Every story shows where it came from.

• Companies' own channels. Publishers' and platform holders' newsrooms, blogs, press-release feeds, official Bluesky accounts and official YouTube channels, plus the Steam announcements of the 100 top-selling games. When a company says something itself, that's where we want to hear it.
• Outlets' headlines. 25 games outlets, read from their own public feeds: the headline and the short summary they publish there. We never open their articles. Outlets that publish on YouTube, like Digital Foundry, count as the same outlet there: titles only, linked.
• The business press. Fortune's and Bloomberg's public story listings, because big industry news often breaks there first. Headlines and times only.
• Reporters' scoop posts. A short list of games reporters on Bluesky. A post counts only when the reporter marks it as a scoop ("BREAKING:", "NEW:"); a reporter sharing their own article isn't a new source.
• Discovery. New and changed listings on Steam and itch.io, for First Light.
• Steam's own numbers. Once a day: every game's price, review counts for the games we watch (top sellers, new releases, our picks) and their Steam Deck rating; hourly, how many people are on Steam.
• Emulation, homebrew and preservation. Official GitHub releases from a list of emulators, game ports, homebrew tools and engines; new retro homebrew listed on PDRoms; and the Video Game History Foundation's posts. Official project releases only; no ROMs, ever, and no emulation of current Nintendo hardware.
• Public records. State WARN layoff notices; SEC filings; National Labor Relations Board filings at game companies; DMCA notices game companies file with GitHub, from GitHub's public record.
• Press releases sent to press@theraytrace.com, from companies that have confirmed they send them. Live showcases such as Nintendo Directs Proposed.

Before we read anything from a site the first time, we read its robots.txt (the file where a site says what automated readers may do) and we check again every week. If a site says no, we don't read it; we ask for permission instead. Some outlets tell AI crawlers no while allowing ordinary readers. From those we keep only what code needs (the headline, the time) and their words never reach an AI model.

Some companies post statements only on X, which we don't read automatically. When one matters, the editor supplies a screenshot and the post's direct link. We transcribe it word for word, keep the screenshot, and link to the post.

Code groups the day's headlines into stories, and an AI model checks the busiest groups for anything that doesn't belong (2.9). Then, for each story, the desk looks for the origin, in this order:

1. A company's own post that came before the coverage. If the company announced it, the company is the origin.
2. A scoop that came before every other outlet. A business-press story or a reporter's marked scoop post, published first.
3. The outlet the others credit. When at least two follow-ups say "according to Fortune", Fortune broke it.
4. A company statement we don't have yet. If follow-ups say the company spoke but we don't hold its words, we wait for them.
5. Otherwise, unknown. We don't write a story whose origin we can't find. Its earliest headline goes to the Rays instead.

We also look for the company's own confirmation by searching the full text of its recent posts, because the confirming line is often buried in a longer letter. Every sentence we quote from it is checked word for word.

A story keeps a dossier all day: every source it has ever had, with times and roles, even after a source drops out of the latest batch. Each time new evidence arrives, the tick-tock is worked out again from everything. For bigger stories, an AI model reads the whole dossier as a desk editor and can ask the editor for things we can't fetch, like a post on X or a stream. It can't change the tick-tock: that's decided by code, from the sources.

A story is written only when we know its origin and at least two outlets have covered it. An AI model writes it (which one) from the sources we may use: the company's own words, the origin's headline, statements the editor supplied. Then code checks the draft before it can run:

• Every quotation appears word for word in a source.
• Every number appears in a source.
• Every reported claim names the outlet that reported it, and the origin is named up top.
• Follow-up outlets stay out of the story; they belong in "How others are covering it".
• No words that imply anyone played the game ("hands-on", "we tried"), and no hype ("insane", "slams").

A draft that fails goes back to the model with the problems listed. After three failures the story is dropped.

Every finished story is then held to the standard: a story the editor approved as the model for all the others, the BioWare lead of October 21, 2026. Code checks that the origin has a link and a time, the tick-tock runs in time order, nothing is called "original" if coverage came before it, and outlets that added details are credited. For now a miss is flagged for the editor; once the site publishes on its own, a story that misses will be held Coming.

