tutor

Projects that follow the best practices below can voluntarily self-certify and show that they've achieved an Open Source Security Foundation (OpenSSF) best practices badge.

There is no set of practices that can guarantee that software will never have defects or vulnerabilities; even formal methods can fail if the specifications or assumptions are wrong. Nor is there any set of practices that can guarantee that a project will sustain a healthy and well-functioning development community. However, following best practices can help improve the results of projects. For example, some practices enable multi-person review before release, which can both help find otherwise hard-to-find technical vulnerabilities and help build trust and a desire for repeated interaction among developers from different companies. To earn a badge, all MUST and MUST NOT criteria must be met, all SHOULD criteria must be met OR be unmet with justification, and all SUGGESTED criteria must be met OR unmet (we want them considered at least). If you want to enter justification text as a generic comment, instead of being a rationale that the situation is acceptable, start the text block with '//' followed by a space. Feedback is welcome via the GitHub site as issues or pull requests There is also a mailing list for general discussion.

We gladly provide the information in several locales, however, if there is any conflict or inconsistency between the translations, the English version is the authoritative version.
If this is your project, please show your badge status on your project page! The badge status looks like this: Badge level for project 16171 is passing Here is how to embed it:
You can show your badge status by embedding this in your markdown file:
[![OpenSSF Best Practices](https://www.bestpractices.dev/projects/16171/badge)](https://www.bestpractices.dev/projects/16171)
or by embedding this in your HTML:
<a href="https://www.bestpractices.dev/projects/16171"><img src="https://www.bestpractices.dev/projects/16171/badge"></a>


These are the Passing level criteria. You can also view the Silver or Gold level criteria.

Baseline Series: Baseline Level 1 Baseline Level 2 Baseline Level 3

        

 Basics 13/13 ●

  • General

    Note that other projects may use the same name.

    The AI side of NovaTrail's tutor: a Go library that reads a learner's work into an observation, reviews the working and words the tutor's replies, with an eval runner on the same code.

    Please use SPDX license expression format; examples include "Apache-2.0", "BSD-2-Clause", "BSD-3-Clause", "GPL-2.0+", "LGPL-3.0+", "MIT", and "(BSD-2-Clause OR Ruby)". Do not include single quotes or double quotes.
    If there is more than one language, list them as comma-separated values (spaces optional) and sort them from most to least used. If there is a long list, please list at least the first three most common ones. If there is no language (e.g., this is a documentation-only or test-only project), use the single character "-". Please use a conventional capitalization for each language, e.g., "JavaScript".
    The Common Platform Enumeration (CPE) is a structured naming scheme for information technology systems, software, and packages. It is used in a number of systems and databases when reporting vulnerabilities.
  • Basic project website content


    The project website MUST succinctly describe what the software does (what problem does it solve?). [description_good]
    This MUST be in language that potential users can understand (e.g., it uses minimal jargon).

    The README says in its first lines what tutor is: the AI side of the tutor of NovaTrail, an open platform for adaptive learning, interactive lessons and AI tutoring. It is a Go library that api, another component of the platform, calls in its own process: it reads the learner's work into an observation, reviews the working, assesses an essay against the module author's rubric and words the hints, and its eval runner is a command-line tool on the same code. https://github.com/altnovatrail/tutor#readme



    The project website MUST provide information on how to: obtain, provide feedback (as bug reports or enhancements), and contribute to the software. [interact]

    The README says how to obtain tutor (clone the repository with git clone --recurse-submodules and work in its devcontainer) and how to use what it has now, starting with just board. GitHub shows the organization's CONTRIBUTING.md and SECURITY.md on this repository's page, and they say how to give feedback and contribute: an issue first, then a pull request whose commits are signed off under the DCO, and a vulnerability reported privately. https://github.com/altnovatrail/.github/blob/main/CONTRIBUTING.md



    The information on how to contribute MUST explain the contribution process (e.g., are pull requests used?) (URL required) [contribution]
    We presume that projects on GitHub use issues and pull requests unless otherwise noted. This information can be short, e.g., stating that the project uses pull requests, an issue tracker, or posts to a mailing list (which one?)

    Changes come as GitHub pull requests: an issue first, then a pull request whose every commit is signed off under the Developer Certificate of Origin. The branch main takes changes only through pull requests whose required checks pass, and its ruleset has no bypass. The process is in the organization's CONTRIBUTING.md, which GitHub shows for this repository: https://github.com/altnovatrail/.github/blob/main/CONTRIBUTING.md



    The information on how to contribute SHOULD include the requirements for acceptable contributions (e.g., a reference to any required coding standard). (URL required) [contribution_requirements]

    CONTRIBUTING.md lists what every contribution keeps to (no learners' data, no secrets, the devcontainer, licenses compatible with Apache-2.0, a DCO sign-off) and points to the rules every NovaTrail repository shares, RULES.md, which this repository's CLAUDE.md imports from spec at the tag it pins and which include the rule on tests: https://github.com/altnovatrail/.github/blob/main/CONTRIBUTING.md and https://github.com/altnovatrail/spec/blob/v0.16.7/RULES.md


  • FLOSS license


    The software produced by the project MUST be released as FLOSS. [floss_license]
    FLOSS is software released in a way that meets the Open Source Definition or Free Software Definition. Examples of such licenses include the CC0, MIT, BSD 2-clause, BSD 3-clause revised, Apache 2.0, Lesser GNU General Public License (LGPL), and the GNU General Public License (GPL). For our purposes, this means that the license MUST be: The software MAY also be licensed other ways (e.g., "GPLv2 or proprietary" is acceptable).

    Apache-2.0, an OSI-approved license, covers the code and the documents. https://github.com/altnovatrail/tutor/blob/main/LICENSE



    It is SUGGESTED that any required license(s) for the software produced by the project be approved by the Open Source Initiative (OSI). [floss_license_osi]
    The OSI uses a rigorous approval process to determine which licenses are OSS.

    Apache-2.0 is approved by the OSI.



    The project MUST post the license(s) of its results in a standard location in their source repository. (URL required) [license_location]
    One convention is posting the license as a top-level file named LICENSE or COPYING, which MAY be followed by an extension such as ".txt" or ".md". An alternative convention is to have a directory named LICENSES containing license file(s); these files are typically named as their SPDX license identifier followed by an appropriate file extension, as described in the REUSE Specification. Note that this criterion is only a requirement on the source repository. You do NOT need to include the license file when generating something from the source code (such as an executable, package, or container). For example, when generating an R package for the Comprehensive R Archive Network (CRAN), follow standard CRAN practice: if the license is a standard license, use the standard short license specification (to avoid installing yet another copy of the text) and list the LICENSE file in an exclusion file such as .Rbuildignore. Similarly, when creating a Debian package, you may put a link in the copyright file to the license text in /usr/share/common-licenses, and exclude the license file from the created package (e.g., by deleting the file after calling dh_auto_install). We encourage including machine-readable license information in generated formats where practical.

