News connected to markets — and someone accountable for the connection.

Who writes this

How a story becomes a score

Nothing here is a hand-picked opinion piece. Every story runs through the same pipeline, and every number on the site is the output of a step you can point at.

1. 01

   Ingest

   Market news is pulled continuously from wire services and publisher feeds. We never republish a publisher's article text — every story links back to the original, and what you read here is our analysis of it.

2. 02

   Extract the claims

   A story is broken into discrete, checkable claims rather than treated as one blob of sentiment. “A six-figure H-1B fee has been proposed” is a claim; the headline around it is not.

3. 03

   Score each claim for market impact

   Each claim is scored on how much it can plausibly move a price, and the reasoning behind the score is stored and shown alongside it. You see why a claim scored what it did, not just the number.

4. 04

   Map claims to tickers

   Scores are attached to companies through a typed relationship graph — suppliers, customers, competitors, subsidiaries, acquirers — so a story about one company reaches the others it actually touches, with the link type named rather than inferred from a keyword match.

5. 05

   Put it on the chart

   Each ticker page plots the scored catalysts on the price history, so a claim's score sits next to what the stock actually did next — including when the two disagree.

6. 06

   Track the theme over time

   Related stories accumulate into topic trackers so a developing story reads as one thread instead of forty disconnected headlines.

Editorial policy

• Sources are credited and linked. Every story names its publisher and links to the original. We analyse the news; we do not reprint it.
• Nobody pays for coverage. No company can pay to be scored, featured, or scored differently. There is no sponsored placement anywhere on the site.
• Predictions are sealed before the outcome exists. Forward predictions in research are hash-locked at the moment they are made. A prediction whose content no longer matches its lock was edited after the fact, and the integrity check surfaces it. This exists because our own earlier work failed retrospectively — the measurement was designed after the data, and three believable numbers were published before the bugs behind them were found.
• Mistakes get corrected in place. When a score, a mapping or a write-up is wrong, it is fixed and the change is noted rather than quietly removed.
• Automation is disclosed. Claim extraction, scoring and summarisation are produced by models running in the pipeline described above, reviewed by Kasper. Where a page is model-generated, it says so.