Skip to content

Collaborate

This site is public on purpose. K-Fish's calibration work, architecture decisions, and research questions all benefit from second opinions. Whether you're a researcher, a developer, a trader, or a friend who's curious what I've been building — there are concrete ways you can help.

Three-lane model

Best for: first-time contributors, questions, opinions, "did you consider X?", pointers to papers you've read.

Where: github.com/ksk5429/quant-notes/discussions

Format: casual — just write the thought. If it turns into a thread, it might get promoted to an issue or incorporated into the site.

Great first topics if you're unsure

  • Anything on the Open Questions page
  • "I don't follow this step" — confusion is useful signal
  • "I read paper X and think it contradicts Y" — bring the citation
  • "My intuition on risk here is different" — dissent is welcome

Best for: "page Z contains factual error W", "this link is broken", "please write a review of topic T."

Where: github.com/ksk5429/quant-notes/issues/new/choose

Pre-built templates exist for:

  • 📝 Correction / typo / broken link
  • ❓ Review question or new-topic request
  • 🐛 Site bug (rendering, navigation, build)

Issues get triaged into the site's backlog. Large asks may become a PR from me; smaller corrections get a fast merge.

Best for: anyone comfortable editing Markdown who wants to actually make the change.

Easiest path: click the "Edit this page" ✏️ pencil at the top-right of any page on ksk5429.github.io/quant-notes. GitHub handles the fork-and-PR flow in your browser.

For larger edits: 

git clone https://github.com/ksk5429/quant-notes
cd quant-notes
pip install -r requirements.txt
mkdocs serve      # hot-reload at http://127.0.0.1:8000

See CONTRIBUTING.md for the full PR checklist.

What you can actually help with

Concrete ways that move the project forward:

  • Push back on an ADR. Each architecture decision record has clear reasoning. If you think it's wrong, the comments on the open-questions page are where that lives — and a single well-argued dissent can flip a design.
  • Reproduce a reported result. Brier parity claims exact match on three legacy runs. If you can run python scripts/brier_parity.py --all and see the same numbers, that's an independent verification.
  • Nominate a paper. If you've read something in LLM calibration / prediction-market research / Bayesian risk modeling that we should engage with, drop it in Discussions.
  • Spot contradictions between the two canonical blueprints and the current build log. They've drifted a little; keeping them honest matters.
  • Suggest clearer phrasing. Half the value of this site is expository. If something didn't land for you, it won't land for the next reader either.

What is and isn't in scope

In scope

  • Architecture, methodology, calibration math
  • Publicly-available code in polymarket-oracle-risk
  • Literature pointers, prior-art corrections
  • Translations (especially 한국어 versions of key pages)
  • Review of ADRs, runbooks, and published reports
  • Site UX (navigation, rendering, accessibility)

Out of scope

  • Access to the private kfish repo. Trading alpha stays private — see ADR-0001. Everything needed for genuine architecture discussion is public.
  • Trading tips / market calls / paid promotion. This is not a trading community. The bot's live trading is blocked pending legal review anyway.
  • Requests for legal advice about running your own bot. See the legal-brief summary — you need your own Korean counsel, not a fork of ours.

Review-credit

Substantive contributors get credit:

  • One good review that changes a design → acknowledged in the build log entry that captures the change.
  • Multiple reviews / sustained engagement → listed in the AUTHORS file (created as soon as there's a second name to add).
  • Port / translation work → author credit on the translated page via the page's front-matter authors: field.

A note for "friend-level" collaboration

If I've shared this site with you specifically — thanks for reading. The most useful thing you can do is comment aggressively on parts you think are wrong. I'd rather have five hard challenges on the calibration math than twenty polite "this looks nice" comments. Direct > polite.

If you're unsure where to start, just open a Discussion thread with whatever's on your mind. I'll engage.

What if I want to chat privately first

Email ksk54299@gmail.com or DM @ksk5429 on GitHub. Fine to scope-out an idea before making it public.