I'm Justinas Čekanauskas, vegan, author and systems builder behind justinasbuilds.com, where I am writing my book What We Tolerate and building The Flourishing Framework. I advocate openly for animals, people, the Earth and moving away from animal products. Nobody funds or backs me, and I have no paid relationships connected to what I write.

Because I am open about what I stand for, I never ask anyone to just take my word for things. This page explains how I research, check, publish and fix mistakes across my articles, maps and investigations. Everything here goes out under my own name, and I take full responsibility for every word, every check and every correction. If I get something wrong, I want you to see the exact evidence I used and tell me straight away.

How my solo research lab and AI checks work

Every piece carries this note: "I write in my own voice and publish independently under my own name. I run my own solo research lab using custom scripts and AI verification checks to dig through data and cross-check public documents, list the sources behind every fact so anyone can check my work, and take responsibility for every word and correction."

What that looks like day to day is simple: my scripts and AI checks help me search and sort official documents, pull source sheets together, and recheck quotes and numbers against saved copies of the files. The writing is mine, I check the quotes I use, and my name is on every line.

We all know AI can get things wrong. That is why I log which checks were run by AI and which ones I checked with my own eyes, and I never pretend an AI check was my own reading.

What counts as proof

For me to state something as a fact, the actual passage has to be checked in context against a saved copy of the original document, with a clear note on who checked it (me, AI, or someone else). Whenever I quote a document, I save the exact words, official link, title, date, and page or paragraph number, along with when the file was saved, when it was checked, and its SHA-256 file fingerprint. That fingerprint just proves which exact file was on my computer when I worked on it. It does not magically make the claim inside the file true.

If I say a government asked for something, I need their own text or an official meeting record that puts that request in their name, along with the date, the office (when named), the reason they gave, and any opposing views in that same record. Also, proposing something only proves it was proposed. Showing it actually passed takes separate proof.

The same goes for saying words were cut out of a text. Two different reports saying different things do not prove anything was deleted. To say wording was removed, I need the saved earlier draft and the saved final version of the exact same document side by side. Anytime I cannot verify a source, quote or speaker, I either leave it out or tell you plainly what is still unverified. And when something is my own opinion or reading, I say clearly that it is mine.

Sources and access

I always start with official primary documents. News articles and NGO or advocacy reports are useful for finding leads, so when I use them I credit them clearly and never mix them up with official records. Five news stories repeating the same number does not mean five people independently checked it. Before I lean on a quote as evidence, the original document gets checked, and if that record also has facts that push against my point, I include those too instead of hiding them.

A big part of the pages and reports I use are opened and saved manually by me in the browser. When I do use scripts to fetch pages automatically, they check the site's robots.txt rules first and respect them. I never try to sneak around blocks, access refusals or sites that say no to automated fetching. If I cannot get a document openly and lawfully (or from a lawfully saved copy), I mark it honestly as a gap instead of guessing. And I link you straight to the original sources rather than reposting whole documents without permission.

How I make counts

When I publish a count, you should not have to trust my math. You should be able to rerun it yourself.

Every count covers a specific, named set of documents. Right next to the result, I share the full list of texts, their dates and versions, official links, saved-file SHA-256 fingerprints, why I chose that set, and anything excluded or missing. I also share the exact search words, the matching rules, and whether I counted individual word hits or full sentences, since those give different numbers.

I put out the counting script and clear steps showing how PDFs or web pages were turned into plain searchable text so anyone can repeat the exact same run. Every match and the text around it also gets read in context to catch things a raw search can miss, with a record of whether AI or I read each passage. No count goes up until its inputs and method can be checked, and if a file is missing or cannot be read, that gets marked as a gap, never counted as a zero.

One fair limit I always keep next to any count: a keyword search only shows what those words hit inside those chosen texts. It does not prove everything an office ever said anywhere else, or whether they talked about the same thing using different words. If I ever update the document list or the script later, I date the new version and explain what changed.

Right of reply

If I am going to publish findings about an office or organisation, fairness means asking them first. Before publishing, I write to them under my own name with the specific points, clear questions and document references. I let them know where I plan to publish, that both my questions and their answers are meant to be published, and the exact deadline (with date, time and time zone).

Normally, when a piece looks at an institution's record or positions, I give seven to ten working days to answer, plus fair extensions if they need a bit more time.

This week's questions: I am sending my questions on Monday 16 November 2026, publishing the questions themselves on Tuesday 17 November 2026, with a reply deadline of Thursday 19 November 2026 at 12:00 CST (Winnipeg time, 18:00 UTC). That is a shorter window than my usual seven to ten working days because COP31 is scheduled to close on Friday 20 November 2026, and these questions are about decisions happening while the summit is still open.

I keep a clear log of who I contacted, when I wrote, the deadline and whether they answered. When a reply comes back, I print it in full alongside my questions and put their main answer right next to the finding it covers. Personal contact details like private emails or phone numbers get removed and marked as redacted, while their actual answer is never cut down. Original emails stay on file privately.

If their answer shows I got something wrong, I fix my line. If the deadline passes with no answer, I just note when I wrote and that no reply had arrived by the deadline or publication time. Not replying does not mean they refused or admitted anything. And if an answer comes in late, I still add it with a dated update.

Corrections

Nobody gets everything right forever. If you think I made a mistake, please email me at j.cekanauskas@justinasbuilds.com with the link to the article or map, the exact part that looks wrong, and the source I should read. I aim to write back within three working days and fix confirmed mistakes within 24 hours once I confirm them. If checking takes longer, I mark that point as under review, or pull back any line I cannot stand behind while I dig into it.

When a fix changes the facts or meaning, I never quietly edit a page and pretend it was always like that. I update the article or map, add a clear dated note showing what it originally said, what changed, why, and the source behind the fix, and put it in the public log below. Small typos that do not change meaning get fixed without cluttering the table, and I only name the person who spotted a mistake if they tell me they want credit.

Corrections log

Log started: 16 November 2026, 00:00 CST (Winnipeg time). This table starts empty because the log is brand new, not as a claim about the correction history of any earlier post.

Date and time (Winnipeg time, CST)Article or map, and passageOriginal wordingCorrection and reasonSupporting source

What I do not claim

I do not pretend to read people's minds or guess hidden motives. If a document states why an office did something, I report that as their stated reason. If the document gives no reason, I simply say no reason is given there.

Timing, money, membership or being in the room do not by themselves prove somebody controlled a decision. Even if someone asks for a change and that change happens right after, that alone does not prove they caused it. Unless a document actually links the request to the decision, I say clearly that cause is not proven.

I do not make health or medical claims. When people appear in my work, it is only in their public, official roles as shown in the records, and I name the office rather than the person whenever a personal name is not needed. Private lives and personal details stay out. Finally, I never claim a piece has had legal review or outside independent review unless that actually happened.