1. The editor's call. When the editor says a story is the big one, it leads, for the rest of the day.
2. News judgment. Otherwise an AI model (2.9) ranks our own written stories against a written standard: how much the news changes what people play or who makes it, and for how long; the stature of the studio, franchise or person; surprise. It sees our headlines and counts, never other outlets' text.
3. Coverage, if the ranking fails. Coverage is evidence of importance, not the measure of it: a studio closing outranks a widely covered patch.

Stories are rewritten only when their facts change: the origin, who added what, a confirmation, a statement. When a story changes after it's been written, it says what changed and when Coming to the page. Every story keeps its full list of sources so a reader can check us.

Corrections. We strive for accuracy, but we and our AI models can make mistakes. If you see something that should be fixed, please email corrections@theraytrace.com.

Once a day, discovery reads Steam's newest releases (from the Steam store's own new-releases list) and itch.io's newest and staff-featured games. Then:

1. Code filters. Software that isn't a game, listings with no description, and Steam's adult sexual-content category are left out.
2. A fast AI model scores. It rates each remaining listing from a compact fact card. Whether a developer disclosed AI use is kept from the model so it can't sway the score; we show that disclosure to you as the listing states it.
3. A second model writes briefs. One for each top scorer, at most 30 a day. Every quote in a brief is checked against the listing.
4. Five run. The top five appear as First Light, described from the store page. "Why it might matter" is labeled as the model's inference.

No sponsor, advertiser or affiliate relationship is ever an input to which games are picked, how they're ranked or how they're described. A brief whose quotes can't all be found in the listing is never shown.

• Fetch or copy other outlets' articles. We use their feed headlines, linked, as they wrote them.
• Ignore a site's robots.txt or terms, or get around bot protection, logins or paywalls.
• Invent a quote, a fact, or a statement from a developer or company.
• Say or suggest that anyone played a game they didn't.
• Present other outlets' repetition as confirmation.
• Write a story whose origin we couldn't find.
• Let money decide coverage: no paid placement, and nothing commercial affects what we pick, rank or say (how we're funded).

The Raytrace is free to read, and it doesn't make money yet. It is entirely self-funded by founder and editor Nicholas De Leon. These are the ways it may make money, and the rules for each:

• Reader support. One-time tips and $4-a-month supporters, through Stripe: the buttons at the bottom of the front page and in the newsletter.
• Sponsors. We're open to sponsorship in any reasonable form. A sponsor is always labeled clearly as one.
• Affiliate links, if we ever use them, are labeled where they appear.
• Consulting. The editor is available for consulting: how The Raytrace is built, or a version of it made for your newsroom or company. Write to editor@theraytrace.com.

Money never decides coverage. Sponsors have no say in what we cover, how we rank it or how it's worded. That's enforced in our code: the parts that find, group, rank and write stories, and that pick First Light's games, can't see who sponsors us, and a test fails if they ever could. Any story that names a current sponsor says so on the story.

The Raytrace's news stories are written by AI. Each one is drafted by an AI model from the primary sources listed with it, and automated checks confirm every quote, number and attribution against those sources before it runs. No person edits each story before publication. The editor, Nicholas De Leon, sets the standards the system follows, decides what leads the front page, and supplies statements we can't collect ourselves.

Every model here is Anthropic's Claude, called through Anthropic's API with The Raytrace's own key. Each call is logged with the model that answered, its tokens and its cost. As of October 22, 2026:

First Light uses Claude Haiku 5.5 to score new store listings and Claude Sonnet 5.5 to write the short briefs, from the store listing only. If Sonnet declines a request, Anthropic's API may answer with another Claude model; the log records which one.

What no model does here: no images are generated (pictures are the companies' own or store art, credited); no model reads text from outlets whose terms or robots.txt bar AI use; and no model decides anything for money. When the models change, this list changes the same day.

The Raytrace is made by Nicholas De Leon in Tucson, Arizona, and is modeled after Tucson Daily Brief, his AI-assisted local news site. He can be reached by email at editor@theraytrace.com or on LinkedIn.

Note: Please send all general pitches, press releases, etc. to press@theraytrace.com. All correspondence ultimately goes to the same inbox, but this helps keep things organized. Thank you.

Follow The Raytrace on Bluesky at @theraytrace.com. You can also message us on Discord: add the username theraytrace.