How every story on the MRSIndex is read
Open news data matches a company wherever its name appears. Name-matching alone over-counts badly: in this edition more than half of the coverage that survived the automated rules was not about the company it had been filed under. This standard is the written test every story is read against before it can count.
Where the read sits in the method
Open news data, 90 days, from a list of 657 credible outlets.
Automated rules remove press releases, market wraps, stock pages, pagination, owned media and known name collisions.
Every remaining story is read against this standard and labelled A, B or C.
Only A and B stories are scored, from a machine reading of the full article text.
A company is published only when at least 100 placements in credible outlets survive all four steps. That floor is never lowered to fit a company in.
The test
One question decides every label: would that company’s own communications team count this article as coverage of their company? The reader judges from the headline, the web address and the outlet.
The tie-breaks are strict by design. Unsure between A and B: choose B. Unsure between B and C: choose C unless the company’s own team would clearly clip it.
Brands and products count as the company
A story about a company’s own brand or product is about the company, even when the company name is absent: Instagram for Meta, Chevrolet for General Motors, Burger King for Restaurant Brands International, Halifax for Lloyds Banking Group, Barbie for Mattel. A company’s own short name counts the same way it does for its peers: “Pepsi” for PepsiCo, “Coke” for Coca-Cola. A different company that shares a name does not: Volvo Trucks is not Volvo Cars, Yum China is not Yum Brands, Lloyd’s of London is not Lloyds Bank.
What is never coverage
Namesakes and venues
- People, places, streets and teams that share the name (Doug Ford; Cadillac, Michigan; Chase Stokes)
- Events at a venue or race named after a sponsor (a match at State Farm Arena, the result of the Coke Zero Sugar 400); the sponsorship deal itself does count
Other companies’ stories
- A bank’s analysts commenting on another company’s stock
- Another company hiring a former executive
- A supplier’s or customer’s earnings
Market noise
- Market wraps, index moves, “stocks making the biggest moves”, stock-price and chart pages
- Earnings-season previews, options ideas, newsletter roundups of unrelated items
Shopping and lifestyle
- Sales, deals, coupons, voucher tips, price comparisons and “best X to buy” pages, even when every item is from one retailer
- “Went viral on TikTok”, creator and trend stories, celebrity style
- Film, TV, game or music plot, cast and review content (a studio’s own business decisions and box office do count)
Incidents the company did not cause
- An outsider’s crime, crash or fire that merely happened at a store, in a car or near an office
- Anniversaries of historical events, unless the company is doing something about it now
Unreadable pages
- Section, tag, author and homepage URLs, and headlines that cannot be read
- Paid or sponsored content
A higher bar for what is shown
Stories shown on a company’s card as evidence of its reputation are read a second time, independently, at evidence grade: any story that would not stand up as an example of the company’s coverage is removed from scoring, not just from display. In this edition that bar took two candidate companies back under the floor once their shopping, creator and trend pages came out, and each was replaced by the next qualifying company.
How the read is checked
- Who reads. Stories are read by AI agents against this written standard. The full MRS® adds human analysts.
- Second pass. Every story shown as evidence is re-read independently.
- Blind audit. A random sample of 400 story-company pairs was judged again blind; the two reads agreed on keep or drop 91% of the time.
- Open to inspection. Every scored story is in the downloadable file with its outlet, codes and pillar scores, so any single decision can be checked.
What this standard does not do
The read decides which stories count. It does not re-score them: sentiment, drivers and prominence come from a machine reading of the full article text, and are not confirmed story by story. That is the line between the public signal and the full MRS®, where analysts read and validate coverage directly.