Evidence and Verification: What Actually Supports a Claim?

Why this matters now

Getting an answer has become easy.

Knowing what deserves trust has not.

Search engines can return thousands of results.

AI can generate polished explanations instantly.

Charts can be created from incomplete data.

Statistics can be copied without their original context.

This changes the central learning problem.

It is no longer only:

Can you find information?

It is also:

Can you tell what that information actually allows you to claim?

Real-Life Readiness argues that in an environment of abundant information and machine-generated responses, knowledge must help people distinguish evidence from plausibility, recognise missing context and judge when a response is reliable enough to become action.

Start with the claim

Verification becomes difficult when the claim is vague.

“This solution is sustainable.”

Could mean:

it uses less energy;

it can be maintained;

it produces lower environmental impact;

it remains affordable;

it can scale.

Each claim requires different evidence.

A good verification process therefore begins by making the statement precise enough to examine.

Claim, source and evidence are different things

CLAIM

What we are asserting.

SOURCE

Where the information comes from.

EVIDENCE

The relevant information that actually supports, limits or contradicts the claim.

A source can be real.

A citation can be accurate.

And the claim can still be unsupported.

Citation is not verification

A reference provides traceability.

Verification requires another step.

Open it.

Read it.

Ask:

Does it actually say this?

Under what conditions?

What population or system was studied?

Is the result observational or causal?

What limitations are stated?

Are we extending the conclusion beyond the evidence?

Citations make verification possible.

They do not perform it for us.

The distance between evidence and claim

One useful way to think about evidence quality is the distance between:

what was actually observed

and

what we are claiming.

Evidence:

“Twenty participants completed the task during the test.”

Claim:

“The interface is intuitive for everyone.”

The claim goes far beyond the observation.

A more defensible statement remains closer to what the evidence can support.

Provenance matters

Where did the information originate?

Who produced it?

When?

Using what method?

Was it original data?

A translation?

A summary?

A repost?

An AI synthesis?

Information can travel through several layers before reaching us.

Original study
→ news article
→ social post
→ AI answer.

At every layer, context can disappear.

Verification often means moving back towards the origin.

Primary does not mean perfect

A primary source is valuable because it is close to the original event, data or decision.

But it can still have interests, limitations or incomplete methodology.

A company is a primary source for what it says its product does.

That is not necessarily independent evidence that the product outperforms alternatives.

Mars Challenge documentation is primary evidence of programme design.

It should not automatically be treated as independent causal evidence of impact.

That distinction protects credibility.

A correct answer can still have weak evidence

A learner may state something true and cite an irrelevant source.

The conclusion happens to be correct.

The reasoning chain is still weak.

Evidence literacy therefore concerns not only whether the final answer is right.

It concerns whether another person can examine why it should be believed.

Context determines relevance

Evidence produced in one context may not automatically travel.

Different:

populations;

environments;

times;

scales;

definitions;

conditions.

A result can be rigorous and still be irrelevant to another claim.

Verification therefore includes transfer:

Does this evidence apply here?

Observation is not explanation

If Y happens after X, we know something about sequence.

We do not automatically know that X caused Y.

Alternative explanations may exist.

This distinction is especially important for impact claims.

A positive participant story can demonstrate that an experience occurred.

It does not automatically demonstrate that the programme caused all subsequent outcomes.

A case is not a prevalence estimate

A case can show possibility.

It can illuminate mechanism.

It can reveal detail.

It does not automatically tell us:

how common the outcome is;

whether it is typical;

whether it generalises.

This is why Case / Field Insight and Research / Evidence should remain distinct content types.

Evidence should be able to challenge us

Weak verification looks for confirmation.

Stronger verification asks:

What could show that our claim is wrong or incomplete?

Suppose a team believes:

“Users prefer this solution.”

Instead of only asking:

“Do you like it?”

they can observe:

Do people choose it when alternatives exist?

Can they use it without support?

