Tools for thinking straight.
Untilt builds small, free tools that catch the errors your reasoning hides from you, in the decisions you make and in the arguments you read.
- 61 cognitive biases
- 153 Crossref-checked references
- No account, no cookies, no profile
The tools
For decisions
Doxa
Walk a decision through the cognitive biases most likely to be skewing it, on your own or with an AI-guided pass. Backed by a searchable bias library.
You get: the biases most likely at play in your decision, each with a question to test it and a technique to counter it.
For arguments
Elenchus
Paste any text and see how its reasoning holds up: logical flaws, a reasoning score, and what it does well. It does not check facts.
You get: a reasoning score (strong, moderate or weak), each flaw quoted from your text, and the argument’s strengths, if it has any.
Made by one person, William Revah. What each tool sends, and where
A look inside each tool
From the bias library
Availability Bias
Overweighting easily recalled information
A question to test it
Have I validated team priorities against objective data — metrics, user research, or revenue impact — rather than relying on what comes to mind most easily?
Read the full entrySaved run · 2026-10-10 · nvidia/nemotron-3-super-120b-a12b:free · prompt v2 · AI-generated
Analysed text
Why remote work is destroying our industry
Read the analysed text
I have now managed remote teams at three companies, and the pattern is unmistakable. In every case, the engineers who worked from home shipped less code, attended fewer design discussions, and were the first to leave when things got difficult. Remote work simply does not function for software teams.
Consider the evidence. Yahoo ended its remote work programme in 2013. IBM recalled thousands of remote staff to offices in 2017. Best Buy abandoned its results-only work environment. Three of the most recognisable names in technology looked at the data available to them and reached the same conclusion. That is not a coincidence; that is a verdict.
The studies claiming productivity gains deserve little weight. Most of them are produced by researchers who themselves work remotely, or funded by the very companies selling video conferencing software. It is hardly surprising that people whose comfort depends on an answer manage to find it.
The most vocal advocate of remote work in our own organisation is a junior developer who has never managed anyone, has been with the company for eight months, and by his own admission has not been in the office since March. It is difficult to take seriously a theory of team management advanced by someone with no experience of managing a team.
None of this is to say people should never work from home. An occasional day is fine. But the idea that a serious engineering organisation can be run without a shared physical space is, on the evidence, wishful thinking.
Summary
The argument relies on personal anecdote, selectively chosen corporate examples, and ad hominem dismissals of contrary evidence, making its overall reasoning weak.
Reasoning score: weak
Elenchus checks reasoning, not facts: a tight argument on false premises scores well here.
Flaws (4)
Anecdote as proof · significant
“I have now managed remote teams at three companies, and the pattern is unmistakable. In every case, the engineers who worked from home shipped less code, attended fewer design discussions, and were the first to leave when things got difficult.”
The author uses personal experience from a limited number of companies as proof for a universal claim about remote work, which does not establish a general pattern.
Cherry-picking · significant
“Consider the evidence. Yahoo ended its remote work programme in 2013. IBM recalled thousands of remote staff to offices in 2017. Best Buy abandoned its results-only work environment. Three of the most recognisable names in technology looked at the data available to them and reached the same conclusion.”
The author selects only companies that discontinued remote work while ignoring many firms that have successfully operated remotely, biasing the evidence.
Ad hominem · significant
“The studies claiming productivity gains deserve little weight. Most of them are produced by researchers who themselves work remotely, or funded by the very companies selling video conferencing software. It is hardly surprising that people whose comfort depends on an answer manage to find it.”
The argument attacks the researchers' motives and funding sources instead of addressing the substance of their studies, dismissing evidence based on perceived bias.
Ad hominem · significant
“The most vocal advocate of remote work in our own organisation is a junior developer who has never managed anyone, has been with the company for eight months, and by his own admission has not been in the office since March. It is difficult to take seriously a theory of team management advanced by someone with no experience of managing a team.”
The author rejects the advocate's position by focusing on their lack of managerial experience rather than engaging with any arguments they might present.
Strengths
- Acknowledges that occasional remote work can be acceptable, showing a qualified rather than absolute stance.
Questions
- Is it free?
- Yes. No account, no payment, no advertising. The Elenchus extension can also run on your own API key, which your provider bills.
- Does Elenchus send my text to an AI?
- Only when you press Analyze. A small proxy forwards it to a language model through OpenRouter and stores none of it. On the free tier the model provider may log it and train on it, so do not paste confidential text.
- What is kept about me?
- No account and no cookie. Your theme and language stay in your browser. To rate-limit requests, a record tied to your network address is kept and deleted the next day: a one-way hash for Doxa, the address itself for Elenchus. On Elenchus web, Cloudflare Turnstile receives your IP address to check for bots. On untilt.app, page views are counted without cookies; Elenchus counts none.
- How far can I trust the score?
- Elenchus checks whether conclusions follow from what came before, not whether the facts are true: a tight argument built on false premises scores well. The score comes from a general-purpose language model and a prompt, so read it as a second opinion.