About implied.news
implied.news is an automated news page. On a schedule, it reads what the wires are covering, finds the prediction markets that price those same events, and publishes the pairing. No one edits it before it goes out. This page explains exactly which parts are a person, which parts are ordinary code, and which parts are a language model — so you can weigh what you read accordingly.
What it's for
You already have news sources. This is not trying to replace them, and it does not report anything of its own. What it adds is a number: for a given story, what price are people currently willing to pay on the outcome? That number is often more specific than the coverage around it, and it moves before the coverage does. When it disagrees with the tone of the reporting, that gap is the interesting part.
Because that's the whole value, the writing tries to stay out of the way. Headlines here are meant to be informative and short, not to compete for your attention.
What a person does
One person builds and maintains this. Their work is upstream of any given edition, not inside it:
- Writes the code, the prompts, and the editorial rules the software follows.
- Chooses the sources, the exchanges, and the standards — for example, the rule that sports results, celebrity news, and outrage bait are excluded.
- Sets the thresholds: how big a price move has to be to matter, how confident the market match must be before a market is shown under a story, how thin a market can be before it's dropped.
- Reads the site after it publishes, and changes the rules when it gets something wrong.
What a person does not do: approve an edition before you see it. There is no pre-publication review. Every headline, summary, story order, and market pairing on the front page went live without a human reading it first. Corrections happen after the fact, in the next cycle.
What ordinary code does
The numbers are not generated by a model. They are fetched, computed, and checked by deterministic software — the same input always produces the same output, and nothing is estimated or invented:
- Pulls live prices, volume, and 24-hour changes from the prediction-market exchanges. Every market shown names the exchange it came from.
- Maintains an index of markets with their price history, and computes moves, relative volume, liquidity, and the charts and probability curves you see.
- Searches that index by meaning rather than keyword alone, to find candidate markets for a story.
- Filters out markets that are settled, expired, unpriced, or too thin to mean anything.
- Deduplicates stories and alerts, and applies fixed significance thresholds before anything is promoted or sent.
If a price is on this page, it came from an exchange through this path. Prices are snapshots from the last build, not a live feed, so they can be several hours stale.
What a language model does
The judgment calls are made by a large language model. Specifically, the model:
- Reads the day's signals — search trends, mainstream wire and newspaper feeds, and a news search index — and groups them into distinct stories.
- Writes the headline and the summary you read. These are the model's words, not a reporter's and not the source outlet's. They are a compression of other people's reporting, and compression can lose or distort things.
- Decides which markets belong under which story. It has to clear a bar to be shown: when nothing does, the story runs with no market rather than a wrong one.
- Orders the page — what leads, what is grouped, what is left off.
- Shortens long market questions into the labels on the pills.
This means a headline here can be wrong in ways a person's headline usually isn't: it can misread a source, flatten a caveat, or state something more plainly than the reporting supports. Every story links to the outlet it came from. For anything that matters to you, read that.
Where the line falls
Roughly: anything that is a number is computed, and anything that is a judgment is generated. Fetching, pricing, charting, and filtering out dead markets are code — they do not involve a model and they do not improvise. Choosing what counts as a story, writing the words you read, deciding which market belongs under which headline, and ordering the page are all the model.
So the failure modes differ by part. A price being wrong here usually means it is stale. A headline being wrong means the model misjudged something — and that is the more likely of the two.
The page rebuilds on a fixed schedule, several hours apart, not continuously.
What this is not
- Not reporting. There are no reporters here. Nothing is independently verified. Every fact traces back to a linked outlet, and the accuracy of a story here is capped by the accuracy of that outlet plus whatever the model lost in summarising it.
- Not a forecast, and not advice. A market price is what traders are currently paying, which is a useful signal and a frequently wrong one. Nothing here is financial advice or a recommendation to trade.
- Not a neutral instrument. Prediction markets have real limits — thin volume, small and unrepresentative trader pools, ambiguous resolution criteria, and occasional manipulation. A confident-looking number can rest on very little money.
- Not reviewed. Worth repeating: nothing on the front page was read by a person before it was published.
Corrections and contact
If something here is wrong — a bad headline, a market that doesn't belong under a story, a stale price — it's worth telling us, because the fix usually goes into the rules and stops the whole class of error rather than just that one instance. Write to admin@implied.media.
Publisher
implied.news is published by Implied Media LLC, Sheridan, WY, United States. Contact: admin@implied.media.