[Intro]

Do not make it sound uncertain
Only when the sentence breaks.
Show uncertainty in numbers.
Show the path the answer takes.

[Verse]

Mark the field that was inferred.
Name the source and name the date.
Separate a likely pattern
From a fact we can validate.

Aisha writes across the board:

Known.
Estimated.
Missing.
Out of scope.

Four ordinary categories
Give uncertainty a home.

[Refrain]

Explain what it cannot know.
Draw the border, show the road.
A limit clearly understood
Is more useful than a guess made good.
Trust begins where claims can end
Without pretending certainty again.

[Verse]

Every summary needs provenance.
Every score requires a test.
Every user sees the evidence,
Not just the sentence ranked as best.

Confidence must be calibrated,
Not a bright percentage badge.
Access follows named permission.
Every action leaves a path.

The model receives a new response
For questions past its ground:

"I do not have sufficient evidence."

No fabricated source is found.

[Chorus]

Explain what it cannot know.
Draw the border, show the road.
A limit clearly understood
Is more useful than a guess made good.
Trust begins where claims can end
Without pretending certainty again.

[Bridge]

[Model]
I found three related records.

[Aisha]
Are they complete? Are they current?

[Model]
Two are current. One has no verified date.

[Aisha]
Then say exactly that.

[Model]
Three related records were found. One may be outdated.
No conclusion should be automated.

[Team]
That answer does not sparkle.

[Aisha]
"That answer can be used."

[Synth Interlude]

[Final Chorus]

Explain what it cannot know.
Mark the border, guard the road.
A limit honestly displayed
Is stronger than a claim inflated.
Give the human room to slow.
Trust is not a perfect voice.

Trust is evidence, context,
And the freedom not to choose.

[Outro]

The four-note motif rises.

This time a fifth note answers.

Not certainty.

Resolution.