    The license is the top-level file LICENSE: https://github.com/altnovatrail/tutor/blob/main/LICENSE


  • Documentation


    The project MUST provide basic documentation for the software produced by the project. [documentation_basics]
    This documentation must be in some media (such as text or video) that includes: how to install it, how to start it, how to use it (possibly with a tutorial using examples), and how to use it securely (e.g., what to do and what not to do) if that is an appropriate topic for the software. The security documentation need not be long. The project MAY use hypertext links to non-project material as documentation. If the project does not produce software, choose "not applicable" (N/A).

    The README says what tutor is and how to use what it has now, each with its command: the check board (just board), which asks for the eval keys without showing them, its record mode (just record), the sheets that code draws, a measurement run and the report on its log. The Environment section of CLAUDE.md says how to start: clone with the spec submodule and open the repository in its devcontainer. RULES.md, which CLAUDE.md imports, says what keeps the work safe: everything inside the devcontainer, exact versions checked against their sums, no secrets and no learners' data in the repository. The packages, the recipes and the checks are in section 3.2 of docs/architecture.md. https://github.com/altnovatrail/tutor#readme



    The project MUST provide reference documentation that describes the external interface (both input and output) of the software produced by the project. [documentation_interface]
    The documentation of an external interface explains to an end-user or developer how to use it. This would include its application program interface (API) if the software has one. If it is a library, document the major classes/types and methods/functions that can be called. If it is a web application, define its URL interface (often its REST interface). If it is a command-line interface, document the parameters and options it supports. In many cases it's best if most of this documentation is automatically generated, so that this documentation stays synchronized with the software as it changes, but this isn't required. The project MAY use hypertext links to non-project material as documentation. Documentation MAY be automatically generated (where practical this is often the best way to do so). Documentation of a REST interface may be generated using Swagger/OpenAPI. Code interface documentation MAY be generated using tools such as JSDoc (JavaScript), ESDoc (JavaScript), pydoc (Python), devtools (R), pkgdown (R), and Doxygen (many). Merely having comments in implementation code is not sufficient to satisfy this criterion; there needs to be an easy way to see the information without reading through all the source code. If the project does not produce software, choose "not applicable" (N/A).

    tutor's external interface is the Go API of its public packages and the command line of its eval runner. Every exported identifier of the public packages, tutor, provider and the adapters provider/claude and provider/gpt, has a Go doc comment, 26 of 26 on 2026-11-01 (https://github.com/altnovatrail/tutor/blob/main/provider/doc.go). tutor-eval run without arguments prints its commands with their flags, each command's -h describes every flag with its default, and the README walks through the commands (https://github.com/altnovatrail/tutor/blob/main/cmd/tutor-eval/main.go). What the library is to become, call by call, is in sections 3.5 to 3.10 of docs/architecture.md.


  • Other


    The project sites (website, repository, and download URLs) MUST support HTTPS using TLS. [sites_https]
    This requires that the project home page URL and the version control repository URL begin with "https:", not "http:". You can get free certificates from Let's Encrypt. Projects MAY implement this criterion using (for example) GitHub pages, GitLab pages, or SourceForge project pages. If you support HTTP, we urge you to redirect the HTTP traffic to HTTPS.

    The project's site is its GitHub repository, served over HTTPS, and Go code takes the module over HTTPS too, through the Go module proxy: https://github.com/altnovatrail/tutor



    The project MUST have one or more mechanisms for discussion (including proposed changes and issues) that are searchable, allow messages and topics to be addressed by URL, enable new people to participate in some of the discussions, and do not require client-side installation of proprietary software. [discussion]
    Examples of acceptable mechanisms include archived mailing list(s), GitHub issue and pull request discussions, Bugzilla, Mantis, and Trac. Asynchronous discussion mechanisms (like IRC) are acceptable if they meet these criteria; make sure there is a URL-addressable archiving mechanism. Proprietary JavaScript, while discouraged, is permitted.

    GitHub issues and pull requests: searchable, addressable by URL, open to anyone with a GitHub account, and nothing to install. https://github.com/altnovatrail/tutor/issues



    The project SHOULD provide documentation in English and be able to accept bug reports and comments about code in English. [english]
    English is currently the lingua franca of computer technology; supporting English increases the number of different potential developers and reviewers worldwide. A project can meet this criterion even if its core developers' primary language is not English.

    The README, CLAUDE.md, AGENTS.md, the documents under docs/ but for the archives of the epics, the code, its comments and the prompts are in English, and issues and pull requests are accepted in English or Russian. The work plan, RUN.md, its technical debt, DEBT.md, and the archives of its epics under docs/plan/ are in Russian, the author's working language.



    The project MUST be maintained. [maintained]
    As a minimum, the project should attempt to respond to significant problem and vulnerability reports. A project that is actively pursuing a badge is probably maintained. All projects and people have limited resources, and typical projects must reject some proposed changes, so limited resources and proposal rejections do not by themselves indicate an unmaintained project.

    When a project knows that it will no longer be maintained, it should set this criterion to "Unmet" and use the appropriate mechanism(s) to indicate to others that it is not being maintained. For example, use “DEPRECATED” as the first heading of its README, add “DEPRECATED” near the beginning of its home page, add “DEPRECATED” to the beginning of its code repository project description, add a no-maintenance-intended badge in its README and/or home page, mark it as deprecated in any package repositories (e.g., npm deprecate), and/or use the code repository's marking system to archive it (e.g., GitHub's "archive" setting, GitLab’s "archived" marking, Gerrit's "readonly" status, or SourceForge’s "abandoned" project status). Additional discussion can be found here.

    Under active development: 104 pull requests merged since the repository was created on 2026-10-19, the latest on 2026-11-01. https://github.com/altnovatrail/tutor/commits/main


 Change Control 9/9 ●

  • Public version-controlled source repository


    The project MUST have a version-controlled source repository that is publicly readable and has a URL. [repo_public]
    The URL MAY be the same as the project URL. The project MAY use private (non-public) branches in specific cases while the change is not publicly released (e.g., for fixing a vulnerability before it is revealed to the public).

    A public git repository on GitHub: https://github.com/altnovatrail/tutor



    The project's source repository MUST track what changes were made, who made the changes, and when the changes were made. [repo_track]

    Git records what changed, who changed it and when, and every commit of a pull request is signed off under the DCO, which a required check verifies.



