# Presentation center design

### Eve two-track edition · 9 September 2026

The [two-track brief](authoring/eve-two-track-edition.md) records the user's
review of the Eve section and the decisions that followed: merge the 25 guides
into twelve chapters on two tracks, _Use Eve_ and _How Eve works_, and give Eve
a second accent. Teal `#1d5c63` now marks Eve-owned surfaces and chapter chrome;
terracotta marks the member and Lilith; ochre and slate mark the agent and the
verifier. The same four actor names, colours and glyphs are the vocabulary on
every Eve diagram, and the `eve-four-planes` graph was corrected to use them.

`designEdition: "eve"` adds a chapter strip (track, position bar, chapter
number, dated source scope), a track-and-chapter footer, and five reader-first
layouts: chapter cover, chapter close with a bridge to the next guide, actor
map, glyph card grid and a real capture with numbered callouts. Titles in this
edition are short names without a terminal full stop. Chapter position and
listening time are derived from the catalog's `tracks`, never authored.

Chapter 1, _Meet Eve_, is the delivered specimen: nine slides, one running
example, one real Explorer capture, no tables. The remaining eleven chapters
carry their inherited slides and gain only the chrome until each is rewritten in
this system. Browser, PDF and narration evidence for this increment is in the
[verification receipt](verification/eve-two-track-edition-2026-09-09/README.md).

### Eve comparison and relationship repair · 8 September 2026

The [Eve design brief](authoring/eve-design-polish.md) addresses two observed
defects: a hard-coded “Question” label for goals/actions, and a validator that
required the final comparison row to receive emphasis. Comparisons now declare
three field headings and an emphasis rationale. Explicit `null` gives peers
equal weight; a named row can receive focus at any position. The original
outcome-evidence pilot retains its purposeful emphasis.

Seven Eve explanations now show nested containment, directed links to a common
subject, a single illustrative work record, a queue gate with a refusal branch,
and visited/skipped/pending accounting. Their source pins, deep notes, questions
and narration remain unchanged. Illustrations are visibly distinguished from
live records and measurements. Incoming arrows terminate on the common subject;
mobile retains each named destination. The progress example uses a common
denominator and patterns as well as labels.

An authored `showMasthead: false` removes repeated guide/brand chrome on these
specimens while preserving the title and dated source scope. Compact layouts
remove padding before reducing text; the established 16px labels and 18px body
roles remain. This is a scoped design repair. It does not complete the full
typography system, real-product imagery, every-slide overhaul or reader review
required in section 10 of `TODOS.md`.

The [rollback-session brief](authoring/v2-rollback-session.md) applies the same
content-led direction to a fourteen-slide worked tick. A legitimate three-step
opening distinguishes restore, replay and the next live frame; unnumbered
responsibilities, exact input tables and selectable source excerpts carry the
supporting explanations. Final review corrects four misleading or ambiguous
visual labels without changing speech. Embedded playback controls remain
reachable by scrolling their normal-flow footer; compact captures do not claim
first-viewport fit. The
[integration receipt](verification/v2-rollback-session-integration-2026-09-07/README.md)
records per-slide semantic, reflow, print and optical evidence. Candidate
compositions still require willing-reader and benchmark acceptance.

The [connection-quality guide brief](authoring/v2-network-quality-guide.md)
applies the content-led direction to a twelve-slide V2 guide. Its opening uses
independent evidence questions; ordered fallback rows preserve branch precedence
without implying a live transition; typed illustrative and policy numbers retain
units and provenance; an exact three-line C++ excerpt separates export metadata
from a performed request. The closing investigation alone uses an ordered flow.
Existing entrance, disclosure and navigation motion remain reduced-motion aware.
No invented game/dashboard image is introduced. Mobile review shortens numeric
headings to A/B/Used while retaining their full caption and explanation. Its
recorded integration now includes all twelve slides in expanded/collapsed
desktop and compact workspaces plus mobile. Playing controls are checked after
scrolling where needed; this does not claim an entire slide fits the first
embedded viewport. That integration's actual-size review found split number/unit
labels on mobile and small prose-table headers at PDF page fit. The shared
repair below resolves those findings; neither stage establishes benchmark
acceptance, complete typography or reader evidence.