Where do they struggle?

Do behaviour and stated preference agree?

The test should allow reality to disagree with the hypothesis.

Corroboration requires independence

Ten webpages repeating the same original report are not ten independent confirmations.

Corroboration asks whether other evidence, ideally independently produced, supports or challenges the claim.

Quantity matters less than independence and relevance.

Different questions require different evidence

If the question is:

How did this participant experience the activity?

The participant's account is highly relevant evidence.

If the question is:

How many participants completed the task?

We need records or observation.

If the question is:

Did the activity cause improved performance?

The evidentiary demand becomes much higher.

There is no single evidence hierarchy that solves every question.

Evidence quality is always tied to what we are trying to know.

Verification in an AI-rich environment

AI and Human Judgment addresses the wider responsibility of working with AI.

Evidence and Verification focuses on the claims inside the output.

An AI response may combine:

facts;

interpretations;

inferences;

errors;

missing context;

invented references;

overconfident language.

A useful process is:

identify material claims;

locate original sources;

open them;

check whether they support the claims;

look for missing context;

corroborate where needed;

mark what remains uncertain.

Real-Life Readiness specifically warns that a fluent answer may appear complete before its evidence is complete.

Verification should be proportional to consequence

Not every statement requires the same depth of checking.

A reversible, low-consequence decision may tolerate lighter verification.

A claim affecting:

safety;

health;

rights;

large resources;

reputation;

irreversible action

should normally require stronger verification.

This creates a direct bridge to Decision-Making Under Uncertainty.

Evidence asks:

How strongly is this supported?

Decision-making asks:

Is that support sufficient for what we are about to do?

A prototype can produce evidence

Mars Challenge's participant guidance frames prototyping around a concrete question:

Can this work?

That matters.

A prototype is not only communication.

It can be a test.

Build.

Observe.

Learn.

The artefact becomes part of an evidence process.

But prototypes have boundaries

A test may establish that:

a mechanism moves;

a workflow can be understood;

a component fits;

a user can complete one task.

It does not automatically establish:

safety;

long-term durability;

economic viability;

sustainability;

scalability.

Good evidence includes a statement of what remains unverified.

Process can also be evidence

Mars Challenge asks teams to preserve sketches, discarded versions and prototype evolution because these artefacts reveal the real work of the team.

A final artefact tells us what exists now.

Process evidence can show:

what changed;

what was tested;

what failed;

which feedback mattered;

why a decision was revised.

This is particularly valuable for evidence of learning.

Evidence of learning should be visible in behaviour

Real-Life Readiness proposes that evidence of readiness should privilege behaviour and artefacts over self-description alone.

Relevant evidence might include whether a person:

distinguishes evidence from assumptions;

identifies missing knowledge;

evaluates external or AI-generated claims;

documents trade-offs;

revises after feedback.

Instead of asking:

“Are you a critical thinker?”

we can examine what the person actually did with a questionable claim.

Confidence should be calibrated

Verification does not always end in:

true

or

false.

Useful conclusions include:

strongly supported;

reasonably supported;

limited evidence;

inconclusive;

contradicted.

Calibrating confidence means matching the strength of the language to the strength of the evidence.

The Evidence Verification Loop

STATE THE CLAIM

What exactly are we asserting?

TRACE THE SOURCE

Where did it originate?

CHECK SUPPORT

What does the source actually show?

CHECK CONTEXT

Under what conditions?

EXPOSE ASSUMPTIONS

What are we adding to the evidence?

CORROBORATE OR TEST

What other evidence can support or challenge it?

IDENTIFY LIMITS

What has not been demonstrated?

CALIBRATE CONFIDENCE

How strongly can we state the conclusion?

MARK THE OPEN QUESTION

What still needs verification?

The full movement is:

CLAIM → SUPPORT → TEST → LIMITS → CONFIDENCE

The Verification Test

Before using an important claim:

What exactly is the claim?

