Byte Journal is written by one person and assembled with a large amount of automation. This page explains the real process, including the parts that are machine-run, so a reader can judge a story on what actually stands behind it.
The three formats
News — a short brief on something that has happened. Written from cited reporting and primary material, published under the Byte Journal name.
Analysis — context on a development: what changed, what it compares to, and what the consequences are. Written from multiple sources.
Field Notes — firsthand work. Experiments, measurements, logs, commands, failures and fixes from running real infrastructure. Written by Arxchitect.
Each article states which one it is at the top of the page.
How a story is found
Feeds, vendor newsrooms, project release channels and security advisories are polled on a schedule. A cheap deterministic pass runs first and throws away anything that obviously fails: duplicates of what is already published, items outside the subject areas, malformed or empty entries, and items whose source cannot be identified. Rejecting early is deliberate — expensive research is never spent on a story that a basic check has already refused.
What survives is scored for newsworthiness (a breach, an outage, a release, a regulation, a pricing change) rather than for keyword volume. There is no article quota. A slow day produces fewer stories, and a day on which nothing passes produces none.
What has to be true before publication
Every news brief must cite its sources at the foot of the page, and those sources are checked rather than trusted:
- Independent corroboration. A claim taken from one outlet is confirmed
against a different one. Byte Journal’s configuration states
publish_requires = "corroborated"andallow_single_source = false, and a brief cannot publish while those conditions are unmet. A vendor’s own announcement can satisfy the factual bar for its own product, but it does not by itself turn into an article about the wider story. - Source reputation. Sources are classified, and a syndicated copy, a press release reprint, an aggregator entry or a scraped mirror does not count as a second independent source. Two copies of one wire story are one source.
- Syndication and copy detection. Near-verbatim reproduction of a source’s text is measured, not eyeballed. A draft whose wording overlaps its source beyond a threshold is rewritten or dropped; it is not published.
- Claim and figure checking. Numbers, prices, dates, versions and percentages in a brief are traced back to the sources. Where sources disagree, the disagreement is stated in the article instead of one value being picked quietly.
- Compliance review. Before publication each brief passes a check covering the publication’s own rules — sourcing present, subjects named honestly, no advice it cannot support.
- Post-publication verification. After publication the live page is fetched again and checked: it exists, it renders, and its sources are present.
What the automation is allowed to do
Automation assists with research organisation, drafting, summarisation, comparison and consistency checking. It is not permitted to relax any of the gates above.
- AI confidence is not evidence.
- Memory is not evidence.
- Search results are not evidence.
- Embedding similarity is not evidence.
- An agent’s agreement is not evidence.
Only the underlying sources satisfy the evidence requirement. If a checking step cannot run, the item is held rather than passed — an unmeasured check is not a passed check. A failed search is a hold, never a pass.
Images
Images are optional. An article without one is correct; an article illustrated with the wrong thing is not. Before an image is used it has to be:
- relevantly about the story’s actual subject — checked for name collisions (an “Apple” story is not illustrated with an apple orchard, an Apple River or Apple Records), and a story about a specific product is not illustrated with a different product from the same brand;
- licensed for reuse, with the licence recorded as the source states it;
- attributed with the creator and licence the source gives, shown with the image on the article page.
Where no image passes all three, the article publishes without one. The publication never generates illustrative imagery for news: a synthetic picture of a breach, a person, a product or an event is not used. Social-sharing previews fall back to a Byte Journal cover card, which is layout, not a depiction.
Corrections
Errors are corrected, not quietly edited away. A correction is added to the article with what was wrong and what the correct position is; a substantive change that is not an error is recorded as an update. Both appear in a short note at the foot of the article. See Corrections for how to report one.
What this page is not claiming
This is a personal publication with automation behind it, and it says so. There is no newsroom, no second editor and no human fact-checking desk. The safeguards above are enforced by code, and their state is not a secret: the publishing configuration that carries them is the same one described here.
The controls are the point. An automated publication that cannot state its own rules is not trustworthy — and one that claims to be something it isn’t is worse.