    To enable collaborative review, the project's source repository MUST include interim versions for review between releases; it MUST NOT include only final releases. [repo_interim]
    Projects MAY choose to omit specific interim versions from their public source repositories (e.g., ones that fix specific non-public security vulnerabilities, may never be publicly released, or include material that cannot be legally posted and are not in the final release).

    Every pull request, its branch and its commits are public before they reach main.



    It is SUGGESTED that common distributed version control software be used (e.g., git) for the project's source repository. [repo_distributed]
    Git is not specifically required and projects can use centralized version control software (such as subversion) with justification.

    The repository is git, hosted on GitHub.


  • Unique version numbering


    The project results MUST have a unique version identifier for each release intended to be used by users. [version_unique]
    This MAY be met in a variety of ways including a commit IDs (such as git commit id or mercurial changeset id) or a version number (including version numbers that use semantic versioning or date-based schemes like YYYYMMDD).

    Every release has a unique version vX.Y.Z, tagged on its commit by .github's shared release workflow, which numbers it; the project's rule is that no tag is made by hand. Releases are immutable, a ruleset forbids moving or deleting a v* tag once it is made, and the Go module proxy and its checksum database hold each version of the Go module to its bytes. https://github.com/altnovatrail/tutor/blob/main/.github/workflows/release.yml



    It is SUGGESTED that the Semantic Versioning (SemVer) or Calendar Versioning (CalVer) version numbering format be used for releases. It is SUGGESTED that those who use CalVer include a micro level value. [version_semver]
    Projects should generally prefer whatever format is expected by their users, e.g., because it is the normal format used by their ecosystem. Many ecosystems prefer SemVer, and SemVer is generally preferred for application programmer interfaces (APIs) and software development kits (SDKs). CalVer tends to be used by projects that are large, have an unusually large number of independently-developed dependencies, have a constantly-changing scope, or are time-sensitive. It is SUGGESTED that those who use CalVer include a micro level value, because including a micro level supports simultaneously-maintained branches whenever that becomes necessary. Other version numbering formats may be used as version numbers, including git commit IDs or mercurial changeset IDs, as long as they uniquely identify versions. However, some alternatives (such as git commit IDs) can cause problems as release identifiers, because users may not be able to easily determine if they are up-to-date. The version ID format may be unimportant for identifying software releases if all recipients only run the latest version (e.g., it is the code for a single website or internet service that is constantly updated via continuous delivery).

    Versions are MAJOR.MINOR.PATCH, as Go modules take them: each release is the next patch of the current line, 0.1, starting at v0.1.0, and a new line is a change of RELEASE_LINE in the release workflow.



    It is SUGGESTED that projects identify each release within their version control system. For example, it is SUGGESTED that those using git identify each release using git tags. [version_tags]

    Each release is the git tag vX.Y.Z, which the release workflow creates together with the release, and the tag is also the Go module's version.


  • Release notes


    The project MUST provide, in each release, release notes that are a human-readable summary of major changes in that release to help users determine if they should upgrade and what the upgrade impact will be. The release notes MUST NOT be the raw output of a version control log (e.g., the "git log" command results are not release notes). Projects whose results are not intended for reuse in multiple locations (such as the software for a single website or service) AND employ continuous delivery MAY select "N/A". (URL required) [release_notes]
    The release notes MAY be implemented in a variety of ways. Many projects provide them in a file named "NEWS", "CHANGELOG", or "ChangeLog", optionally with extensions such as ".txt", ".md", or ".html". Historically the term "change log" meant a log of every change, but to meet these criteria what is needed is a human-readable summary. The release notes MAY instead be provided by version control system mechanisms such as the GitHub Releases workflow.

    A release follows every merge to main that changes what the Go module is built from, and its notes on GitHub list the pull requests it brings, linked to each pull request's description: an English summary of what changed and why. https://github.com/altnovatrail/tutor/releases



    The release notes MUST identify every publicly known run-time vulnerability fixed in this release that already had a CVE assignment or similar when the release was created. This criterion may be marked as not applicable (N/A) if users typically cannot practically update the software themselves (e.g., as is often true for kernel updates). This criterion applies only to the project results, not to its dependencies. If there are no release notes or there have been no publicly known vulnerabilities, choose N/A. [release_notes_vulns]
    This criterion helps users determine if a given update will fix a vulnerability that is publicly known, to help users make an informed decision about updating. If users typically cannot practically update the software themselves on their computers, but must instead depend on one or more intermediaries to perform the update (as is often the case for a kernel and low-level software that is intertwined with a kernel), the project may choose "not applicable" (N/A) instead, since this additional information will not be helpful to those users. Similarly, a project may choose N/A if all recipients only run the latest version (e.g., it is the code for a single website or internet service that is constantly updated via continuous delivery). This criterion only applies to the project results, not its dependencies. Listing the vulnerabilities of all transitive dependencies of a project becomes unwieldy as dependencies increase and vary, and is unnecessary since tools that examine and track dependencies can do this in a more scalable way.

    No publicly known vulnerability of tutor has been fixed in any release. If one is, its release notes will name it and link its GitHub security advisory.


 Reporting 8/8 ●

  • Bug-reporting process


    The project MUST provide a process for users to submit bug reports (e.g., using an issue tracker or a mailing list). (URL required) [report_process]

    Bugs are reported as GitHub issues: https://github.com/altnovatrail/tutor/issues



    The project SHOULD use an issue tracker for tracking individual issues. [report_tracker]

    GitHub issues track individual reports. The project's own planned work is in its work plan, RUN.md.



    The project MUST acknowledge a majority of bug reports submitted in the last 2-12 months (inclusive); the response need not include a fix. [report_responses]

    No bug report has been submitted yet (2026-11-01); the maintainer answers each report in its issue.



    The project SHOULD respond to a majority (>50%) of enhancement requests in the last 2-12 months (inclusive). [enhancement_responses]
    The response MAY be 'no' or a discussion about its merits. The goal is simply that there be some response to some requests, which indicates that the project is still alive. For purposes of this criterion, projects need not count fake requests (e.g., from spammers or automated systems). If a project is no longer making enhancements, please select "unmet" and include the URL that makes this situation clear to users. If a project tends to be overwhelmed by the number of enhancement requests, please select "unmet" and explain.

    No enhancement request has been submitted yet (2026-11-01); the maintainer answers each request in its issue.



