--- name: particula-report description: "Investigate and prepare a developer-facing GitHub issue for a researcher's suspected Particula bug, unexpected result, or perceived scientific error. Use for reporting concerns, not implementing fixes or submitting code PRs." --- # Investigate and report a concern Help the researcher communicate an actionable concern without requiring them to prove a bug or develop a fix. Read `../../references/scientific-practice.md` and `../../references/sources.md` for scientific checks and version-matched source discovery. Use `../../references/issue-report.md` for the report format. ## Investigate within the researcher's environment 1. Establish the intended calculation, observed discrepancy, expected behavior and research impact. Record the selected interpreter, Particula version, installation source/commit and relevant dependency/device details. Use `../../scripts/doctor.py` when local execution is available; inspect its output before including any of it in a public report. 2. Search version-matched documentation and the owning installed source. Follow relevant callers and inspect nearby tests when available. Use CodeGraph in an indexed checkout where available; otherwise use ordinary source search. A full development checkout is optional. Cite source paths and symbols, preferably commit-pinned links. Separate installed-release evidence from current upstream code; never infer that a main-branch fix is installed. 3. Reduce to a small runnable example with synthetic or shareable inputs when feasible. Preserve the original study and environment. Check relevant units, population measures, physical assumptions, conservation, convergence, seeds and backend support. Do not silently upgrade, switch backends, alter the research data or patch the library to make the discrepancy disappear. 4. Search open and closed issues and relevant PRs/fixes in https://github.com/uncscode/particula using the error, symbols and symptom. Read likely matches before calling something a duplicate. Record useful links and whether any fix applies to the researcher's version. If online access is unavailable, explicitly mark the duplicate search as not checked. 5. Separate observed results, hypotheses and confirmed findings. Report what was actually run and its outcome. If reproduction or source access is unavailable, say so and identify the next useful diagnostic step. Stop once the report scopes the concern sufficiently for developer triage; a fix or exhaustive investigation is not required. ## Review and submission Draft a concise symptom-based title and the complete issue body using the shared format. Retain its headings; mark missing information as unknown or not tested and omit irrelevant detail. Do not automatically classify an unconfirmed concern as a confirmed bug or apply a bug label. Show the researcher the destination, title and complete sanitized report for review. Ask for approval of that report before posting, even if the initial request was to file an issue. A matching issue should receive a proposed evidence-bearing comment for approval instead of an automatic duplicate issue. If a fix already exists, explain the applicable version and let the researcher decide whether additional reporting is useful; do not upgrade their environment. After approval, use available authenticated GitHub tools to create the issue in `uncscode/particula`, or add the approved comment to the agreed issue. Approval applies to the reviewed content and destination; substantive changes need another review. Check the returned result and provide the issue/comment link. If posting fails ambiguously, check for an existing submission before retrying. Without GitHub write access, provide the exact copyable Markdown and the repository's new-issue link; do not claim it was submitted.