A Nalarasa Lab productdecide better
A decision tool for founders, built by Nalarasa Lab.
It asks the questions. You do the thinking.
You are probably here because you saw de.Think somewhere and wanted to know who is behind it. Fair. This page answers that.
What calibration means
If you say you are 80 percent sure about ten things, about eight of them should turn out true. Most people say 80 percent and get five or six. The distance between how sure you feel and how often you are right is your calibration gap. It shows up in every forecast you make about your own business, and almost nobody measures it.
The confidence slider runs from 50 percent to 100 percent, not from zero. Each question has two answers, so 50 percent means you are guessing. Anything below that would mean you actually believe the other answer, in which case you would pick that one instead.
Who builds this
de.Think is made by Nalarasa Lab, the product studio of PT Nalarasa Intervensi Digital, an Indonesian company.
The name comes from two Indonesian words. Nalar means reason, logic, the way you think. Rasa means feeling, intuition, empathy. Not opposites. Both are in every decision a person makes.
We build behavioural products that start from existing research rather than assumption. de.Think is one of them. Runtut, a reflection tool for Indonesian users, is another.
We are small and still early. We would rather say that plainly than pretend to be bigger than we are.
Why de.Think exists
Almost every founder has started a decision journal or tool at some point. A Notion page, a spreadsheet, a notes file. Almost every one of them stopped after a few entries.
The reason is not laziness. An empty journal never answers back. You write down a decision, you close the file, and nothing happens. The payoff, if it ever comes, arrives months later when you remember to look.
de.Think answers on the very first entry. It shows you what your own reasoning skipped, and asks about it, before you commit to anything. The prediction gets checked later, once there is an outcome to check.
That is the whole point. You get something back on day one, not in six months.
How it works
You bring a real decision. Something with a cost attached.
You frame it
What are you actually deciding, and how hard would it be to undo? A choice you can reverse next week gets a short path. A choice that takes a year to unwind gets the full one.
You dig out your assumptions
What has to be true for this to be the right call? How confident are you in each one, and is that confidence built on data, on experience, or on a guess?
You pick a lens
Pre-Mortem, First Principles, Inversion, or Second-Order Thinking. Each one opens a different set of questions.
You compare the options properly
Side by side, against criteria you weighted yourself, including the option of doing nothing.
You commit to a number
Not "probably fine". An actual percentage, a definition of what success looks like, and a date to check back.
Then, months later, it checks
What actually happened, how far off you were, and whether the assumption you leaned on held up.
Over enough decisions, patterns appear. Where your confidence runs ahead of your accuracy. Where you keep choosing the status quo even when your own scoring says otherwise.
Full detail is at de-think.app.
You do not have to wait months to learn something
Most decision tools only pay off in the future. de.Think starts paying off while you are still sitting with the decision.
As you work through it, de.Think watches the process itself, not only what you wrote. The assumption you left blank. The question you could not answer. The criterion you skipped past. Those gaps get flagged while you can still do something about them, because a question you cannot answer is usually the weakest link in the whole decision.
Where your reasoning thinned out, which parts you avoided, and how solid the ground under this decision actually is.
Whether the assumption you leaned on held up, and how far your confidence sat from what happened.
The patterns that only appear across a body of decisions. Not advice about you, just your own record, read back.
Where AI fits, and where it does not
We are not against AI. We are against handing over the thinking. Those are different positions, and de.Think is built on the second one.
It works fully without AI, and that is the default. Business decisions carry things you would rather not send anywhere. Pricing you have not published, a hire you have not made, a pivot nobody knows about yet. So nothing leaves your device unless you decide it should.
The AI layer is optional, and it runs on your own key. You choose the provider, you choose the model, and you choose whether to switch it on at all. Your text goes to the provider you trust, never through our servers. Turn it off and nothing in the product breaks.
What it adds when you do switch it on. Without it, the questions are the same well built questions for everyone. With it, the questions follow what you actually wrote, so the follow ups get sharper and more specific to your situation. The patterns across your decisions get read in context instead of arriving as numbers you have to interpret alone.
What it never does. It only ever asks. It does not answer, score, recommend, or decide. There is no AI at all in the steps where you compare options, make the call, or reflect on the outcome. That part stays yours.
