Skip to content

Contributing

Ryan Bayne edited this page Oct 31, 2026 · 2 revisions

Contributing

Status: active Category: E (product / user) Created: 2026-10-31 Last updated: 2026-10-31 Owner: TCS team

The code is AGPL-3.0: use it, run it, read it, fork it, sell a service built on it — and if you run a modified version as a network service, publish your changes. That is deliberate. Improvements come back to everyone instead of being resold as a closed product.

⚠️ One person maintains this. Reviews and replies may take time; that is the honest trade for a codebase this size. A clear, reproducible report is worth more than a large speculative pull request.


Where things go

You have Put it in
A question, or "how do I…" Discussions → Q&A
An idea, or a "should it…" Discussions → Ideas
A reproducible defect Issues — with the command, the output and what you expected
A fix A pull request, small and scoped

What a useful defect report contains

The project's whole culture is evidence over assertion, and it applies to reports as much as to code.

  • The command or click path, verbatim.
  • What you saw — the actual output or number, not a summary of it.
  • What you expected, and why.
  • Whether it is active or merely possible. "The UI lets me pick an option the backend ignores" is reachable; "25 of 26 records are configured to use it" is active. The second is a different priority and a different fix.

Numbers make a report fixable: "the score was 71, the threshold was 75, and it still dispatched" is a bug report. "It behaves oddly" is not.

Before you change code

npm run build                       # frontend: fast, always safe
npm --prefix nest run typecheck     # backend types, no Prisma generate
npm --prefix nest run build:api     # backend: stop → build → restart → health

⚠️ Never npm --prefix nest run build while the API is up — it fails with an EPERM naming a temp file and never mentions the running process. Use build:api.

⚠️ Never prisma db push. See Troubleshooting.

Conventions that will surprise you

Convention Why it exists
Comment blocks are long and explain why The reasoning is deliberately at the code site — the next person editing that file will not open an issue
Numbers in documentation are generated A number typed into prose is a second copy, and the second copy rots. If you need a count, it belongs in docs/GENERATED-FACTS.md
"Cannot judge" is preferred to a confident answer An empty list and a failed query must never render the same
Guards ship with a self-test that can fail A check that has never been observed to fail is decoration. --self-test on any validator should prove it can refuse
Changing a scoring weight or threshold needs a copied configuration Live maths is compared side by side, not edited in place — otherwise there is nothing to measure against
State what you did not do A partial change reported as complete is the one thing worse than a partial change

Areas that need help

The open issues are the honest list. The ones that would change the project most:

  • Packaging — an installer or container so the app can be run without a developer (#808)
  • Live-account safety — the real-account gate and risk acknowledgement (#1153)
  • Data coverage — more providers verified, more symbols with a confirmed route
  • Documentation — a step that confused you is a documentation bug, and you are the only person who can still see it

Licence of contributions

By contributing you agree your contribution is licensed under the AGPL-3.0, the same as the project.

Clone this wiki locally