    The project MUST have a publicly available archive for reports and responses for later searching. (URL required) [report_archive]

    GitHub issues are public, permanent and searchable: https://github.com/altnovatrail/tutor/issues?q=is%3Aissue


  • Vulnerability report process


    The project MUST publish the process for reporting vulnerabilities on the project site. (URL required) [vulnerability_report_process]
    Projects hosted on GitHub SHOULD consider enabling privately reporting a security vulnerability. Projects on GitLab SHOULD consider using its ability for privately reporting a vulnerability. Projects MAY identify a mailing address on https://PROJECTSITE/security, often in the form security@example.org. This vulnerability reporting process MAY be the same as its bug reporting process. Vulnerability reports MAY always be public, but many projects have a private vulnerability reporting mechanism.

    The security policy, which every repository of the organization shares, says how to report a vulnerability: privately, through GitHub's private vulnerability reporting. It also says that the maintainer gives a first response within 14 days. https://github.com/altnovatrail/tutor/security/policy



    If private vulnerability reports are supported, the project MUST include how to send the information in a way that is kept private. (URL required) [vulnerability_report_private]
    Examples include a private defect report submitted on the web using HTTPS (TLS) or an email encrypted using OpenPGP. If vulnerability reports are always public (so there are never private vulnerability reports), choose "not applicable" (N/A).

    Private vulnerability reporting is enabled for tutor: a report goes over HTTPS to the maintainers and opens a draft advisory that only they and the reporter see. https://github.com/altnovatrail/tutor/security/advisories/new



    The project's initial response time for any vulnerability report received in the last 6 months MUST be less than or equal to 14 days. [vulnerability_report_response]
    If there have been no vulnerabilities reported in the last 6 months, choose "not applicable" (N/A).

    No vulnerability has been reported in the last six months: tutor has no security advisory, draft or published (2026-11-01).


 Quality 13/13 ●

  • Working build system


    If the software produced by the project requires building for use, the project MUST provide a working build system that can automatically rebuild the software from source code. [build]
    A build system determines what actions need to occur to rebuild the software (and in what order), and then performs those steps. For example, it can invoke a compiler to compile the source code. If an executable is created from source code, it must be possible to modify the project's source code and then generate an updated executable with those modifications. If the software produced by the project depends on external libraries, the build system does not need to build those external libraries. If there is no need to build anything to use the software after its source code is modified, select "not applicable" (N/A).

    tutor is Go. just build runs go build ./..., which builds every package from source, the eval runner tutor-eval among them, and go run ./cmd/tutor-eval builds and starts the runner. The Go is pinned: Go 1.27.3 in the toolchain stage of .devcontainer/Dockerfile, from an archive checked against its SHA-256, with GOTOOLCHAIN=local, and go.sum holds the hash of every module the build takes. CI builds that stage from a base image pinned by digest on every run and compiles every package in its checks: the tests, the linters and CodeQL's build of the Go code. https://github.com/altnovatrail/tutor/blob/main/justfile



    It is SUGGESTED that common tools be used for building the software. [build_common_tools]
    For example, Maven, Ant, cmake, the autotools, make, rake (Ruby), or devtools (R).

    The go command builds, vets and tests the code, with golangci-lint, govulncheck and go-licenses beside it; Docker builds the environment; and just runs every recipe.



    The project SHOULD be buildable using only FLOSS tools. [build_floss_tools]

    All FLOSS: Go (BSD-3-Clause), golangci-lint (GPL-3.0, a development tool that ships nothing), govulncheck (BSD-3-Clause), go-licenses (Apache-2.0), Docker Engine (Apache-2.0), Node.js (MIT), which runs the documents check, and just (CC0-1.0).


  • Automated test suite


    The project MUST use at least one automated test suite that is publicly released as FLOSS (this test suite may be maintained as a separate FLOSS project). The project MUST clearly show or document how to run the test suite(s) (e.g., via a continuous integration (CI) script or via documentation in files such as BUILD.md, README.md, or CONTRIBUTING.md). [test]
    The project MAY use multiple automated test suites (e.g., one that runs quickly, vs. another that is more thorough but requires special equipment). There are many test frameworks and test support systems available, including Selenium (web browser automation), Junit (JVM, Java), RUnit (R), testthat (R).

    tutor's automated test suite is the go command's own, go test, over 117 test files beside the code they test, in this repository under Apache-2.0. The tests run on synthetic strokes and sheets that the code draws and call no AI model: the adapters are tested against a local fake server. just test runs them, and just ci-test runs them with the race detector and without the test cache, as the required check Go / test does on every pull request and on main: https://github.com/altnovatrail/tutor/blob/main/.github/workflows/ci.yml



    A test suite SHOULD be invocable in a standard way for that language. [test_invocation]
    For example, "make check", "mvn test", or "rake test" (Ruby).

    go test ./..., the standard way for Go, runs every test. just test runs it, just ci-test runs it as CI does, with the race detector on every package that starts goroutines, and just --list names the recipes.



    It is SUGGESTED that the test suite cover most (or ideally all) the code branches, input fields, and functionality. [test_most]

    On 2026-11-06 the tests covered 95.2% of the statements of the module, as go test -coverprofile measures them. Every package is at 86% or more except the provider port, provider, at 54%. CI sends the coverage of each pull request and each commit to main to Codecov, and its check codecov/patch asks for 90% of new and changed lines: https://codecov.io/gh/altnovatrail/tutor



    It is SUGGESTED that the project implement continuous integration (where new or changed code is frequently integrated into a central code repository and automated tests are run on the result). [test_continuous_integration]

    GitHub Actions runs every check and every test on each pull request and on each commit to main. main takes a pull request only when they pass and its branch is up to date with main. https://github.com/altnovatrail/tutor/blob/main/.github/workflows/ci.yml


  • New functionality testing


    The project MUST have a general policy (formal or not) that as major new functionality is added to the software produced by the project, tests of that functionality should be added to an automated test suite. [test_policy]
    As long as a policy is in place, even by word of mouth, that says developers should add tests to the automated test suite for major new functionality, select "Met."

    RULES.md, the rules every NovaTrail repository shares, which this repository's CLAUDE.md imports, says that code comes with its tests: a pull request that adds or changes what code does adds or changes the tests of its automated suite too. https://github.com/altnovatrail/spec/blob/v0.16.7/RULES.md



    The project MUST have evidence that the test_policy for adding tests has been adhered to in the most recent major changes to the software produced by the project. [tests_are_added]
    Major functionality would typically be mentioned in the release notes. Perfection is not required, merely evidence that tests are typically being added in practice to the automated test suite when new major functionality is added to the software produced by the project.

    The latest major changes came with their tests: the essays assessed in two calls, with tests of the call, its report and its command (https://github.com/altnovatrail/tutor/pull/101); the assessment by the rubric (#91); and the leak check of every hint, with tests that run spec's 17 leak vectors (#83).



