In short
Every article starts from sources you can open. An AI model drafts the text from a fact sheet, programs check what it wrote, and Vynuo, the editorial lead, approves each text before it goes live.
The data on Radar pages works differently: it is refreshed automatically from official sources or public databases, named on each page, with checks, and is not approved one change at a time.
This page says what each part does and what it does not do. Nothing here is a promise that there will be no mistakes. If there are, the corrections policy says what happens.
Sources
Every fact in an article comes from a registered source, and the article lists those sources at the end, with the type of each. We use four tiers:
- Official primary source: a company announcement or blog, a changelog, the API of whoever owns the data, a verified official channel, a public body. It can support a fact on its own.
- A reliable database or aggregator that does not own the information. On its own it cannot support a fact: it helps us spot news and confirm it at the official source, or counts as confirmation when two independent ones agree.
- Another news outlet: a lead, never proof.
- Relevance signals, such as audience interest: they only help us decide what to cover and never support a fact.
We only fetch from sources the site has registered in advance, and we respect each one's robots.txt and terms of use. When a source blocks automated access, it is marked unavailable and we do not try to get around the block. We do not use leaked material or material obtained by improper means.
The confirmation rule
- A fact about a product (version, price, feature): one official source is enough. If a second independent source exists and the two disagree, the article does not run.
- A fact about the world (an acquisition, a delay, a cancellation): needs an official source or two independent unofficial ones. Two language versions of the same blog count as one.
- A future date in a changelog is a target, not an accomplished fact. An announcement date is not an availability date. Alpha, beta and release-candidate builds are filtered out.
- When sources contradict each other, the article is blocked until someone resolves it.
- Every time is stored in UTC and only converted for display.
- We do not translate titles ourselves: we use the title given by the publisher, distributor or platform and, when there is none, the international title. "Available in Brazil" appears only when a Brazilian source says so.
How we use artificial intelligence
Articles are drafted with the help of AI models from the Claude family, made by Anthropic. The model receives a fact sheet in which each fact carries the literal passage from the source, the source itself, and who is stating it. It does not read the source pages or the text of other outlets: only the sheet.
Each article is written in Portuguese and in English, in separate calls, and each version starts from the same facts. The second is not a translation of the first. A local subject, such as a price in Brazilian reais, may exist in Portuguese only.
What the AI does not do:
- It does not type quotations. Today our articles carry no verbatim quotes: what a source states appears as a paraphrase attributed to it.
- It does not translate the title of a work, game or product.
- It does not invent numbers, dates, prices, versions or proper names. Those values arrive typed from the sheet and are formatted by a program.
- It does not decide what gets published. It only proposes a draft.
- It does not pick addresses to visit or browse on its own.
- It does not take on a persona: the text belongs to the newsroom, not to an invented person.
Automated checks
Before a draft reaches Vynuo, it goes through checks run by programs and by an AI call separate from the one that wrote it:
- every number, date, price, version and name in the text must exist in the fact sheet;
- overlap with the source text is measured and capped, and the headline must differ from the source's headline;
- links in the text must respond;
- the Portuguese and English versions must carry the same numbers, dates, names and facts;
- a separate AI call, without seeing the draft, answers factual questions by reading only the sources, and a program compares those answers with the text; afterwards it hunts for flaws in the draft;
- sources are fetched again at approval and at publication: if any has changed, the article goes back to the queue.
These checks lower the risk of error but do not remove it. That is why no text goes out without the human approval described below.
Human approval
Every news text is approved by Vynuo in the admin panel before it is published. The system does not publish text on its own. The two language versions are approved separately: one can go out while the other waits.
The approval screen shows the sources and the fact sheet first, and only then the result of the checks. To unlock approval, the panel requires some of the article's facts to be opened at the source. The word we use is "approved", not "reviewed": the label "Approved by Vynuo" appears only when the record shows the approval took place.
The Radar and automatic data
Radar pages are data tools, not articles. Their data is refreshed automatically from official sources or public databases and goes through checks before it goes live:
- each page has a closed data format and plausible ranges: a price outside the range or an empty list is not published;
- big changes, such as a price that collapses or a catalog that loses half its titles, become drafts awaiting approval instead of going straight live;
- a change only counts once it shows up in two consecutive reads;
- if the source fails, the page keeps the last good data and shows since when.
Each page shows two dates: "Updated", which changes only when the value changes, and "Verified", the last time the source was looked at. The fixed text of each page is approved once per language.
Each relevant change also becomes a short line in the "Just changed" strip, written from a template, without AI.
Images
Images are official promotional material, taken from an official source: the company's website, its press kit, the product page in an official store, or the official channel. Every image carries a credit, in the form "Image: Press/company", and is only resized or cropped, with no text, filter or montage on top. When there is no official image, or when the data is the news, the site generates the cover itself.
We do not use images from other outlets, from photo agencies or stock libraries, from fan sites or from third-party social networks, nor from databases whose terms forbid it. We do not use AI-generated images that look like a photo of a real person, product or event.
If you hold rights to an image and ask for it to be taken down, the image comes down within 24 hours and the generated cover goes in.
Use of other outlets
Other outlets can serve as leads. When an article starts from a lead, we read only the outlet's title, address and date: its text is neither stored nor used. The article rests on the primary source, names the outlet that broke the news, with a link, in the first paragraphs, and carries a headline different from the original.
News that exists in only one outlet, with no confirmation from a primary source, is not published.
Labels and dates
- "Approved by Vynuo, editorial lead": the recorded human approval.
- "jkit.tools newsroom, with AI": the text was drafted with the help of artificial intelligence.
- Source type: official, data (database or aggregator) or the outlet that broke the news.
- Date and time of publication and of update. The update date changes only when the content changes in a relevant way.
- A dated update or correction note at the top of the article. See Corrections.
The site's ads and affiliate links are described in the privacy policy.
Questions and disagreements
If something on this page does not match what you see in an article, write to [email protected] or use the form.