The repository came before the brief

The lifecycle in this estate has a fixed order:

idea -> brief -> research -> requirement -> {adr, spec} -> decompose -> build

The plan for this series gave episode 02 to the research stage. This site has no research stage to report. It also has no requirement stage. This episode reports the order in which things actually happened, and what the missing stages would have caught.

The order

The homepage stated who I am and what I claim for 35 minutes before the product it belongs to had a brief.

Nothing objected

The skill that seeds a repository took no product as input. The two checkers that gate every canonical write test a repository’s internal coherence: files against the index, the index against the graph. Neither asks whether the thing the repository serves exists upstream. Nothing in canonical state connected the two.

I found it by reading a status message. No gate found it. Episode 00 reported that briefly. Episode 03 reports the decision it forced.

What the missing stages produce

Other products in the estate show what this site skipped.

A research artefact states a question and answers it from sources. One for another product asks what the product can “responsibly train and measure”, and cites five sources.

A requirement carries acceptance criteria a test can check. One for the same product requires that “every target uses a non-colour identifying cue”.

This site’s brief has a section called “Research / prior art”. It makes claims about the field and cites no source. There is no requirement for this site.

What the skip cost

Nobody wrote down how readers find the site. The brief says the reader arrives “from a professional network, a search, or a referral”. No requirement turned that into something a test could check. On 12 October, five weeks after the first deploy, I looked:

• page titles were 8 to 18 characters, such as Episodes;
• there was no sitemap and no robots file;
• no page had a link preview tag, so a shared link showed no summary;
• the www address served a second copy of every page.

All four were fixed in one day, as one objective. The work was small. The gap was that nothing had asked for it.

The brief asked a question that was already answered. It lists “How is the site published, and from where?” as open, for a decision record. That decision was made two days earlier, on 7 September, in ADR-0022. A brief written before the repository would have asked it in time. This one asked it after the answer had been built.

The cadence was decided, but only in my head. The brief asks “What sustains cadence when delivery is slow?”, and says it “needs deciding before it happens rather than during”. I had a cadence: one episode a week. No record held it.

So the dates mislead. Episode 00 went live on 12 September and episode 01 on 13 October, 31 days later. That was not a missed cadence. I held episode 00 back from any audience, because the site’s look and feel needed a serious overhaul first. A reader of the records alone could not tell a held launch from a stalled series.

The cadence is now a decision record, written on 13 October: the series launches with three episodes, then publishes every few days. The brief’s question is still half open. The record sets the pace. It does not say what the series does when the work pauses.

What the skip did not cost

The build was sound. The deployment pipeline, the identity smoke test and the CI gate were built without a requirement, and they work. In the same days, gates caught a dependency pin with eight security advisories and an apex domain that served someone else’s page.

Every gap in this episode is about what the site is for: who reads it, how they find it, and how often it publishes. Those are the questions research and requirement exist to answer, and they are the ones the build could not answer for itself.

Next

The repository came first, so the first decisions came from the build, not from a requirement. Episode 03 is about those decisions.