    It is SUGGESTED that this policy on adding tests (see test_policy) be documented in the instructions for change proposals. [tests_documented_added]
    However, even an informal rule is acceptable as long as the tests are being added in practice.

    The rule on tests is in RULES.md, which the organization's CONTRIBUTING.md names as the rules every contribution follows. The checklist of the pull request template asks that code come with its tests: https://github.com/altnovatrail/.github/blob/main/.github/pull_request_template.md


  • Warning flags


    The project MUST enable one or more compiler warning flags, a "safe" language mode, or use a separate "linter" tool to look for code quality errors or common simple mistakes, if there is at least one FLOSS tool that can implement this criterion in the selected language. [warnings]
    Examples of compiler warning flags include gcc/clang "-Wall". Examples of a "safe" language mode include JavaScript "use strict" and perl5's "use warnings". A separate "linter" tool is simply a tool that examines the source code to look for code quality errors or common simple mistakes. These are typically enabled within the source code or build instructions.

    golangci-lint runs its standard set (errcheck, govet, ineffassign, staticcheck and unused) with errorlint, misspell, nolintlint and unconvert, and checks gofmt and goimports formatting. just lint runs it, and CI runs it as the required check Go / lint (https://github.com/altnovatrail/tutor/blob/main/.golangci.yml). The recipes run bash with set -euo pipefail. SonarCloud analyses the code on every pull request, and its quality gate is a required check: https://sonarcloud.io/summary/new_code?id=altnovatrail_tutor



    The project MUST address warnings. [warnings_fixed]
    These are the warnings identified by the implementation of the warnings criterion. The project should fix warnings or mark them in the source code as false positives. Ideally there would be no warnings, but a project MAY accept some warnings (typically less than 1 warning per 100 lines or less than 10 warnings).

    On 2026-11-01 golangci-lint reports 0 issues, the code has no //nolint, and SonarCloud shows no open issue. A finding of SonarCloud or CodeQL is fixed in code before the next task starts; marking one false is a last resort, with the reason written down. https://sonarcloud.io/project/issues?id=altnovatrail_tutor



    It is SUGGESTED that projects be maximally strict with warnings in the software produced by the project, where practical. [warnings_strict]
    Some warnings cannot be effectively enabled on some projects. What is needed is evidence that the project is striving to enable warning flags where it can, so that errors are detected early.

    Go / lint fails on any finding of the linters, and nolintlint reports a //nolint that suppresses nothing. SonarCloud's quality gate holds a merge on any new bug or vulnerability, and the answer to a finding is a change to the code.


 Security 16/16 ●

  • Secure development knowledge


    The project MUST have at least one primary developer who knows how to design secure software. (See ‘details’ for the exact requirements.) [know_secure_design]
    This requires understanding the following design principles, including the 8 principles from Saltzer and Schroeder:
    • economy of mechanism (keep the design as simple and small as practical, e.g., by adopting sweeping simplifications)
    • fail-safe defaults (access decisions should deny by default, and projects' installation should be secure by default)
    • complete mediation (every access that might be limited must be checked for authority and be non-bypassable)
    • open design (security mechanisms should not depend on attacker ignorance of its design, but instead on more easily protected and changed information like keys and passwords)
    • separation of privilege (ideally, access to important objects should depend on more than one condition, so that defeating one protection system won't enable complete access. E.G., multi-factor authentication, such as requiring both a password and a hardware token, is stronger than single-factor authentication)
    • least privilege (processes should operate with the least privilege necessary)
    • least common mechanism (the design should minimize the mechanisms common to more than one user and depended on by all users, e.g., directories for temporary files)
    • psychological acceptability (the human interface must be designed for ease of use - designing for "least astonishment" can help)
    • limited attack surface (the attack surface - the set of the different points where an attacker can try to enter or extract data - should be limited)
    • input validation with allowlists (inputs should typically be checked to determine if they are valid before they are accepted; this validation should use allowlists (which only accept known-good values), not denylists (which attempt to list known-bad values)).
    A "primary developer" in a project is anyone who is familiar with the project's code base, is comfortable making changes to it, and is acknowledged as such by most other participants in the project. A primary developer would typically make a number of contributions over the past year (via code, documentation, or answering questions). Developers would typically be considered primary developers if they initiated the project (and have not left the project more than three years ago), have the option of receiving information on a private vulnerability reporting channel (if there is one), can accept commits on behalf of the project, or perform final releases of the project software. If there is only one developer, that individual is the primary developer. Many books and courses are available to help you understand how to develop more secure software and discuss design. For example, the Secure Software Development Fundamentals course is a free set of three courses that explain how to develop more secure software (it's free if you audit it; for an extra fee you can earn a certificate to prove you learned the material).

    The maintainer designs for least privilege, fail-safe defaults and a small attack surface. Every workflow starts with a read-only token, and a job gets only the write it needs: CodeQL and Scorecard to upload their results, and Scorecard an OIDC token to publish them; the workflows hold no secret. The branch main has no bypass, and a merge waits for the secret scan, the DCO sign-off, the documents check, the four Go checks, CodeQL's analyses of the Go code and of the workflows, and SonarCloud. The library's request types have no fields for a learner's or an account's identifiers, each AI provider's SDK is imported by its adapter alone, and tutor gives the platform no verdict: a program decides each one from what tutor reads. The check board serves only a loopback address, turns away a request whose Host is not local or whose Origin is another site, and writes the author's handwriting readable by its owner alone. A paid eval run starts only after its estimate and a typed yes, and starts no call that could take it past its dollar cap. The decisions behind these are in https://github.com/altnovatrail/tutor/blob/main/docs/decisions.md



    At least one of the project's primary developers MUST know of common kinds of errors that lead to vulnerabilities in this kind of software, as well as at least one method to counter or mitigate each of them. [know_common_errors]
    Examples (depending on the type of software) include SQL injection, OS injection, classic buffer overflow, cross-site scripting, missing authentication, and missing authorization. See the CWE/SANS top 25 or OWASP Top 10 for commonly used lists. Many books and courses are available to help you understand how to develop more secure software and discuss common implementation errors that lead to vulnerabilities. For example, the Secure Software Development Fundamentals course is a free set of three courses that explain how to develop more secure software (it's free if you audit it; for an extra fee you can earn a certificate to prove you learned the material).

