0001 — Voice-first input is a primary design constraint

• Status: Proposed
• Date: 2026-07-29

Context

The workbook asks for extended free-text reflection — 443 discrete blanks across the five-day and rigorous tracks, most of them wanting sentences rather than words. That volume of writing is a barrier to anyone who cannot type comfortably for long stretches, and every design decision so far has quietly assumed a keyboard and an uninterrupted hour.

This is stated as a general requirement. It describes a real class of users and does not depend on any individual’s circumstances.

Decision

Treat completion without extended typing as a first-class requirement rather than an accommodation retrofitted later. Every surface that accepts input must be usable by dictation. Feature work whose primary justification is this constraint — notably the agent bridge in 0007 — is in scope, and should say so plainly rather than dressing itself up as something else.

Consequences

C1. Autosave must never re-render or re-focus the field being edited. A debounced save that replaces the DOM node or resets the caret destroys an in-progress dictation session — minutes of speech, gone, in a way that is hard to reproduce and maddening to hit. This is a hard constraint on the storage work, not a polish item.

C2. Fields should be large and few per screen. Dictating into a dense grid of short inputs is considerably worse than dictating into one generous textarea.

C3. Progress must be durable at field granularity. A dictated session gets interrupted, and resuming should never cost re-speaking an answer.

C4. Some conventional interactions get re-examined rather than inherited. Anything that assumes rapid keyboard correction — inline validation that fights you mid-sentence, aggressive autocomplete — needs justifying against this constraint.