matbench-composition-stack
public since Oct 2026- Python
- Language
- Unlicensed
- Licence
- Oct 2026
- Last push
Second paper (private until submission): cross-fitted stacking + local OOF residual correction on Matbench composition tasks
Open source
Everything public on this GitHub account that has code in it: what each repository is, the licence it ships under, where it is published, and what the tree actually carries. Read from the GitHub API rather than typed out by hand.
There are no star counts on this page. These repositories were published between 2024 and 2026 and none has an adoption history worth quoting yet — the licence, the packaging, the CI and the governance are the parts that are verifiable today, so those are what is shown.
Second paper (private until submission): cross-fitted stacking + local OOF residual correction on Matbench composition tasks
Numerical edge-case testing for array code. Runs your functions on NaN, ±inf, overflow, subnormals, cancellation and float16/32/64, and compares them with high-precision references and metamorphic properties. Works as a pytest plugin, a CLI and a GitHub Action. Optional LLM agent suggests fixes that must pass every check. pip install numguard-ci
edge-cases · floating-point · github-actions · jax · llm-agents · numerical-analysis · numpy · pytest-plugin
Check AI-written statistics with evidence. Re-runs analyses and notebooks, tests each statistical test's assumptions on the real data, recomputes reported p-values (statcheck-style) and lints analysis code. Posts the result on pull requests; an optional LLM agent proposes corrected analyses that are re-run to verify them. pip install stataudit
data-science · github-actions · hypothesis-testing · jupyter · llm-agents · p-values · python · reproducibility
Check whether a pull request actually does what it claims. Runs fail-to-pass tests, replays issue reproducers and maps which diff hunks the evidence needs, then posts a verdict card on the PR. Optional LLM agent writes reproduction tests that are only counted once they're executed and verified. pip install proofcheck-ai
ai-generated-code · ci · code-review · developer-tools · fail-to-pass · github-actions · llm-agents · maintainers
SIG-Net: similarity-informed local-competence mixtures of cross-fitted experts for OOD molecular property prediction (JCIM manuscript, code, per-fold results)
Research software engineering postings ask for open-source Python packaging, project governance and reproducible harnesses by name. These are the ones that exist in the repositories above, not a list of things worth doing some day.
fork field. Also excluded: the profile README repository, and repositories created but never pushed to./users/bhaskargurram-ai/repos, unauthenticated and fetched on the server, cached for six hours. Language, licence, description, topics and dates are GitHub’s. Registries, governance files and the notes under each description are from profile/master-profile.yaml and each repository’s own tree. If the API fails, the page serves a dated snapshot and labels it.