### Shared table repair · verified implementation

The [shared typography brief](authoring/table-typography.md) addresses those two
integration findings without changing authored content or narration. The shared
`--table-header-size` token gives slide tables 16px authored-case headers with
normal letter spacing, including compact playback. Recognized number/unit
expressions remain intact without converting their exact characters. Narrow row
identities reserve a readable value column; ordinary rows use 120–146px and
quantity/exact-source rows use 132–146px, with 14px right padding.

Print retains natural table height and visible overflow, correcting a wrapper
that could clip final-row ink despite passing text extraction. All 40 fresh PDFs
preserve exact table text and order. Their 322 headers across 114 table
instances meet the 12px effective-size floor at 960px page fit; Poppler reports
embedded, Unicode-mapped Liberation Sans. This is integer point metadata, not
precise glyph height or reader-approved typography.

Source excerpts keep their separate 15px text and independent scrolling. A 160px
cell minimum prevents compressed ordinary source labels, while the exact
revision wraps inside the narrow dialog. They do not inherit sticky row
identity. The source regression checks actual Tara disclosure and
keyboard/axe/focus behavior, plus every embedded source-table specimen at three
widths. Its isolated inventory geometry is not a complete source-flow or optical
pass.

The current evidence includes actual 200%/400% browser zoom, enlarged text,
explicit font probes, full center exports, and bounded browser/PDF review. The
explicit fallback probe resolves to the same Liberation Sans as the preferred
stack on this host; distinct-face coverage is not claimed. The full 421-slide
bank passes all twelve checks with observed exit zero, including per-track
playback and final artifact stability. The
[verification receipt](verification/shared-table-typography-2026-09-07/README.md)
separates directly inspected captures, exact-PNG inheritance and exact
unchanged-region PDF comparisons. The
[short-viewport navigation repair](verification/center-short-viewports-2026-09-07/README.md)
subsequently addresses the 400% center header, full title and focus transitions.
Complete typography, optical alignment and willing-reader acceptance remain
open.

## Content-led pilot · work in progress

The section 10 contract in `TODOS.md` supersedes the historical layout choices
below. The baseline at `ed2780cfb7` plus the in-progress Psyche manuscript has
387 unique slides, 413 guide instances and 33 four-column openings. These counts
locate review candidates; they are not a visual-quality score.

Visual thesis: keep the cream, ink and terracotta workspace, using a spacious
open return path for a member's optional activities and aligned evidence rows
for independent availability questions. Emphasis belongs to the relationship,
not automatic numbering or identical boxes.

Content plan: establish the outcome-evidence distinction; introduce Lilith's
four-room scope; show optional activity and return; then explain shell, release,
installation, account, design, maturity and an interrupted PWA case. The
complete eleven-slide web/PWA guide is the pilot's rhythm specimen. Add focused
architecture, runtime decision and quantitative examples to cover communication
jobs that the guide does not contain. The first six candidate treatments are an
implementation step, not pilot acceptance or a whole-library migration.

[The complete web/PWA guide brief](authoring/v1-web-pwa-guide.md) now records
each slide's job, evidence anchor, relationship, emphasis, reading order and
transition. Its remaining treatment choices are plans, not completed designs;
the canonical slide-order test guards against silently omitting a pilot slide.

Interaction thesis: retain the existing slide entrance, source-dialog reveal and
sidebar-width transition, all reduced-motion aware. The new diagrams need no
motion or interaction to convey their complete meaning in HTML or PDF.

Canonical per-guide direction and pending pilot coverage live in
`authoring/composition-pilot.json`. A slide's `brief` is authoring-only; it must
not appear in visible slides, search text or narration. New semantic treatments
require an explicit layout, node reading order, emphasis, complete relational
description and valid labeled edges. Unknown and missing layouts fail with the
slide ID instead of becoming generic columns. Legacy layouts remain explicit
migration targets; accepting their names does not count as an editorial pass.

