# Eve design repair · 8 September 2026

The review began at `a60560f3aa`. The user identified a misleading “Question”
heading and unexplained final-row emphasis. The renderer required that emphasis;
all 37 capability comparisons inherited it. Of 94 capability slides, 73 used
tables or the same comparison treatment. These are audit findings, not a
comprehension score. The whole-library design overhaul in section 10 remains
open.

## Visual thesis

Keep the cream, ink and terracotta identity, with Georgia claims and Arial
explanations. Make relationships visible: nested containment, directed graph
verbs, a real-shaped work record, a queue gate with a refusal branch, and
separate progress categories. Use accent for an authored teaching focus, never
for an item's position. Independent alternatives receive equal visual weight.

## Content plan

Retain the capability route: choose work → understand the graph → interpret
coverage → capture work → discuss decisions → inspect delivery → walkthroughs →
operator analysis → studio reads/actions → member scope → one complete practice.
Within the graph guide, establish containment before typed links and
projections; within work tracking, inspect the record before its lifecycle and
recovery. These prerequisites are useful. The repair changes how their
relationships are taught rather than rearranging the course merely to produce a
different order.

Use the illustrative Tara / courses / enrollment case consistently. Display
names are examples, never invented catalog IDs or asserted product defects.
Preserve source pins, exact tool boundaries, detailed notes and review
questions. Do not put implementation-review metadata into the member's learning
flow.

Before choosing a treatment, each changed slide records its job, evidence,
relationship, reading order and emphasis rationale in its authoring brief.
Comparisons receive authored headings that describe their own contents. Existing
exact-value and field-reference tables remain tables. The new visual specimens
supplement the narration's existing explanation; recheck speech fingerprints.

## Interaction thesis

Keep the existing restrained entrance, source disclosure and workspace controls.
Every diagram must explain the complete relationship without animation, hover or
color alone. Narrow screens reflow named relationships instead of shrinking a
fixed canvas. PDF retains the same labels, branches and evidence limitations.

## Design standards and review scope

1. **References:** retain before/after specimens and record rejected treatments.
   This is an implementation review, not an accepted benchmark library.
2. **Typography:** align comparison headings with their actual fields; preserve
   readable body/label roles at embedded, compact, mobile and PDF sizes. The
   complete library typography and optical pass remains open.
3. **Quantities:** illustrate walkthrough accounting explicitly; distinguish
   visited, skipped and pending. Never present illustrative values as telemetry.
4. **Continuity:** repeat the Tara case and stable graph vocabulary across
   graph, coverage, task capture and practice. Do not reuse a highlight as a
   slide motif.
5. **Imagery:** use code-native diagrams and an explicitly illustrative record.
   No screenshot is fabricated. Broader real-product image treatments remain
   open.
6. **Viewing size:** inspect full guide sequences and representative slides at
   their actual browser and PDF sizes, separate from automated bounds checks.
7. **Reader evidence:** the user's two concrete defects are observed feedback.
   Neutral willing-reader tasks and benchmark acceptance remain pending.
8. **Subtraction:** remove positional emphasis and generic headings. Replace
   redundant abstract rows with the relationship or object being explained.
   Preserve content that qualifies authority, availability or evidence.

The verification receipt must distinguish implemented repairs, model optical
review, automated checks and pending human/whole-library work. Counts or passing
tests alone cannot close the eight standards.