    For this kind of software, a library and a command-line tool that send a learner's work to AI providers, with a local web page for the author, the maintainer counters: instructions hidden in the work for the model (on the platform a program, not the model, decides every verdict, the model's answer must follow a JSON schema and is decoded with unknown fields refused, and the runner checks every hint for the answer with the leak check); cross-site requests and DNS rebinding against the local board (it checks Host and Origin); writes in the wrong place (the board refuses any folder of the work tree outside samples/ and writes each sheet and picture as a new file, never over an older one); regular expressions that could backtrack on a long text (the leak check scans in loops); keys in the shell's history (the recipes ask for them without echo); script injection and untrusted checkouts in workflows (none, and Scorecard's Dangerous-Workflow check scores 10); and supply-chain tampering (actions and shared workflows pinned by commit, downloads checked against SHA-256 sums pinned in the Dockerfile, Go modules held by go.sum).


  • Use basic good cryptographic practices

    Note that some software does not need to use cryptographic mechanisms. If your project produces software that (1) includes, activates, or enables encryption functionality, and (2) might be released from the United States (US) to outside the US or to a non-US-citizen, you may be legally required to take a few extra steps. Typically this just involves sending an email. For more information, see the encryption section of Understanding Open Source Technology & US Export Controls.

    The software produced by the project MUST use, by default, only cryptographic protocols and algorithms that are publicly published and reviewed by experts (if cryptographic protocols and algorithms are used). [crypto_published]
    These cryptographic criteria do not always apply because some software has no need to directly use cryptographic capabilities.

    tutor implements no cryptography. It relies on published and reviewed algorithms and protocols: TLS, through Go's crypto/tls, for every call to the AI providers, which the official SDKs make over HTTPS (https://api.anthropic.com and https://api.openai.com by default); and SHA-256 (FIPS 180-4), which names the version of each prompt in a run's log, places each eval sheet in the split of a seed and names, by its digest, each bank of tasks whose open views a release carries. The Dockerfile of the devcontainer, which CI builds too, checks every archive it downloads against a SHA-256 sum pinned in the file.



    If the software produced by the project is an application or library, and its primary purpose is not to implement cryptography, then it SHOULD only call on software specifically designed to implement cryptographic functions; it SHOULD NOT re-implement its own. [crypto_call]

    It calls Go's standard library (crypto/sha256, crypto/rand and, through net/http, crypto/tls) and sha256sum, and never implements cryptography itself.



    All functionality in the software produced by the project that depends on cryptography MUST be implementable using FLOSS. [crypto_floss]

    Go's standard library is under BSD-3-Clause and sha256sum is part of GNU coreutils, and the SDKs that make the HTTPS calls are under MIT (anthropic-sdk-go) and Apache-2.0 (openai-go): all FLOSS.



    The security mechanisms within the software produced by the project MUST use default keylengths that at least meet the NIST minimum requirements through the year 2030 (as stated in 2012). It MUST be possible to configure the software so that smaller keylengths are completely disabled. [crypto_keylength]
    These minimum bitlengths are: symmetric key 112, factoring modulus 2048, discrete logarithm key 224, discrete logarithmic group 2048, elliptic curve 224, and hash 224 (password hashing is not covered by this bitlength, more information on password hashing can be found in the crypto_password_storage criterion). See https://www.keylength.com for a comparison of keylength recommendations from various organizations. The software MAY allow smaller keylengths in some configurations (ideally it would not, since this allows downgrade attacks, but shorter keylengths are sometimes necessary for interoperability).

    tutor has no keys of its own and sets no key length: its HTTPS connections take the defaults of Go's crypto/tls. An API key for eval is a token the provider issues, not a cryptographic key.



    The default security mechanisms within the software produced by the project MUST NOT depend on broken cryptographic algorithms (e.g., MD4, MD5, single DES, RC4, Dual_EC_DRBG), or use cipher modes that are inappropriate to the context, unless they are necessary to implement an interoperable protocol (where the protocol implemented is the most recent version of that standard broadly supported by the network ecosystem, that ecosystem requires the use of such an algorithm or mode, and that ecosystem does not offer any more secure alternative). The documentation MUST describe any relevant security risks and any known mitigations if these broken algorithms or modes are necessary for an interoperable protocol. [crypto_working]
    ECB mode is almost never appropriate because it reveals identical blocks within the ciphertext as demonstrated by the ECB penguin, and CTR mode is often inappropriate because it does not perform authentication and causes duplicates if the input state is repeated. In many cases it's best to choose a block cipher algorithm mode designed to combine secrecy and authentication, e.g., Galois/Counter Mode (GCM) and EAX. Projects MAY allow users to enable broken mechanisms (e.g., during configuration) where necessary for compatibility, but then users know they're doing it.

    No broken algorithm is used: tutor's code calls SHA-256 alone, never MD5, SHA-1, DES or RC4, and sets no TLS option of its own, so its HTTPS connections take Go's crypto/tls defaults.



    The default security mechanisms within the software produced by the project SHOULD NOT depend on cryptographic algorithms or modes with known serious weaknesses (e.g., the SHA-1 cryptographic hash algorithm or the CBC mode in SSH). [crypto_weaknesses]
    Concerns about CBC mode in SSH are discussed in CERT: SSH CBC vulnerability.

    No algorithm with known serious weaknesses is used: SHA-256 only, never SHA-1 or MD5, and no TLS configuration of tutor's own.



    The security mechanisms within the software produced by the project SHOULD implement perfect forward secrecy for key agreement protocols so a session key derived from a set of long-term keys cannot be compromised if one of the long-term keys is compromised in the future. [crypto_pfs]

    tutor implements no key agreement; its HTTPS connections are Go's crypto/tls.



    If the software produced by the project causes the storing of passwords for authentication of external users, the passwords MUST be stored as iterated hashes with a per-user salt by using a key stretching (iterated) algorithm (e.g., Argon2id, Bcrypt, Scrypt, or PBKDF2). See also OWASP Password Storage Cheat Sheet. [crypto_password_storage]
    This criterion applies only when the software is enforcing authentication of users using passwords for external users (aka inbound authentication), such as server-side web applications. It does not apply in cases where the software stores passwords for authenticating into other systems (aka outbound authentication, e.g., the software implements a client for some other system), since at least parts of that software must have often access to the unhashed password.

    tutor stores no passwords: it authenticates no users, and the board, which serves only this machine, has no sign-in. The eval API keys are credentials for the providers, which the criterion does not cover; they reach the process in environment variables, and the repository holds none.



    The security mechanisms within the software produced by the project MUST generate all cryptographic keys and nonces using a cryptographically secure random number generator, and MUST NOT do so using generators that are cryptographically insecure. [crypto_random]
    A cryptographically secure random number generator may be a hardware random number generator, or it may be a cryptographically secure pseudo-random number generator (CSPRNG) using an algorithm such as Hash_DRBG, HMAC_DRBG, CTR_DRBG, Yarrow, or Fortuna. Examples of calls to secure random number generators include Java's java.security.SecureRandom and JavaScript's window.crypto.getRandomValues. Examples of calls to insecure random number generators include Java's java.util.Random and JavaScript's Math.random.

    tutor generates no cryptographic keys or nonces; those of TLS are Go's crypto/tls. The board takes a random tail for a file name from crypto/rand, and math/rand/v2 only draws synthetic eval samples from a seed, so that the same seed draws the same sheets.