The initial recurrence treatment groups optional peers without arrows between
them. Choice, pause/finish, a no-activity path and the return are distinct
relationships. Independent evidence questions use aligned definition rows, with
the final outcome emphasized; they have no directed edges or ordinal numbers.
Responsive HTML preserves labels and reading order without scaling a fixed
canvas. The return connector is a native SVG with live HTML labels.

The priority-decision candidate replaces Psyche's runtime fallback overview with
a source-ordered ladder. Absence proceeds down; presence joins selected-signal
validation; valid and invalid outcomes terminate independently. The no-match
result remains separate. Native CSS routing and full named destinations preserve
meaning on narrow screens and in accessible text. This specific first-match
contract does not pretend to implement every decision-tree or ownership shape.
[The decision brief](authoring/decision-pilot.md) records its source-backed
communication job and neutral reader tasks. It remains a candidate, not accepted
reader evidence.

Compact active narration has a separate height budget. The priority-decision
slide uses an 8px section gap and 12px caption separation in that state, keeping
the complete playback controls visible without reducing its label sizes. The
focused pilot matrix plays every candidate at 1280×720; static fit alone does
not cover the additional status and seek controls.

Pilot typography uses the established Georgia/Times/serif title and
Arial/Helvetica/sans-serif body stacks. The semantic bodies use 18–22 CSS px,
headings 23–29 px (priority-condition names 20 px), and captions/connector
labels at least 16 px; compact and print bodies use 18 px. These are provisional
engineering floors, not reader-validated thresholds. Exact font availability,
fallback/enlargement, actual embedded sizes, page-fit PDF legibility and optical
balance need their own evidence. No font files or remote assets are introduced.

Reader review remains pending. Do not promote these treatments as accepted
benchmarks, migrate the library by imitation, or claim comprehension from
Playwright or model inspection. Continue independent implementation and source
work while arranging neutral, version-bound review tasks for willing readers.

### Exact quantities and table continuity

The [table pilot brief](authoring/table-pilot.md) retains a table for configured
latency targets: readers need exact values across different measurement
contracts, not bar lengths that invite a ranking of unlike intervals. Typed
numbers, explicit missing-state objects, kind/unit and cited/pinned source
identities make the data reviewable. The visible caption distinguishes missing
configuration from zero; seven numeric values and five missing entries bind to
the four pinned exports. Number strings, booleans, null, nonfinite values,
ambiguous missing markers and incomplete provenance are rejected. Generic tables
can preserve legitimate negative/zero values without inventing them for this
latency example. This is not the complete chart repertoire or an accepted
quantitative benchmark.

Numeric columns use right alignment and tabular figures; numeric headers,
captions and cells remain at least 16px in the focused viewing matrix, including
compact playback. Exact values are not rounded by the renderer. Full provenance
records source symbols, conversion/aggregation policy, population and period;
configuration records explicitly identify sample/window fields as inapplicable.

Native row/column semantics remain unchanged. At widths up to 600px, an opaque
sticky first column retains row identity. Container-relative fixed columns give
each prose column the available reading width; compact numeric columns can fit
together where space allows. Long labels wrap within their column. The quantity,
unit and missing-value caption also stays within the scroll viewport. Source
dialog tables are excluded. A resize observer shows the scroll hint only when
the actual table overflows. Print retains the complete static table.

The shared readability check tests real text ranges against padded cells and
visits every narrow column beside its anchored row name, including caption
visibility. A deliberately disabled sticky rule must fail this check. These are
geometry/semantic guards, not evidence of willing-reader comprehension.

