real load on them; two already refused automated reads.
Cost: each article is a full agent loop with research, plus storage for 40 to 60 pictures an hour.
Silent failure: a job that runs while nobody watches needs a visible log, a stop switch and a cap.

The options

A, as asked: hourly, 3 to 5 per site, 432 to 720 a day. Needs a model key and budget, and titles invented by the job. High risk: a ranking penalty on the shared domain, wrong claims and cost.
B, recommended: daily, 1 per site, 6 a day or 42 a week. One timer on the Mac starts a full /seo-routine run at a set hour, with a lock, a daily cap, a log and an off switch. Low risk: the same checks as today at a volume that can be read.
C, middle: every 4 hours, 1 article, cycling through the sites: 6 a day, spread out. Low to medium risk: more runs for the same output.

With B or C, the volume goes up once the measure step shows which sites' posts bring visitors. That needs Search Console and PostHog connected first.

Decided on 9 October 2026

Publishing: scheduled posts go live on electrified.vlvd.net once they pass every check, instead of waiting as drafts.
This proposal is published here, on Electrified.

Still open

Cadence and volume: A, B or C.
Writer: a headless Claude Code run on the Mac, or a model key for the kit with a monthly budget.
Merging: whether each scheduled run's one pull request merges on green checks.
Sites: whether these four vlvd.net sites are all of them.