DrobyshevDev
Checks
What is checked, and what is not
Every project here publishes numbers. This page says, for each of them, which file holds it to the code and what happens when the two disagree — and then lists the claims nothing checks yet, because a page like this is worth reading only if the second list is real.
Numbers pinned to code
A number in prose rots quietly; a failing test does not. Each row below is a claim that appears in a README, a citation file or on this site, and the file that makes it answerable to the package that ships.
| Claim | Where it is stated | What holds it |
|---|---|---|
| decisionrl ships 32 algorithms, 24 environments, 9 of them applied | README prose and badges, CITATION.cff, .zenodo.json, pyproject.toml, this site |
tests/test_documented_counts.py |
| The applied tasks where a learned policy beats the classical baseline still do | decisionrl README results table | .github/workflows/verify.yml, nightly, 3 seeds |
| decisionrl line coverage does not fall below 80% | this site, Codecov badge | fail_under in pyproject.toml |
| glia needs no dependency outside the standard library | glia README badge, this site | dependencies = [], offline test suite |
| glia line coverage does not fall below 90% | glia CI | --cov-fail-under=90 |
| The English and Russian mlango READMEs carry the same badges and commands | both READMEs, and the PyPI landing page | tests/test_pipelines.py::TestReadme |
| The praxis golden set is the size the README says it is | praxis README, both languages | tests/test_eval.py |
| praxis's offline retrieval figures are what a run actually produces | the quality table in both praxis READMEs | tests/test_eval.py, under PRAXIS_OFFLINE=1 |
| praxis depends on this organisation's own packages, not on strangers' names | pyproject.toml extras |
tests/test_packaging.py |
| This site makes no third-party request of any kind | the legal page | scripts/check_site.py, on every change |
| The figures on the front page match the repositories they describe | the front page, both languages | scripts/check_figures.py, weekly |
What runs on a schedule
A check that runs only on a commit stops telling the truth the moment the world moves without one — a link rots, a model is withdrawn, a dependency is patched. These run whether or not anyone pushed.
| Check | How often | When it fails |
|---|---|---|
| Re-verify decisionrl's published results | nightly | the build goes red — the claim in the README is no longer reproducible |
| OpenSSF Scorecard: signed releases, pinned actions, patch latency | weekly, per repository | the grade drops, publicly, and findings land in code scanning |
| CodeQL static analysis | on every change and weekly | an alert opens in the repository's security tab |
| Outbound links on this site and in the research notes | weekly | one standing issue lists what broke |
| The figures on this site against their sources | weekly | one standing issue names the page, what it says, and what the source says |
| Organisation-wide maintenance triage | weekly | one issue lists what needs a decision, ordered by that rather than by repository |
Test suites themselves run on Linux, macOS and Windows, across every interpreter each package claims to support, because "works on my machine" is the failure mode all of this exists to rule out.
What nothing checks yet
This is the part of the page that makes the rest worth reading. These claims are published and are not held by anything; they are listed here rather than quietly counted among the checked ones.
- praxis on real models
- recall@5 0.92 and MRR 0.94 on the full corpus are stated in prose and measured on a GPU that CI does not have. The offline column of the same table is now held by a test — the offline path has no models, no network and no seed, so a run of it is reproducible — but the GPU column needs a scheduled re-run on real hardware, the way decisionrl re-verifies its applied claims nightly.
- The praxis corpus against the official publication
- The six codes shipped in the repository are transcriptions from Wikisource, marked in the corpus as pending a check against pravo.gov.ru. The parser for that source is how an edition gets confirmed, and it has not been run against every article.
- Anything about adoption
- No download counts, no star counts, no user numbers appear anywhere on this site, because none of them would survive the standard the rest of this page sets.
Each of these is a gap to close where it lives, not something to reword here. When one closes, it moves to the first table.
This page describes the checking regime rather than its current status; for the state of any individual run, the Actions tab of the repository in question is the source. History of this page — in the repository.