Full-height mobile table evidence uses an explicit same-width expanded viewport
and verifies unchanged cell geometry and scroll position before and after the
PNG. This avoids a reproduced Chromium 148 beyond-viewport screenshot side
effect that moved a 235px scroll offset to 101px. The original tested viewport
is restored and both sizes are recorded; a tall reading image is not evidence
that the whole slide fits inside a phone's initial viewport. Embedded captures
remain at the actual frame size. Prose row headings have a 120px minimum width
to keep ordinary identifiers such as “Interruption” intact; quantitative rows
retain their compact source-label widths.

The [companion-label repair](authoring/v2-companion-label.md) demonstrates why
cell bounds alone cannot establish prose readability: a word can break inside
its cell. Its visible row names use “Auth contract” and “Mobile app”, preserving
the full OAuth and planned-client explanation beside them and in notes. The
browser audit checks each word's line geometry and the unchanged 16px floor on
this reviewed slide, including narrow and enlarged row text. It does not impose
a no-wrap rule on exact API/code identifiers elsewhere. Broader typography and
reader acceptance remain open.

### Named responsibility handoffs

The [ownership pilot](authoring/ownership-pilot.md) adds a concrete Tara read
detail after the conceptual architecture overview. Three named owners identify
caller context, shared orchestration/composition and the configured API read
dependency. They are code responsibilities, not asserted deployment units. The
three reads start together; their successful fulfillment precedes composition.
The failure caption preserves rejection without a partial envelope or automatic
cancellation. Exact URL and validation boundaries remain in notes and sources.

`ownership-handoffs` separates stable owner IDs and ordered exchange IDs from
geometry. Each exchange declares request/return, endpoints and a label; optional
detail must add information. Repeated owner pairs are legitimate. Missing
references, orphan owners, invalid roles/kinds and duplicate exchange IDs fail
authoring. A scoped scenario states what the order means; the treatment does not
model arbitrary branches or elapsed time.

Wide views align requests and returns to owner centers with native CSS
arrowheads. Duplicate endpoint names remain available nonvisually, becoming
visible at widths up to 1000px. More than four owners use the named-exchange
expression instead of squeezing lane labels. Owner headings use 22px, owner
responsibilities 18px, transfer labels 20px and supporting labels/captions at
least 16px. Compact playback uses the same 8px section gap as the priority
candidate. Redundant transfer detail was removed before changing the height
budget; no semantic label was reduced below its floor.

The browser audit checks actual attachment points, arrowhead orientation, named
recipients and visible labels, including print mode. Negative controls reverse
an arrow, detach a route and hide/clip a label. PDF text checks retain scope,
owners, responsibilities, payloads and exchange order; exported arrow direction
additionally requires the print-geometry check and visual review. Synthetic long
names wrap without clipping, but an unbroken identifier can still leave a short
fragment: prefer concise display names and preserve exact identifiers in
supporting text. This does not close the full typography or reader-comprehension
standard.

### Annotated exact-source artifacts

The [artifact pilot](authoring/artifact-pilot.md) replaces the PWA slide's
implied install/cache/update/clear chain with an exact function-body excerpt.
The artifact and its annotations have distinct roles: original code establishes
the local guard/message/return boundary; short adjacent interpretations explain
what that boundary does and does not prove. Existing installation, native-scope,
route-sensitive reading, queue and privacy explanations remain in the notes.

`annotated-source` requires a cited immutable source, symbol, excerpt label,
positive starting line and exact line strings. The builder compares them with
the same Git snapshot used by source excerpts. Ordered annotation ranges must
partition the entire specimen without missing, duplicated or out-of-range lines;
one annotation is permitted when appropriate. Annotation IDs, ranges, emphasis
and reading order are semantic data. No directed edge or annotation-count quota
is implied. The excerpt's omitted signature/brace and source-versus-device scope
are explicitly labeled.

