Governance, trust, and why the solver is private

Public position

Software can claim compatibility with almost anything. Physics is less accommodating. swmmrs is an independent implementation of EPA SWMM that puts fidelity first. It changes the implementation, ownership, packaging, and interfaces without quietly changing the model underneath them.

EPA SWMM is the behavioral reference. When the two engines disagree, assume that swmmrs is wrong. That assumption changes only when comparison finds an upstream version difference, a documented EPA issue, or an intentional departure. Non-EPA formulations must be explicit, opt-in, tested, and documented. They belong in clearly marked, opt-in branches of the timeline.

A new implementation does not inherit EPA SWMM's authority by borrowing its name. Compatibility claims must come from measured behavior. swmmrs preserves the model while changing how people build, isolate, and use it. This is not an attempt to put different hydraulics in a familiar uniform and hope nobody notices.

Why the solver source is private

I began swmmrs to learn Rust by translating a program I already knew and cared about. I still build it for myself first. The project is not a service owed to other developers, and its existence does not promise that the solver source belongs to the public.

Private source is a social boundary, not a force field. I do not have the resources to enforce a custom restrictive source license, and I will not pretend that a sternly worded file can patrol the galaxy. Requiring contact before access gives interested developers a reason to introduce themselves, exchange ideas, and contribute. It also discourages silent forks into parallel timelines with no contact between their inhabitants.

People who genuinely intend to contribute can get access to the solver repository. Contributions can be code, validation models, discrepancy investigation, benchmarks, review, documentation, public interfaces, or tooling. The point is communication and reciprocal work, not exclusivity.

The public repository contains the Python and command-line interfaces, output reader, documentation, packaging, installers, and release checks. Original public-repository material is licensed under Apache-2.0. Official wheels and command-line archives may be used, modified, and redistributed under the swmmrs Binary Runtime License.

Privacy proves correctness about as well as a sealed box with a blinking light. It also limits independent review of the implementation. That cost is real, and the project does not ask users to ignore it.

Trust model

Private implementation. Public receipts.

While the source is private, swmmrs has to make its case with evidence that anyone can inspect:

• reproducible regression comparisons against EPA SWMM.
• published benchmark workloads, engine versions, methods, hardware, timings, and result distances.
• documented differences from EPA SWMM, known upstream issues, and solver compatibility changes.
• focused tests and comparison artifacts for intentional behavior changes.
• versioned release artifacts, licenses, and checksums.
• public discussion of reproducible discrepancies and corrections.

These materials make scrutiny possible. They do not prove equivalence across every model, routing condition, platform, numerical edge case, or neighboring dimension. Compare critical production results with a trusted EPA SWMM release. Report unexplained differences with the input model, both engine versions, and the result artifacts.

There is no Jedi mind trick exemption for numerical software. You do not owe swmmrs trust because its maintainers wrote it. swmmrs owes you specific claims, reproducible evidence, visible limits, and swift corrections.

Stewarded development while private

While the solver source is private, lead maintainers control repository access, project scope, and releases. Access begins with a conversation about the work someone wants to pursue. There is no score, entrance exam, or sentient admissions computer.

Repository access is not a Jedi rank. Significant changes to numerical methods, model semantics, defaults, or compatibility expectations need reproducible evidence, focused validation, and documentation. Another qualified solver maintainer must review the change when one is available. The project will not describe a change as independently reviewed when nobody independently reviewed it.

No captain's chair can make a numerical claim correct. Reference behavior, reproducible models, benchmarks, published literature, and engineering reasoning settle technical disagreements. Seniority and repository access do not.

Request solver access

Useful work can begin on the public side of the airlock:

• report and reduce a behavioral difference from EPA SWMM.
• contribute a difficult validation model or benchmark workload.
• improve comparison methods or investigate a numerical edge case.
• improve public APIs, tooling, packaging, or documentation.
• participate constructively in technical review and design discussion.

A completed public contribution helps, but it is not a formal prerequisite for access. To start the conversation, email admin@swmm.rs with your GitHub username, relevant experience, and the work you want to pursue.

Do not include private credentials or solver source in public issues, pull requests, logs, or artifacts.

Path to Apache-2.0

The solver source is not meant to remain frozen in carbonite on principle. Privacy is the current collaboration model, not a promise of permanent closure. If interest in swmmrs grows enough, the source will move to Apache-2.0.

The decision will be guided by:

• solver repository access requests.
• GitHub stars.
• broader interest expressed through use, discussion, issues, validation work, and contributions.

No ceremonial number of stars opens the vault. Together, these signals show when public collaboration would help the project more than the current privacy boundary.