  • Secured delivery against man-in-the-middle (MITM) attacks


    The project MUST use a delivery mechanism that counters MITM attacks. Using https or ssh+scp is acceptable. [delivery_mitm]
    An even stronger mechanism is releasing the software with digitally signed packages, since that mitigates attacks on the distribution system, but this only works if the users can be confident that the public keys for signatures are correct and if the users will actually check the signature.

    tutor is delivered over HTTPS only: its repository, its releases and their files, and its Go module, which the go command fetches through the Go module proxy and checks against the checksum database, which holds each version to its bytes. Releases are immutable, GitHub attests each one with its files, which gh release verify and gh release verify-asset check, and each file is signed with cosign, its Sigstore bundle beside it in the release.



    A cryptographic hash (e.g., a sha1sum) MUST NOT be retrieved over http and used without checking for a cryptographic signature. [delivery_unsigned]
    These hashes can be modified in transit.

    No hash is retrieved over HTTP. Every SHA-256 sum of an archive the devcontainer downloads is pinned in .devcontainer/Dockerfile next to its version, and the curl that fetches them is restricted to HTTPS, redirects included (--proto =https --proto-redir =https). The go command checks the Go modules against go.sum and the checksum database, over HTTPS. A release's SHA256SUMS comes signed, with its Sigstore bundle, in the same release as the files it sums.


  • Publicly known vulnerabilities fixed


    There MUST be no unpatched vulnerabilities of medium or higher severity that have been publicly known for more than 60 days. [vulnerabilities_fixed_60_days]
    The vulnerability must be patched and released by the project itself (patches may be developed elsewhere). A vulnerability becomes publicly known (for this purpose) once it has a CVE with publicly released non-paywalled information (reported, for example, in the National Vulnerability Database) or when the project has been informed and the information has been released to the public (possibly by the project). A vulnerability is considered medium or higher severity if its Common Vulnerability Scoring System (CVSS) base qualitative score is medium or higher. In CVSS versions 2.0 through 3.1, this is equivalent to a CVSS score of 4.0 or higher. Projects may use the CVSS score as published in a widely-used vulnerability database (such as the National Vulnerability Database) using the most-recent version of CVSS reported in that database. Projects may instead calculate the severity themselves using the latest version of CVSS at the time of the vulnerability disclosure, if the calculation inputs are publicly revealed once the vulnerability is publicly known. Note: this means that users might be left vulnerable to all attackers worldwide for up to 60 days. This criterion is often much easier to meet than what Google recommends in Rebooting responsible disclosure, because Google recommends that the 60-day period start when the project is notified even if the report is not public. Also note that this badge criterion, like other criteria, applies to the individual project. Some projects are part of larger umbrella organizations or larger projects, possibly in multiple layers, and many projects feed their results to other organizations and projects as part of a potentially-complex supply chain. An individual project often cannot control the rest, but an individual project can work to release a vulnerability patch in a timely way. Therefore, we focus solely on the individual project's response time. Once a patch is available from the individual project, others can determine how to deal with the patch (e.g., they can update to the newer version or they can apply just the patch as a cherry-picked solution).

    No vulnerability is open. Dependabot has raised no alert for tutor, govulncheck, a required check, found no vulnerability in the code tutor reaches on 2026-11-01, and neither CodeQL nor SonarCloud has an open finding. Scorecard's Vulnerabilities check scores 10.



    Projects SHOULD fix all critical vulnerabilities rapidly after they are reported. [vulnerabilities_critical_fixed]

    Dependabot security updates are on, govulncheck runs on every pull request and on main as a required check, and a finding of SonarCloud or CodeQL is fixed in code before the next task starts.


  • Other security issues


    The public repositories MUST NOT leak a valid private credential (e.g., a working password or private key) that is intended to limit public access. [no_leaked_credentials]
    A project MAY leak "sample" credentials for testing and unimportant databases, as long as they are not intended to limit public access.

    gitleaks scans the whole history on every pull request and on main (Secrets / gitleaks, a required check), and GitHub's secret scanning and push protection are on. The workflows hold no secret, and the keys for eval stay outside the repository: a run gets them in its environment.


 Analysis 8/8 ●

  • Static code analysis


    At least one static code analysis tool (beyond compiler warnings and "safe" language modes) MUST be applied to any proposed major production release of the software before its release, if there is at least one FLOSS tool that implements this criterion in the selected language. [static_analysis]
    A static code analysis tool examines the software code (as source code, intermediate code, or executable) without executing it with specific inputs. For purposes of this criterion, compiler warnings and "safe" language modes do not count as static code analysis tools (these typically avoid deep analysis because speed is vital). Some static analysis tools focus on detecting generic defects, others focus on finding specific kinds of defects (such as vulnerabilities), and some do a combination. Examples of such static code analysis tools include cppcheck (C, C++), clang static analyzer (C, C++), SpotBugs (Java), FindBugs (Java) (including FindSecurityBugs), PMD (Java), Brakeman (Ruby on Rails), lintr (R), goodpractice (R), Coverity Quality Analyzer, SonarQube, Codacy, and HP Enterprise Fortify Static Code Analyzer. Larger lists of tools can be found in places such as the Wikipedia list of tools for static code analysis, OWASP information on static code analysis, NIST list of source code security analyzers, and Wheeler's list of static analysis tools. If there are no FLOSS static analysis tools available for the implementation language(s) used, you may select 'N/A'.

    CodeQL, with the security-extended queries, analyses the Go code and every workflow, and SonarCloud analyses the repository, the board's HTML pages and their scripts among them, on every pull request and on main. golangci-lint runs staticcheck, go vet and the other linters of just lint (Go / lint), and govulncheck checks the code against the Go vulnerability database (Go / vuln). All of them are required checks of main's ruleset, CodeQL for each of its languages.



    It is SUGGESTED that at least one of the static analysis tools used for the static_analysis criterion include rules or approaches to look for common vulnerabilities in the analyzed language or environment. [static_analysis_common_vulnerabilities]
    Static analysis tools that are specifically designed to look for common vulnerabilities are more likely to find them. That said, using any static tools will typically help find some problems, so we are suggesting but not requiring this for the 'passing' level badge.