Native `pre`/`code` retains whitespace and selectable characters; live source
bands connect to accessible interpretations by ID. Georgia titles and Arial
source/body text preserve the two-family identity. The code has no ornamental
syntax palette, fake terminal chrome or text baked into a bitmap. Display-only
soft wrapping retains exact characters and source line identities. Native `wbr`
opportunities at camel-case, dot, slash and underscore boundaries reduce broken
identifier fragments without inserting spaces or hyphens. Source line ranges
remain together; character and provenance checks run after reflow. Source text
and interpretation use 18px, headings 20px, symbol 22px and provenance/location
labels 16px. These remain provisional engineering floors, not reader evidence.

Wide bands align code with its interpretation; at widths up to 1000px each
interpretation follows its source block. Line ranges sit beside headings to
remove repeated location rows. Artifact identity shares a line where space
allows; full source path/pin remain visible and wrap. Compact view uses an 8px
slide gap, 4px band padding and 6px artifact-heading separation, preserving code
sizes. Sparse rules separate source bands; the returned boolean receives the
single terracotta emphasis. Print keeps the complete static specimen and
annotations.

Source tests distinguish posted requests from worker/page effects, including
posting exceptions and repeated calls. Browser checks bind exact characters,
line numbers and annotations, exercise narrow/enlarged reflow and reject hidden,
clipped or altered source content. PDF text checks preserve provenance, excerpt
order and interpretations; actual-size optical and reader evidence remain
separate. This is a code-mechanism candidate, not completion of image,
screenshot, full-guide or benchmark requirements.

## Visual thesis

A quiet cream-and-ink Oshun reading workspace, with terracotta navigation and
the presentation as the dominant visual plane.

## Content plan

A topic/subtopic sidebar or mobile drawer; a minimal selected-topic header; the
actual focused presentation filling the right workspace; source explanations,
narration and PDF access where useful. No giant master presentation as a
delivered reading route.

## Interaction thesis

- A restrained sidebar width transition that gives space back to the
  presentation.
- Clear disclosure and selected-topic transitions for browsing the hierarchy.
- A brief content loading reveal; all transitions respect reduced motion.

## Short-viewport navigation

At viewport heights of at most 480 CSS pixels, a single content-sized row keeps
Topics, the complete guide title at 14px and Guide details. The iframe fills the
remaining viewport. Route, position, measured duration, summary, prerequisites,
previous/next, separately-open and PDF actions remain in the guide dialog, which
starts at its actions whenever opened. Its heading and close control stay
reachable as the content scrolls. The short sidebar scrolls as one document,
retaining its browsing position and a sticky brand/close row.

Breakpoint focus handling remembers a control hidden by CSS before Chromium's
media-query event, without retaining an intentional blur from a visible control.
Selecting a guide from either dialog moves focus to the presentation. Ordinary
height layouts retain the two existing toolbar rows.

The [implementation brief](authoring/center-short-viewports.md) and linked
receipt keep actual tab zoom separate from supplemental viewport emulation. The
480px breakpoint and 44px control geometry are implementation constraints, not
reader-approved legibility thresholds or whole-slide first-screen fit.

## Coverage and enrichment

The Eve/V1 library retains the 244 reviewed foundation slides and adds 105
topic-specific explanations, worked cases and failure analyses across 35 guides.
Each guide includes the context it needs to stand alone. Recommended preparation
links support a deeper reading sequence. The source map accounts for every
foundation slide and connects all 191 reviewed sources to the guides.

Narration follows the current script and the requested "oh-shuhn" pronunciation.
Guide and ancestor listening times use verified audio durations; shared context
counts in each guide where it is played. Audio fingerprints prevent an older
narration from playing against changed content.

The portfolio orientation introduces V1–V10 before the existing Eve/V1 reading
sequence. Twelve slides give each product one clear purpose and connect the
portfolio through shared ownership and evidence. It combines the existing column
and flow language with diagrams of product composition and V3 world/tenant
boundaries. The center brand returns to the first guide in the catalog; the
deeper guides retain their current sequence and prerequisites.

V2 orientation is split by reader task: product commitments and acceptance,
game/web/companion surfaces, then contracts and verification. Comparisons use
tables, process explanations use flows, and authority relationships use labeled
diagrams. Detailed evidence remains in the explanation and source context.

