Skip to content

Changelog

Substantive changes only — new pages, revised claims, corrections, retractions. Typos and styling are not recorded.

A public changelog exists here for the same reason last verified dates do: a knowledge base that changes silently cannot be trusted, and a reader deserves to know whether a page was revised after they read it.

2026-09-09 — Deployed

The site is live at probatio.yevgen.ponomarenko.io. Until this afternoon it had never been deployed, and the README claimed it was — everything published existed only locally and in a private GitLab.

Applying the infrastructure surfaced three faults that had been sitting latent because nothing had ever run it:

  • The root Terragrunt config was invalid and failed init outright — a Terraform-only argument in a Terragrunt block.
  • The S3 bucket name collided globally. It was composed from generic parts, and another AWS account already owned it. Bucket names are unique across all of AWS, not per account.
  • A CloudFront function redirected the site root to the CV, which would have made this page, and the home page, unreachable. The same redirect had been removed from page JavaScript hours earlier; the infrastructure held a second copy.

Verified after deploy: the home page returns its own content rather than redirecting, and the family memoir platform — which shares this AWS account and DNS zone — is unaffected.

2026-09-09 — The site has a name: Probatio

  • Named the work. Probatio — Latin for proof, testing, trial. It names what this site argues: that claims need evidence, and that evaluation is the capability which produces it. The site ships at probatio.yevgen.ponomarenko.io.
  • Why not Praxis, which was the working name: praxis is Greek rather than Latin, PRAXIS® is a registered trademark of ETS, and the bare word is crowded. Probatio is uncrowded and closer to the argument. Praxis stays as the internal concept — the framing of work as motion rather than display — and does not appear in published copy.
  • The brand names the work, not the person. The author remains prominent; Probatio is what the body of work is called.
  • Found and removed a redirect that would have hidden the home page. The CloudFront function 301'd the site root to the CV — the same bug that had been removed from page JavaScript earlier the same day, still present in the infrastructure. It had never been caught because the site has never been deployed.

2026-09-09 — Roadmap published, and an origin story added

  • A visual roadmap, prioritised and with its dependencies stated. Whitepaper and profile alignment come before any new building: the site now holds a case study, four method pages and a public build note, and almost nobody has been told.
  • Built in CSS, not JavaScript — the roadmap is the page most likely to be read on a phone, printed, or retrieved by an answer engine, and client-side rendering fails all three. The single diagram is Mermaid with the same information in a table beneath it.
  • The artifact pipeline moved to "not committed" — it solves drift between three documents that can be kept in step by hand.
  • About gained the engagement that changed direction: an end-to-end ML product shipped without a data engineer. It succeeded, and every hard problem we hit late was a data problem in disguise. That is why the data layer became the work, and the generative-AI wave keeps reproducing the pattern.

2026-09-09 — First publicly inspectable project added

  • Published a build note for the memoir publishing platform — a real public repository with 200 commits, four languages, three output formats, edge-gated deployment, and two licences in one repository. It is the first artefact on this site a visitor can open and read the source of.
  • Featured on the home page, replacing the entry about this site's own pipeline. A public repository with a deployment is stronger evidence than a description of the machinery you are already looking at.
  • Recorded the pattern it shares with the enterprise work: its content-completeness checker is the fail-loud pattern applied to a book instead of a knowledge corpus.

2026-09-09 — Lab rebuilt around build notes, not demos

  • Lab entries now carry the full arc — intent, build story, the trade-off exposed, an explicit split of what AI did and what I decided, evidence, and limits. An app without that treatment does not get a page.
  • First build note published: the publishing pipeline behind this site, including the confidentiality guard generalised for reuse.
  • Status labels extended with unsupported and conceptual. Designed-but-unbuilt work is listed as such, in those words, rather than quietly implied.
  • Three candidate builds published as conceptual, with the reasoning for the order: a semantic metric explorer first because it needs no model calls and no secrets, a retrieval evaluation workbench second, a decision-log generator third.
  • A joint project with my son added — an ontology of a fictional zone and its real-world lineage, built to test whether the provenance machinery holds up outside enterprise software.
  • Safety rules stated in advance for anything that ships with a live demo, so the claims stay matched to the validation.

2026-09-09 — Restructured around work in motion

The site was a CV with three placeholder pages — selective, retrospective, finished. It is now organised around five layers: position, proof, method, inquiry and contribution. The evidence is the same; the framing changed from a display case to a working practice.

  • New sections: Work, Methods, Lab, Writing, Now, About, Changelog.
  • Removed: the projects and research placeholder pages. Neither contained content; both were replaced by Work and Writing.
  • First case study published — a governed knowledge platform for a global CPG manufacturer, covering the prototype delivered between May and September 2026. States its maturity as a prototype and records what was not achieved.
  • Four method pages published, generalised from that engagement: fail-loud pipelines, intent routing, competency-question evaluation, evidence-gate investment cases.
  • Content status model adopted. Every substantive page now carries machine-readable status front matter and a visible status strip, so maintained guidance can be told apart from a working note.
  • Home page rewritten around a positioning statement rather than a keyword string.

2026-09 — Infrastructure

  • Migrated the site generator from MkDocs Material to Zensical.
  • Added a confidentiality guard to CI: the build fails if the private tier becomes tracked in git, or if a client-identifying term appears anywhere in the published tree. Negative-tested before being relied on.
  • Moved raw client and personal source material into an ignored tier. Nothing client-identifying has been published; the guard now makes that structural rather than a matter of care.