EVERYTHING SOUNDS EQUALLY CERTAIN

Believed. Measured. Unknown.
All spoken with confidence.

A generated paragraph, a measured result and an open question arrive in the same voice, at the same size, with the same confidence. The interface has no way to say which is which, so the reader supplies the difference — usually wrongly.

How should an interface show what it actually knows?

By giving certainty a form rather than an adjective. HAUSE separates the states structurally: a Claim is a belief and carries a status mark; Evidence is a finding with its receipt, its date and its machine; a Question is what remains open, at the same size as an answer; a Refusal is what the system will not assert at all. Strip the styling away and a reader — or a machine — can still tell known from believed from unresolved.

01 / EXPERIENCE THE DIFFERENCE

Make the difference visible.

FICTIONAL LISTENING STUDY / TEN RECORDED CHOICES

Let the record change what can be asserted
01?02?03?04?05?06?07?08?09?10?

CLAIM

Every listener preferred room A.

OPEN

The assertion is testable. No responses have been revealed in this view.

01A02A03A04A05A06A07A08A09B10B

CLAIM

Every listener preferred room A.

REFUTED

Two listeners chose B. One counterexample would already defeat “every”.

EVIDENCE

Eight chose A; two chose B

All ten fictional responses are shown above. No population-level conclusion is claimed.

SUPPORTED
01A02A03A04A05A06A07A08A09B10B

CLAIM

Most listeners in this ten-person study preferred room A.

SUPPORTED

8 of 10 chose A. A narrower claim matches this record; preference beyond this sample remains unknown.

The type did not become more confident. The evidence changed what the sentence could responsibly say.

02 / HOW YOU MEET IT

Confidence is a property of the sentence, not of the system that produced it.

The summary reads exactly like the measurement. The inference reads exactly like the retrieved fact. Nothing in the layout separates a number someone measured on a named machine on a named date from a number that merely sounded right — and by the time a reader learns which was which, they have already quoted one of them.

03 / BENEATH THE SURFACE

Why the failure keeps returning.

Interfaces inherited their vocabulary from documents, where confidence was carried by prose and by knowing who was writing. Generated content breaks both halves: there is no author to know, and the prose is uniformly fluent. Ordinary design systems have states for loading, error and empty, and none for believed, measured or open — so the distinction has nowhere to live, and quietly disappears somewhere between the design review and the implementation.

TAKE THE RESPONSE WITH YOU

Give it a form.

Claim

An assertion that knows it must answer to evidence.

Evidence

Rows of labelled findings with status marks — receipts, not decoration.

Question

An open question given the same typographic dignity as an answer.

Refusal

Fail-closed rendered as design language, not as an error state.

Agreement

N independently-derived values that must be identical — with a FAIL row.

PUBLISHED 31 AUG 2026 · REVISED 9 SEP 2026 · VERSION 1.1

CITE

site build fb3ec31 · built 2026-09-26

CITE THIS

Article · 1.1

Hay, C. (2026). Everything sounds equally certain. In HAUSE — the problems (Version 1.1). hause.design. https://hause.design/problems/everything-sounds-equally-certain

An argument is a published object: dated, versioned, and referenceable — which is the subject of one of these pages and the practice of all of them.

HAUSE · THE SPECIMEN BOOK / 2026

GITHUB →VINDEX3.ORG →CHRISHAYUK →