The contracts draft keeps the same quiet cream, ink and terracotta workspace.
Its sequence starts with a traceable claim and reference conventions, compares
artifact kinds, follows a concrete determinism contract through build/cook and
CI, then traces proto metadata generation and closes with a bounded evidence
claim. Tables carry exact comparisons; flow slides carry the execution and
review sequences. Existing navigation, disclosure and focus transitions provide
the interaction model, with reduced motion retained.

The [eighteen-slide contracts integration](authoring/v2-contracts-guide.md) adds
aligned comparisons for executable value schemas versus artifact checks and the
reproduced stale-descriptor case. Independent declaration/consumer/proof and
checker-boundary columns have no ordinals. Enlarged-text inspection keeps the
new row labels concise while exact identifiers and isolated-execution detail
remain in notes and sources. This applies the content-led and actual-size
workflow; it is not reader acceptance or a complete typography benchmark.

Three contracts source tables explicitly opt into `sourceWrap`. Native `wbr`
opportunities at camel-case, dot, slash, underscore and section-sign boundaries
preserve the exact selectable text, including quoted anchor notation. Their
mobile row column is 132–146px with 14px cell-right padding; each value column
still fits beside its sticky row identity. Multi-value tables retain the
existing horizontally scrollable region. Neither source characters nor type size
are changed to force a fit. Tables without the opt-in retain their previous
HTML. The boolean is validated and cannot silently apply to another layout or
typed quantities. Browser word-range checks exercise narrow and enlarged source
segments, while the PDF gate retains non-whitespace characters and cell order
after line reflow. This is an explicit source-reading treatment, not a universal
rule for all table prose or a complete typography acceptance claim.

The complete isolated contracts review replaces the cache slide's linear arrows
with a route-comparison table. Explicit BuildGraph Cook entries, BuildCookRun
commands and direct wrappers expose different configuration evidence; they must
not look like one uniform command path. Compact-desktop and mobile table-end
captures accompany desktop, mobile initial-state and PDF evidence for manual
review.

The opening evidence slide compares four distinct availability states in plain
columns, without implying a progression from proposal to shipment. The
task-family and context-layer slides retain their contract comparison and
four-row table; explanations supply routing choices, retention and user controls
without crowding the visual. Existing navigation, disclosure and reduced-motion
behavior remain the interaction model for this source-review pass.

The Tara playback correction retains the existing three-column comparison of
audio controls, rate policy and accessible alternatives. Precise normalization
examples belong in the explanation, keeping the visible comparison readable; the
correction does not introduce a new visual treatment or decorative motion.

The architecture source review keeps its existing cream, ink and terracotta
workspace. The partial-failure slide retains its assembled fan-out diagram,
illustrating two successful reads and one typed error, followed by a recovery
comparison table in the guide's closing example. The diagram is one permitted
mixed-result case, not a promise that every error envelope includes success.
Exact empty-result and structural-guard counterexamples stay in the explanation.
Existing navigation, disclosure and reduced-motion behavior remain the
interaction model; no additional decorative motion is needed for these semantic
corrections.

The curation review retains the five-step keep-to-revisit flow as its dominant
visual. Package metadata, corrected inventories and the distinction between
returned library artifacts and persisted platform effects stay in the notes and
source panel. This keeps the member journey scannable without hiding the code
boundary behind a decorative package diagram.

The Sophia review keeps the five-step extractive flow, the three-column
enhancement comparison and the four-row failure table. Each has one job:
ordering source text, separating cleanup from proof, or comparing actual failure
outcomes. The blue/green counterexample, retained-prose caveat and exact numeric
thresholds live in the explanation, keeping the canvas readable. Existing
navigation, disclosure and reduced-motion behavior remain the interaction model;
no decorative imagery or additional motion is needed for this semantic review.

