Building a publishing pipeline that keeps releases and acquisition apart
Evidence, inputs, outputs, costs and media sit in bounded production packs, a Go manifest drives a streaming Rust renderer, and activation is atomic — with 22 of the 24 acquisition techniques deliberately left manual.
Publishing at scale is two problems that usually get solved badly together: producing the pages, and deciding what happens to them afterwards. I kept them apart, because the second one is where a mistake becomes public.
Producing a release
Each release is a bounded production pack: the evidence behind it, its inputs and outputs, what it cost, the media it carries, and the approve/hold/reject state of every article in it. A Go process writes a manifest, a Rust renderer streams that manifest into pages, a validator checks the result, and activation is atomic — so a half-published release is not a state the site can be in. Multilingual scheduling runs through the same path, which means a translated release is held to the same bounds as the original. I built eight renderer architectures along the way. Radiant, Signal and Atlas are the working examples, and a validated 50,000-page stream is the scale test.
Acquisition, and where I stopped automating
The catalogue carries 24 versioned acquisition techniques. I automated two of them — IndexNow and XML-RPC, through a Rust distribution worker. The other 22 stayed manual or proposal-only, because a technique that publishes to somebody else's platform should be a decision a person makes. Every one of them carries an explicit opt-in, a budget, a pause control and a recorded outcome. Nothing here claims traffic, rankings, leads or revenue: what I have evidence for is what the pipeline did, not what it earned.
One measurement, and its scope
A 50,000-row cannibalization-audit persistence diagnostic measured 6,558 ms on the earlier run and a three-run median of 1,010.058 ms on the later one. That is how long it took to write audit rows to the database. It is not renderer or release timing, and it should not be read as one.
Where it stands
The pipeline and its previews are real; the results are the results of the pipeline. The other publishing projects are linked as related work rather than presented as the same product.
Have a system like this?
Describe the workflow, the data or the system you need built. Scope and any quote are handled privately after the enquiry.
Published