A green dashboard is an argument, not evidence

A green dashboard looks like a readout. The colour is a conclusion, and the tile does not show whether anything was observed to reach it.

A readout reports something that was measured. An argument asks you to accept a conclusion. A green tile looks like the first: a status, a colour, a question apparently settled. What it actually is depends on what sits behind it. Sometimes a system was checked. Sometimes a person decided the control was in place, and the dashboard displayed that decision in the same colour it would use for a measurement. The tile does not tell you which.

An assertion and a piece of evidence are different things

The two are easy to conflate, and a dashboard that shows both in one colour conflates them for you.

An assertion is a statement about the world: this control is operating. It costs nothing to make, and it can be made in good faith and still be wrong. A piece of evidence is an observation that would look different if the assertion were untrue: the log that would be empty, the configuration that would read otherwise, the ticket that would still be open. Evidence can fail to support the assertion, and that is what makes it worth having. A claim nobody can check is not weak evidence. It is not evidence.

Financial auditing writes this distinction into its standards. ISA 500, the international standard on audit evidence, sets out general rules for weighing it in paragraph A35: “Audit evidence obtained directly by the auditor (for example, observation of the application of a control) is more reliable than audit evidence obtained indirectly or by inference (for example, inquiry about the application of a control)”, and “The reliability of audit evidence is increased when it is obtained from independent sources outside the entity.” ISA 330 is specific about controls. When an auditor tests whether a control operated, paragraph 10(a) requires other procedures “in combination with inquiry”, so asking someone whether the control works is never enough on its own. A green tile fed by self-assessment is inquiry without the other procedures.

Follow the green back to its source

If the trail ends at a person asserting the thing the tile now shows, the tile is reporting an assertion. A questionnaire answered “yes”. A control marked “implemented” in a setup wizard. A policy uploaded, which shows that a policy exists and nothing about whether anyone follows it. In each case the tool collected a statement and displayed it as status.

Self-attestation is a claim. Observation is what tests it. A tool that treats the claim as the test is reporting what you told it.

Nothing in that pipeline can disagree with the person filling it in. The assertion goes in and the colour comes out. The step that could catch a gap, checking the claim against something that would look different if it were false, is not part of it.

A tool that credits self-reports produces the confidence it exists to test

An assurance tool exists to test confidence: to find the places where what an organisation believes and what is true have come apart. A tool that turns self-reports green does the opposite. It takes the claims it was meant to examine and returns them as results. This is a property of the design, not of anyone’s intentions, and it holds however carefully the self-reports were made.

It matters most when the claim is wrong. A tool that repeats what it was told agrees with you in exactly the case it needed to catch.

That is why Frank gives no credit for self-reports. Self-assessment has a use: it is a reasonable way to decide where to look. But once a self-assessment can earn positive confidence, the confidence records what someone entered, not what was observed.

More integrations do not settle the question

Pulling an artefact from a live system is better than a questionnaire, because it is an observation. Connecting more systems is the right direction, but it does not answer the question on its own. If a self-reported “yes” can still turn a control green where no integration covers it, or a captured artefact keeps counting after the thing it shows has changed, an assertion is still being graded as evidence alongside the observations. The number of integrations is not the test. The test is whether a claim can go green without an observation that would have looked different if the claim were false.

What has to be true before Frank shows a posture

In Frank, only observation can earn positive confidence. A claim can lower posture without evidence, as a self-reported issue does, but it cannot raise it.

• A claim earns positive confidence only when something observed backs it. An assertion with no covering evidence stays unsupported until a current observation supports it. The missing basis is shown as a gap, not filled in with a colour.
• Every posture carries an expiry. Holds through YYYY-MM-DD, tied to the evidence behind it. A fact checked six months ago is evidence about six months ago, so the confidence decays with it.
• The basis is always inspectable. Each posture shows the evidence behind it, the exceptions weighing on it, and the person accountable for it, so anyone relying on it can check the basis instead of the colour.

The point of these limits is that nobody can produce a green posture by asserting one.

Put it to any dashboard you are handed

One question separates a readout from an argument: what did this tile observe, and would it look different if the claim behind it were false? Where the answer is “someone said yes”, the green is evidence that someone said yes, and nothing more.

You can see which of your claims are backed by something observed. Start by seeing where you stand. There is no score in it. It shows which of your claims have evidence behind them, and which rest on a statement alone.