Glean 拾遗
Editorial standards

What “human review” means

What “every link human-reviewed” actually involves · what AI may touch · what it can’t.

1 · What gets in

✓ Accepts

  • Engineering content with a new claim or new data
  • Author is the first-hand implementer
  • Official changelogs / design docs / RFCs
  • Postmortems & retrospective deep-dives
  • First-hand non-English sources (CN / JP / DE …)

✗ Rejects

  • Listicles (“10 best …”)
  • Medium re-blogs / content farms
  • AI-summarized AI-summary posts
  • Articles 80% behind a paywall
  • Obvious affiliate / SEO link farms

2 · What AI is allowed to do

DeepSeek does exactly three things: extracts the body, translates both ways, drafts bullet points. It does not select picks, write editor’s notes, or order the weekly. Its drafts are all editable in our admin panel — on average 60% of cards get at least one human edit before publish.

AI · YES

Translation draft

First-pass EN↔ZH. A human checks term & tone.

AI · NO

Editor’s note

“Why read this” is always human-written. AI never takes a stance for us.

AI · NO

Selection

AI may score candidates as a reference. Pick / reject is always a human click.

AI · YES

Tag suggest

Picks 1–3 from existing taxonomy. Editor may replace all.

3 · Conflicts of interest

If an editor or contributor has an employment / advisory / investment relationship with the author or company of a pick, the card carries an editor’s disclosure line at the bottom. We take zero paid promotion — and we’d announce it on /about if we ever did.

4 · Corrections & takedowns

Spot something wrong? Reach out via the site footer. Technical mis-statements fixed within 24h with a footnote; factual errors trigger a corrigendum.

5 · What I won’t pretend to do

This isn’t full-stack tech media. Coverage: systems · tooling · design. Hardware, academic papers, startup news — out of scope.