The research behind the questions
Every framework in de.Think traces to a primary source. A few of them:
Reversibility
One way door versus two way door, from Jeff Bezos's 2015 shareholder letter. This is what decides whether you get the short path or the full one.
Assumptions mapping
From Bland and Osterwalder, Testing Business Ideas (Wiley, 2019). The idea of finding the assumption that is both load bearing and least evidenced.
Pre-mortem
Klein, G. (2007), Harvard Business Review 85(9). Imagining the failure before it happens, which Kahneman called his favourite debiasing technique.
Calibration
Fischhoff, Slovic and Lichtenstein (1977) for the academic foundation, and Tetlock and Gardner's Superforecasting (2015) for the practical case that precise numbers beat vague words.
Locus of control
Rotter, J. (1966), Psychological Monographs 80(1). This is what separates "I misjudged this" from "nobody could have seen that coming".
Decision quality
Spetzler, Winter and Meyer (2016). Decision quality is judged by the weakest link in the chain, not by how the outcome happened to land.
The full citation list, including the cognitive biases and their sources, is in our research document. Coming soon.
What we do not claim
It does not hand you a verdict on your biases
de.Think surfaces bias two ways. Through questions placed at the point where a particular bias tends to bite, and through patterns in your own data once there is enough of it. What it will not do is announce which bias you have. A tool that tells you that is guessing.
It is not clinically tested
de.Think is built on established frameworks from decision science and cognitive psychology, and it has been through testing with real users before launch. What it has not been through is a clinical trial. Those are different things, and we will not blur them.
It is not coaching
It never gives advice, recommendations, or an assessment of you. It asks, you answer.
A short test is not a diagnosis
That applies to the free calibration test. It is a quick check on one specific thing, not a measurement of who you are.
Your data
Local first. What you write stays on your device, and no account is needed to work that way. That is the default, not a setting you have to find.
If you want your decisions available across more than one machine, a synced option will be available on a different tier. It is a choice you make deliberately, never something switched on for you.
If you use the optional AI layer with your own key, your text goes to the AI provider you chose, not to us. That is stated on the screen where you add the key.
You can export everything as Markdown, JSON, or PDF whenever you want.
Pricing
Pricing will be announced at launch, set at a level that matches the quality of the work rather than at whatever undercuts the market.
There will be a version built for group use, on its own tier. The core mechanic is never locked behind a tier. What separates them is how you are allowed to use it, not how much of the thinking you get.
Where things stand
Being specific about this matters more than looking finished.
The free calibration test at de-think.app. Ten questions, three minutes, no account, and you see your score without handing over an email.
Built and verified
The full decision flow, both the short and the long path, the history and patterns view, the guided tour, and the licence system.
Verified means something specific here. The licence check is cryptographically signed, and we tested it by deliberately tampering with it to confirm the tampering gets caught. Colour contrast was computed rather than eyeballed, which is how we found that our own secondary text colour failed accessibility standards in dark mode at 2.99 to 1, and replaced it. The interface was checked by measuring where elements actually sit on screen, not by looking at screenshots, which is how we caught a layout collision that only appeared on narrow phones.
Built, and being revised
Parts of the flow are getting reworked before launch. The bias library is moving from a reference list to questions that appear at the point where they matter, because a tool that tells you which bias you have is guessing. Several steps are being sharpened for the kind of decision that carries real cost.
Decided and designed, not yet coded
The process report that answers on your first entry, the optional AI layer, encrypted sync between devices, and group mode. Each has a written specification and a settled design. None of them exist in the product yet, and we will not describe them as if they do.
Launch
We have not set a public date. When there is one, the calibration test list hears first.
Want to know when it launches?
Leave an email and we will tell you once, on the day. Nothing else.
One message, at launch. No newsletter, no drip sequence.
FAQ
Is this another AI wrapper?
How is this different from a Notion template or a spreadsheet?
What happens to my data if the company disappears?
Do I need to be technical to use this?
Does it work offline?
Will it remind me when it is time to check a decision?
Can I use it with my team?
Why is the free test only about business trivia?
Who do I contact?
Ten questions, three minutes, no account.
See how close your confidence runs to your accuracy.
Try the free calibration testAlso from Nalarasa Lab
Runtut
A reflection tool for decision making, in Indonesian, for a general audience.