    CodeQL's security-extended queries and SonarCloud's security rules look for common vulnerabilities, and govulncheck for the known vulnerabilities of the Go vulnerability database in the Go code and the standard library it reaches (Go / vuln, a required check). SonarCloud found, and the pull request fixed before it merged, a Go tool installed in the Dockerfile without a lock file (docker:S8545, https://github.com/altnovatrail/tutor/pull/13). On main it found an npm install that would have run the packages' install scripts (docker:S6505), fixed on 2026-10-20.



    All medium and higher severity exploitable vulnerabilities discovered with static code analysis MUST be fixed in a timely way after they are confirmed. [static_analysis_fixed]
    A vulnerability is considered medium or higher severity if its Common Vulnerability Scoring System (CVSS) base qualitative score is medium or higher. In CVSS versions 2.0 through 3.1, this is equivalent to a CVSS score of 4.0 or higher. Projects may use the CVSS score as published in a widely-used vulnerability database (such as the National Vulnerability Database) using the most-recent version of CVSS reported in that database. Projects may instead calculate the severity themselves using the latest version of CVSS at the time of the vulnerability disclosure, if the calculation inputs are publicly revealed once the vulnerability is publicly known. Note that criterion vulnerabilities_fixed_60_days requires that all such vulnerabilities be fixed within 60 days of being made public.

    The ruleset of main holds a merge on any CodeQL alert at error level or of high severity and above, and on SonarCloud's quality gate. CodeQL has raised no alert in tutor, and SonarCloud shows no open issue (2026-11-01): https://sonarcloud.io/project/issues?id=altnovatrail_tutor



    It is SUGGESTED that static source code analysis occur on every commit or at least daily. [static_analysis_often]

    On every pull request and on every commit to main.


  • Dynamic code analysis


    It is SUGGESTED that at least one dynamic analysis tool be applied to any proposed major production release of the software before its release. [dynamic_analysis]
    A dynamic analysis tool examines the software by executing it with specific inputs. For example, the project MAY use a fuzzing tool (e.g., American Fuzzy Lop) or a web application scanner (e.g., OWASP ZAP or w3af). In some cases the OSS-Fuzz project may be willing to apply fuzz testing to your project. For purposes of this criterion the dynamic analysis tool needs to vary the inputs in some way to look for various kinds of problems or be an automated test suite with at least 80% branch coverage. The Wikipedia page on dynamic analysis and the OWASP page on fuzzing identify some dynamic analysis tools. The analysis tool(s) MAY be focused on looking for security vulnerabilities, but this is not required.

    The automated test suite runs on every pull request and on main with Go's race detector (go test -race, the required check Go / test), which watches the code for data races as it runs: on every package whose code or tests start goroutines; the rest start none, but for the eval runner's parallel tests, which share nothing they write, and run without it, and a test fails once one of them starts one. Some tests vary their inputs: they run the generators of eval sheets over as many as 200 seeds and check every result against its rules. On 2026-11-01 the suite covered 95.1% of the statements; Go measures statements, not branches, and there is no fuzz test yet.



    It is SUGGESTED that if the software produced by the project includes software written using a memory-unsafe language (e.g., C or C++), then at least one dynamic tool (e.g., a fuzzer or web application scanner) be routinely used in combination with a mechanism to detect memory safety problems such as buffer overwrites. If the project does not produce software written in a memory-unsafe language, choose "not applicable" (N/A). [dynamic_analysis_unsafe]
    Examples of mechanisms to detect memory safety problems include Address Sanitizer (ASAN) (available in GCC and LLVM), Memory Sanitizer, and valgrind. Other potentially-used tools include thread sanitizer and undefined behavior sanitizer. Widespread assertions would also work.

    tutor is written in Go, a memory-safe language, without cgo or the unsafe package; the board's pages run their scripts in the browser.



    It is SUGGESTED that the project use a configuration for at least some dynamic analysis (such as testing or fuzzing) which enables many assertions. In many cases these assertions should not be enabled in production builds. [dynamic_analysis_enable_assertions]
    This criterion does not suggest enabling assertions during production; that is entirely up to the project and its users to decide. This criterion's focus is instead to improve fault detection during dynamic analysis before deployment. Enabling assertions in production use is completely different from enabling assertions during dynamic analysis (such as testing). In some cases enabling assertions in production use is extremely unwise (especially in high-integrity components). There are many arguments against enabling assertions in production, e.g., libraries should not crash callers, their presence may cause rejection by app stores, and/or activating an assertion in production may expose private data such as private keys. Beware that in many Linux distributions NDEBUG is not defined, so C/C++ assert() will by default be enabled for production in those environments. It may be important to use a different assertion mechanism or defining NDEBUG for production in those environments.

    Go has no assert statement. The tests check their expectations with the testing package's t.Error and t.Fatal and run with the race detector, which CI turns on for the tests of every package that starts goroutines. The code checks what it reads, the files of samples, banks and plans and the models' answers, decoded with unknown fields refused, and returns an error.



    All medium and higher severity exploitable vulnerabilities discovered with dynamic code analysis MUST be fixed in a timely way after they are confirmed. [dynamic_analysis_fixed]
    If you are not running dynamic code analysis and thus have not found any vulnerabilities in this way, choose "not applicable" (N/A). A vulnerability is considered medium or higher severity if its Common Vulnerability Scoring System (CVSS) base qualitative score is medium or higher. In CVSS versions 2.0 through 3.1, this is equivalent to a CVSS score of 4.0 or higher. Projects may use the CVSS score as published in a widely-used vulnerability database (such as the National Vulnerability Database) using the most-recent version of CVSS reported in that database. Projects may instead calculate the severity themselves using the latest version of CVSS at the time of the vulnerability disclosure, if the calculation inputs are publicly revealed once the vulnerability is publicly known.

    A failing test fails a required check, Go / test, and the pull request cannot merge until it passes. No vulnerability has been found this way.



You can use tools and AI systems to propose changes via a simple URL, such as https://www.bestpractices.dev/en/projects/16171/choose/edit?osps_ac_01_01_status=Met&osps_ac_01_01_justification=GitHub+enforced. See our automation proposals system for how to do that. This data is available under the Community Data License Agreement – Permissive, Version 2.0 (CDLA-Permissive-2.0). This means that a Data Recipient may share the Data, with or without modifications, so long as the Data Recipient makes available the text of this agreement with the shared Data. Please credit Aleksandr Riazanov and the OpenSSF Best Practices badge contributors.

Project badge entry owned by: Aleksandr Riazanov.
Entry created on 2026-11-01 19:00:10 UTC, last updated on 2026-11-16 18:35:25 UTC. Last achieved passing badge on 2026-11-01 21:03:47 UTC.