What is the original source?

Does the source actually say this?

What kind of evidence does it contain?

Is it relevant to this context?

What assumptions are we adding?

Is there independent corroboration?

Is there contradictory evidence?

What limitations remain?

What level of confidence is justified?

What Evidence and Verification is not

It is not collecting citations.

It is not trusting prestigious sources automatically.

It is not rejecting lived or local knowledge because it is not published academically.

It is not treating every source as equally useful for every question.

It is not demanding perfect certainty.

It is not only about fact-checking AI.

It is not turning every learning activity into formal academic research.

It is not hiding uncertainty.

And it is not allowing evidence to make the decision for us.

Evidence informs judgement.

Humans remain responsible for what becomes action.

What the evidence says

Real-Life Readiness explicitly defines epistemic judgement as the ability to distinguish evidence, assumptions and plausible claims, identify what needs verification and calibrate confidence.

It also identifies plausible certainty as a risk when people or technologies produce answers that appear complete before their underlying evidence is complete.

Its proposed evidence architecture privileges behaviours and artefacts such as identifying missing knowledge, evaluating external or AI-generated claims and revising decisions after feedback.

Mars Challenge's participant process also provides first-party evidence of a design built around questions, prototypes and documented iteration.

These sources support the conceptual framework.

They do not establish a causal effect on participants' evidence literacy.

What this means for educators

Teaching verification can go beyond:

“Use reliable sources.”

Give learners a claim and ask:

What exactly is being stated?

What source supports it?

Does the source actually support it?

What assumption has been added?

What evidence would contradict it?

What is still unknown?

How strongly should the conclusion be written?

The objective is not scepticism for its own sake.

It is disciplined confidence.

Related concepts

• Decision-making under uncertainty
• AI and human judgment
• Real-Life Readiness
• Experiential learning
• Agency
• Reflection and feedback

Projects and evidence in practice

Mars Challenge Participant Process

Prototype work is framed around testable questions and documentation of sketches, versions and feedback.

Evidence status: first-party evidence of programme process; not causal evidence of learner outcomes.

Frequently asked questions

Is a citation the same as evidence?

No. A citation tells you where information came from. Verification asks whether that source actually supports the specific claim being made and under what limitations.

What is the difference between evidence and an assumption?

Evidence has relevant observable or documented support. An assumption is something temporarily accepted without sufficient verification. Assumptions are not automatically bad; hidden assumptions are the larger problem.

How many sources are enough?

There is no universal number. A simple factual claim may be well supported by one appropriate primary source. A complex, controversial or high-consequence claim may require multiple independent sources or different forms of evidence.

How should I verify an AI answer?

Identify its material claims, locate original sources, open and inspect those sources, confirm that they support the claims, check context and date, and seek independent corroboration where the consequences justify it.

Does a prototype prove that a solution works?

It can provide evidence about the specific function or question tested. It does not automatically validate the entire solution, its safety, sustainability, adoption or scalability.

What does calibrated confidence mean?

It means expressing a conclusion with strength proportional to the evidence. Some claims can be strongly supported; others may remain tentative, limited or inconclusive.

Why preserve failed versions?

Because they make reasoning visible. They can show which assumptions were tested, what failed, what feedback changed the project and why the final version is different.

References

Real-Life Readiness — A Foundational Framework
Adelino Sousa / Mars Challenge / Virtual Educa, 2026.
Epistemic judgement, evidence and assumptions, plausible certainty, verification and confidence calibration.

Mars Challenge Participant Guide 2026
Mars Challenge, 2026.
Prototyping as a testable question and documentation of the development process.

Explore next

Next step

Deepen: Explore Real-Life Readiness, AI and Human Judgment and Decision-Making Under Uncertainty.
Explore the process: Examine how investigation, prototypes and documented iteration make evidence visible in Mars Challenge.
Act: Apply the Verification Test to an important claim before turning it into a decision.