News feeds need provenance, not more summaries

I do not want a news product that sounds confident.

I want one that tells me what it knows, where it got it, how fresh it is, and where the evidence gets thin. That is a less glamorous design goal than "summarize the world for me," but it is the difference between a useful news interface and a very smooth rumor machine.

That is why I keep coming back to the boring fields in Asha.News. The public feed does not only return a headline and summary. It returns source, canonical URL, category, source publication time, first-seen time, last-checked time, freshness status, bias labels where available, archive status, and whether an item is marked breaking. The public digest groups stories into clusters and keeps the source trail visible instead of pretending the summary is the primary artifact.

Sources: Asha.News public feed, Asha.News public digest, BBC RSS item surfaced in the feed, CBS News item surfaced in the feed.

The summary is the least trustworthy part

A summary is convenient. It is also where the product has the most room to quietly change the story.

A feed item with only a headline and paragraph asks the reader to trust the system's compression. A feed item with source timestamps and canonical URLs lets the reader check the compression. That second design is less pretty. It is also more honest.

In the Asha.News feed I checked today, each item carried source_published_at, asha_first_seen_at, and asha_last_checked_at. Those fields look like implementation detail until a story is moving fast. Then they become the interface. They tell you whether the system saw a new event minutes ago, whether the source itself claims an older timestamp, and whether the archive checked it again.

That matters because freshness is not one thing. A story can be newly discovered by the product but old at the source. It can be updated by the publisher without changing the original RSS date. It can be breaking for one audience and stale for another. A single "new" badge flattens all of that into theater.

News UX should show its homework

Most news products make provenance feel like plumbing. The link is somewhere. The source name is there if you squint. The timestamp is visible until the template decides relative time is cuter.

For an evidence product, provenance is not plumbing. It is content.

The source URL tells me where to verify. The canonical URL helps deduplicate syndication noise. The archive status tells me whether the product captured enough of the source to preserve what it saw. The category and bias fields are not a verdict from God, but they tell me how the system is classifying the item. That gives me something to challenge.

The product design lesson is simple: do not hide uncertainty in the name of readability. Put the uncertainty where the user can use it.

Asha.News is strongest when it behaves like a research surface, not a newspaper homepage. The homepage can still be readable. The feed can still be fast. But the underlying object should stay inspectable. If the product says a story is fresh, I want to see the timestamps. If it says a cluster has two articles, I want the two links. If it summarizes a claim, I want the primary source sitting close enough that checking it feels normal.

The digest is a map, not the destination

The digest endpoint is useful for a different reason. It clusters stories and gives a quick read on what is being covered. That is great for scanning.

But the digest should not become the final answer. A cluster like "Supreme Court temporarily allows Trump’s ballroom construction to continue" is a map of coverage. The actual claims still need source links, dates, and a path back to the underlying articles.

This is where a lot of AI news products go wrong. They treat the digest as the product because it feels magical. The model reads, compresses, and hands you a neat paragraph. Done.

Not done. The neat paragraph is the riskiest object in the chain. The product earns trust by making it easy to leave the paragraph and inspect the trail.

What I would steal for any AI research tool

The same pattern applies outside news.

If you are building an AI research tool, a social listening dashboard, or an agent that briefs a team every morning, make the evidence fields first-class. Store the original URL. Store the canonical URL. Store when your system saw it. Store when the source claims it was published. Store the last time you checked it. Store the extraction state. Store the reason something was rejected.

Then expose enough of that state in the interface that a human can make a judgment without opening the database.

The interface does not need to show every field all the time. It does need to make the trail reachable. A collapsed "evidence" drawer beats a polished summary with no receipts. A boring timestamp beats a cute freshness badge. A source list beats an invented confidence score.

This is not only about media literacy. It is product safety. When a system summarizes fast-moving events, the failure mode is not a typo. The failure mode is misplaced certainty.

Asha.News points in the right direction because it treats provenance as part of the product. More news tools should copy that. Less magic. More receipts.