Repository navigation
Description
What
A Comments section at the bottom of every dataset detail page, mirroring the issues and their replies from that dataset's repository (nemarDatasets/<id>).
Read-only. The website never writes to GitHub. Instead the section carries a note telling people where to comment:
Comments are mirrored from this dataset's GitHub repository. To add one, open an issue or reply to an existing one — it will appear here.
Labelled "Comments", not "Issues": for a data user, "issue" reads as "something is broken with this dataset", which suppresses exactly the ordinary questions and usage notes this is meant to collect. The note makes it plain that comments are public and live on GitHub, so nobody is surprised.
Why read-only
Dropping the write path removes the two hardest problems at once, at the cost of one extra click for the commenter:
- No attribution decision. A comment posted on GitHub is authored by its author, full stop. No GitHub OAuth bolted onto ORCID, and no NEMAR bot relaying comments under its own identity with moderation and deletion routed back through us.
- No abuse surface. There is no authenticated NEMAR path that writes to a public repository, so no new rate limiting, no policy for what happens to a suspended user's comments, and no impersonation risk.
Moderation, editing, and deletion stay where the content lives, which is also where curators and depositors already work.
Why push, not fetch-on-render
The mirror is populated by GitHub pushing to NEMAR when an issue or comment changes, not by NEMAR polling or proxying. See nemarOrg/nemar-cli#1043.
This matters because fetch-on-render does not survive production. Unauthenticated GitHub is 60 requests/hour per source IP; server-rendered from the Cloudflare Worker that IP is Cloudflare's and shared across an enormous number of tenants, so the budget is permanently gone rather than occasionally. Moving the call into the browser makes it 60/hour per visitor IP, which fails for anyone behind institutional NAT — much of the intended audience.
With a push-populated mirror, rendering a dataset page reads NEMAR's own store. GitHub is not in the request path at all, so the page stays edge-cacheable and there is no rate limit to reason about.
Frontend scope
- Section renders from
GET /datasets/:id/comments. - Degrades gracefully: no comments, endpoint unavailable, or a dataset whose repo has never had an issue should all render something sensible rather than an error — per the existing null-safety conventions for backend-shaped data.
- Comment bodies are user-authored Markdown from a public source. They must go through the existing
src/lib/markdown.tssubset and be escaped — never injected raw. This is the first place the site renders text written by arbitrary third parties, so it is the first place an XSS would land. githubRepoSlugis already derived insrc/pages/dataset/[id].astro, andActionBar.astroalready links<repo>/issues. That link and this section should agree rather than tell two different stories.- Threads: issues with their replies nested, newest activity first. Closed issues probably still worth showing, visually distinguished — a resolved question is often the useful one.
Blocked by nemarOrg/nemar-cli#1043. Supersedes the /community placeholder removed in #222.
Activity
neuromechanist commentedon Jul 31, 2026
Backend dependency filed:
nemarOrg/nemar-cli#1043. That issue also carries the attribution-model decision (GitHub OAuth vs a relaying bot), which should be settled there before either side builds.Context: this supersedes the
/communityplaceholder, removed in #222 / PR #224.[-]Dataset page: Comments section mirroring the dataset repository's GitHub issues[/-][+]Dataset page: read-only Comments section mirroring the dataset repo's GitHub issues[/+]