Last updated: 13 September 2026
IT Pro Insights is an independent, one-person technical publication. This page
explains exactly how an article here gets made, checked, published, and kept
current — and who is answerable for it.
AI assistance and human accountability
Drafts are produced with AI assistance. This is disclosed at the top of every
single article, in the same words: "Drafted with AI assistance. We review and
correct articles on an ongoing basis."
Nothing publishes itself. Every article is read, corrected, and explicitly
approved before it goes live by Thorsteinn Halldorsson, who writes, edits, and
publishes this site and is the named human responsible for everything on it. An
article that has not been through that review does not reach the site — the
publishing script only ever reads from the approved queue, so skipping the step is
not possible by accident.
Research and sourcing
Topics come from real operational problems: changes shipping in the platform, gaps
in the vendor documentation, and the topic requests readers send in.
Version-specific and behavioural claims have to be traceable to an authoritative
source — vendor documentation (Microsoft Learn, Azure docs), project
documentation (Kubernetes, Terraform, Flux, Argo CD), standards bodies (IETF/RFC,
W3C, NIST, CISA, OWASP), or the relevant language and database docs. This is not
a good intention: the pre-publish check counts the authoritative citations in
tutorials, how-tos, guides, and deep dives, warns at one, and fails the article
at zero. Blog posts and vendor marketing pages do not count as sources.
"Tested against"
Many articles end with a "Tested against" footer naming the exact product,
API, CLI, and module versions the steps and commands were written and checked
against — because a command that is correct on one API version can be wrong on the
next.
Treat it as an expiry label rather than a badge: if your versions differ from the
ones listed, verify before you run. Where a platform is known to move fast, that
footer says so explicitly and tells you to check the current vendor documentation
first.
The two-stage check
Every draft passes through an automated gate before a human ever sees it as a
candidate for publication.
- Automated quality gate. A scripted check runs over the draft and reports
errors and warnings. It verifies structure and length, meta description
length, the presence of real code examples and a diagram, no duplicate H1,
working internal links, authoritative citations, headline and keyphrase rules,
readability, and exactly three related-article links. It also flags generic
filler and AI boilerplate — phrases like "as an AI", "I hope this helps",
leftover TODO markers, and placeholder text. Errors are not advisory: the
approval script refuses to move a failing draft into the approved queue. - Human review. The gate can only check form. A draft that passes is then
read end to end, corrected, and approved by hand. Approval is a deliberate,
separate step — the gate passing is a precondition, never a decision.
Articles that cannot be made accurate are not published, and there is no volume
target that overrides that.
Publishing cadence
Approved articles are published in small batches on a schedule, not dumped in
bulk. Each approved article carries a planned publish date; a scheduled weekday
job flags the ones that have come due, and a human then promotes them.
The promotion script caps an ordinary run at two articles, and refuses a
larger batch unless the operator explicitly overrides it. Flipping many posts
live at once gives them all the same publication timestamp, which is both
misleading to readers and a pattern that quality reviews flag. Publishing steadily
is a deliberate constraint enforced in code, not a habit.
Updates, revisions, and superseded articles
Technology moves, and published guidance goes stale. When an article is
meaningfully revised after publication, a "Reviewed" date appears beside the
byline showing when it was last updated. Every article always shows its original
publication date.
Articles that no longer earn a place in search results are not deleted. A
thin or superseded article is marked noindex and dropped from the sitemap while
staying reachable at its original URL — links that people saved keep working, and
anything already cited elsewhere does not turn into a dead end. Where a newer
article genuinely supersedes an older one, the old URL redirects to its
replacement. Quietly deleting published work to tidy up an archive is not
something this site does.
Independence from advertisers
The site is self-funded and earns revenue from display advertising. Advertisers
have no input into what gets covered, how a product is assessed, or when an
article is published, and no advertiser sees an article in advance. No affiliate
links are currently used; if any are introduced, each will be disclosed where it
appears and will never change a recommendation. Recommendations are made on
technical merit alone. Guest posts, sponsored articles, and link-insertion
requests are not accepted.
Corrections
A confirmed factual error gets fixed in the article as soon as it is verified,
and corrections are prioritised over all other mail. How to report one, and what
happens next, is on the corrections page.
Questions
Anything about how this site is made:
thorsteinn.halldorsson@gmail.com. See also
about, the terms of use, and the
privacy policy.