One human author and AI models used as tools. The roles are separate and are recorded separately, because merging them would misstate all of them.
KalevIT — author, maintainer, decision owner.
Originated the project. Defined its purpose, scope and target certification. Set the requirements for each mode. Made accessibility a first-class requirement rather than an afterthought. Commissioned the audit against the official objectives. Decided the config-driven direction and accepted the restructure it required. Raised the licensing question that produced the current content boundary. Reviewed, accepted or rejected every artefact.
Every decision in decisions/ is his, including the ones taken against the
advice given.
Claude Sonnet 4.6 (Anthropic), April–May 2026 — the application layer that survives in
this repository: the five mode pages, style.css, and a11y.js including every
accessibility mode it provides.
An earlier question bank and the concept maps were also produced in that phase. Neither is part of this repository: the bank was retired under ADR-0004 and the maps are held back pending a provenance decision. This file credits what is here, not what once was.
Claude Opus 5 (Anthropic), July 2026 — audit against the official objectives, the config-driven engine, the weighted sampler, both CI gates, the content-licensing boundary, the 40 original sample questions, and this documentation.
Neither wrote a commit directly: neither has repository access. Everything was reviewed and committed by the author.
| Verified | Method | Result |
|---|---|---|
| Sample bank structural integrity | tools/validate-questions.js |
0 errors |
| Sample coverage vs official weights | cross-checked against lpi.org objectives v1.6 | exact match, 0.0 pt deviation on all 19 |
| Weighted sampling | 2,000 simulated exams | 7/9/9/8/7 per topic, 40.00 per exam |
| Learning material licence | read from the imprint of the official PDF | CC BY-NC-ND 4.0, confirmed |
| Page load path | browser load order simulated in Node | all seven globals resolve |
Not verified: the factual correctness of every question. Structural validity is mechanical; a wrong answer key is not machine-detectable. Check anything surprising against the official learning material.
Responsibility for this repository — code, questions, the claims in the README and any defects — rests entirely with the human author. Models are tools. Tools do not answer for their output.
AI-assisted commits carry an Assisted-by: trailer naming the model:
Assisted-by: Claude Opus 5 (Anthropic)
Co-Authored-By: is deliberately not used. It asserts co-authorship of a work by a tool,
and the address conventionally paired with it is not controlled by Anthropic — commits
using it have been misattributed to unrelated GitHub accounts. See
ADR-0005.
Enable the hook locally:
git config core.hooksPath .githooks