AI-assisted, community-corrected news for internet culture.
Signals become structured first drafts.
Sourced, Sketchy, or Cap with reasons.
Corrections and receipts create history.
Signal selection. We monitor public platforms for culture-relevant signals. A signal needs at least two independent sources (or a source independence ratio ≥ 2) to become a draft. Weak-only social sources are rejected unless they overlap with known entities.
AI pipeline. A language model writes the first pass — headline, native tone, what happened, why it matters, and a confidence score. The model does not publish. It drafts. Every draft starts as unverified until checked.
Community vetting. Readers vote Sourced ✓, Sketchy 👀, or Cap ✕. A story with >60% Cap and ≥10 votes is hidden from feeds. Corrections and receipts go into a review queue; nothing goes live without moderator approval.
Auto-publish gate. High-confidence drafts (≥0.82 score, confirmed label, culture-relevant, diverse sources) in eligible categories can auto-publish. Everything else waits for human review.
Adversarial review. Before a confirmed story publishes, a second model pass challenges the draft. If the challenger finds contradictions or weak sourcing, the story is held for review.
Every story begins as a signal detected by our pipeline, then shaped by reader votes and moderator review into something you can trust.
Every story should show linked source evidence.
Key facts can be sourced, sketchy, or cap.
Approved changes create a visible version trail.
Community fixes do not go live until reviewed.
Mark it Sourced, Sketchy, or Cap.
Point at the exact headline, claim, or context issue.
Bring a better source into the review queue.
Big changes should leave a readable version trail.
Readers should be able to tell whether a fix came from new receipts, moderator review, or community pressure.
Unresolved arguments should stay visible instead of being papered over.
The model writes the first pass, not the final truth.
Source evidence should outrank virality, tone, or momentum.
Community fixes stay in review until a moderator approves them.
Readers push weak or misleading framing into review instead of letting it sit.
New evidence should change the story, not just the comments around it.
The live version should improve only after moderator sign-off.
Standard and Native should preserve the same underlying claims.
The product should help people understand internet-native stories without already living inside the discourse.
Language adaptation should make stories easier to follow, not blur what the receipts actually say.