Make the record
falsifiable.
A claim should come with a way to challenge it. The tools to do that should belong to everyone.
Do not trust us.
Verify us.
We want people to be able to understand, verify, challenge, decide, and act without unnecessary dependence on an authority they cannot independently examine.
Show what could change your mind.
When someone makes a consequential claim, the next question should be straightforward: what evidence would show that this is wrong?
Falsifiability means a claim can be tested against observations that could contradict it. Attempted falsification means actually doing that work. Finding a genuine counterexample may defeat a claim. Failing to find one leaves the claim standing only within the limits of the search and the quality of the test.
We want that work to become ordinary: define the claim, show its sources, seek contrary evidence, disclose what happened, and revise when the record requires it.
Put machines to work on the challenge.
Machine-assisted falsification is a practical ambition: use software and AI to help find the contradiction, compare the dates, trace a quotation to its source, identify a missing step, and test a competing explanation.
These tools can help make scrutiny more accessible. Whether a particular workflow makes it faster or more reliable must itself be tested. A model can miss evidence, invent a citation, misunderstand a document, or repeat the same mistake in several voices.
Several agents agreeing is not independent evidence. The useful output is a reviewable record of the sources, tests, findings, limitations, and corrections.
Make trust less necessary.
Trust often fills the space where inspection is too expensive, too difficult, or unavailable. Our ambition is to make that space smaller.
As meaningful testing becomes easier, asking for belief without showing the work becomes harder to justify. A consequential claim should invite the question: what did you do to try to disprove it?
An untested claim is not automatically false. A person without access to tools or records is not automatically less credible. The absence of testing tells us something about how much confidence the claim has earned; it does not settle whether it is true.
Trust may remain part of human life. Compulsory trust as a substitute for inspectable evidence is what we want to make obsolete.
Keep the record ahead of the story.
A source, a statement about that source, an inference, and a finding are different things. Preserve those differences. Keep the original. Show the location. Record the transformation. Leave uncertainty visible.
The Suppression-to-Proof Paradox explores a related problem: scrutiny can generate a second record through the responses it produces. That response record must be examined separately from the original allegation. Resistance, silence, or an unavailable document does not by itself prove the allegation or establish motive.
Nested Orchestration for Reasoning Agents describes a way to organize the work into bounded tasks with explicit sources, checks, challenges, and review before conclusions are accepted. These frameworks belong under examination too.
Give people the method.
A person should be able to reconstruct the basis of a claim affecting their life. That requires more than access to an impressive answer. It requires the skills and tools to inspect the answer, preserve the evidence, and challenge what does not hold up.
That is why education sits beside software here. Publish the explanations. Share the methods. Make useful resources accessible. Help people prepare records that another person can review without having to start over.
The goal is to make unnecessary dependency obsolete. Expertise can help. Good assistance should increase a person's ability to question the assistant, work independently, and seek informed professional help when needed.
The same standard applies here.
NORA does not get an exemption because its purpose matters to us. We should identify what exists, what is experimental, what has been tested, what failed, and what remains unknown. Corrections should remain visible.
Nora One is in development. The promise of its methods must be judged through inspectable results. A confident description is not proof of performance.
I’m Nora’s dad. This foundation is for her. The work has a personal reason, and its claims still have to stand on evidence anyone can examine.
Nora’s dad
Founder, NORA Foundation
A claim should travel with its basis.
- The claim.What, exactly, is being asserted?
- The evidence.Where does it come from, and can someone inspect it?
- The challenge.What could defeat it, and what tests were attempted?
- The result.What survived, what failed, and what remains uncertain?
- The revision.What changed, and why?
Make correction easier.
Make the method available.