The Iris review separates relevance scoring and operator inspection into focused
slides, rather than making the recall overview carry both mechanisms. The
scoring table compares four factors; the inspection flow follows request,
purpose, decision, snapshot and closure. The worked case uses two plain columns
because write policy and recall deduplication are different paths, not stages of
one guaranteed correction flow. Entry layers and the failure table retain the
established cream, ink and terracotta composition. Caller obligations and
controlled counterexamples stay in the explanation. Existing navigation,
disclosure and reduced-motion transitions remain the interaction model. The
shared Eve memory slide retains its three-column recall/confirm/forget
comparison, but frames enforcement as an integration obligation. Its revised
title and explanation are reviewed in both Eve context and Iris so the reused
overview agrees with the detailed worked case.

The Psyche/Lilith review retains the quiet cream, ink and terracotta reading
surface. Its content sequence starts with ownership, separates stream admission
from turn outcomes, compares four latency metrics, then follows overload,
downgrade and recovery before the persona-policy and worked-session material.
Focused four-row tables carry turn outcomes and measurement targets; short flows
carry caller actions. Recovery's three possible outcomes use columns, because
explicit re-engagement can skip the ordinary wait and offer. Detailed
counterexamples remain in the explanation. The closing failure comparison asks
what additional evidence each returned plan needs. Existing slide navigation,
source disclosure and reduced-motion-aware selection transitions remain the
interaction model; no decorative motion or imagery is added to these
implementation explanations.

The [Agentic Studio application brief](authoring/agentic-studio-guide.md)
follows one eligible execution path between responsibility and outcome tables,
then separates saved telemetry from an external effect. Its envelope, handoff
and interruption comparisons explicitly set `numbered: false`; ordinal markers
remain on the actual ordered flow. The column renderer validates this optional
boolean rather than silently accepting a malformed or inapplicable flag. Other
legacy columns retain their previous output pending their own content-led
review. The obsolete admission SVG override is removed so it cannot replace the
authored table. These bounded refinements are not whole-library composition
acceptance.

## Eve field-guide edition

The 204 Eve slides use `designEdition: "field-guide"`. Georgia carries the main
claim; Arial carries evidence. Cream, ink and terracotta remain the shared
palette. Repeated mastheads are removed; release/source scope stays visible. The
case-study, product-view and recovery-map families extend the existing semantic
compositions. Use their meaning, not a quota, to select a layout.

New semantic evidence body is 18px, headings 23px (21px in compact recovery
rows), captions 16px. The main case claim is 30–44px. Source scope is secondary
12px metadata; it is not the only place an implementation date is disclosed.
Four inherited architecture graphs use larger nodes, wrapped 22px detail and an
unconstrained SVG canvas. Narrow center frames use native labels and the
relationship description; browser checks measure the transformed font size and
label containment, not just the SVG's declared font. Footer metadata and
controls retain their existing roles. Body copy never shrinks to repair
overflow.

The viewing matrix covers 1440×900 and 1280×720 standalone, 390×844 and 320×740
reflow, and the actual expanded/collapsed/mobile center frames. Compact desktop
slides include their controls in the viewport budget. Center frames and narrow
screens may scroll vertically. Product captures can pan horizontally with the
keyboard; each has alt text, native explanatory annotations and an
original-image link. The ten-column Work board uses a 2800px reading width on
screen. Print fits the real board on the page while native annotations retain
every state. Missing imagery does not remove the explanation.

Images come from unmodified pinned React components with declared fixtures, not
generated mock interfaces. `assets/eve-product/provenance.json` binds the
source, fixture, capture harness, browser version, pixel dimensions and PNG
hashes. The readiness example executes the pinned pure evaluator: 3 of 4 gates
are cleared (pass or waived), but a blocking safety failure and missing signoff
still return no_go. These are hypothetical inputs, not a release measurement.

The Enrollment graph/work case and related Enrollment guide provide continuity.
Records state their illustrative status. Native diagrams preserve ownership,
legal branches and source direction; alternative model configurations do not
receive containment/fallback arrows. The quality ledger retains the reason for
every table and marks model review separately from human-reader evidence.
