# Wireframe critique log

Newest on top. Each round is a table of what was found, who found it, and what it
became. A finding that did not survive a second reading stays in the log, marked
"dropped at verification", with the reason: an audit that quietly deletes its own
misses cannot be judged.

---

# The 360 measure (2026-08-10)

One finding, taken in a browser, and it is about the instrument rather than about
a screen. It came out of a question that had nothing to do with layout: whether
the thinnest state pages were too thin to be a product.

They were not. Each of the five smallest pages carries the shell, exactly one
`h1`, real copy from `microcopy.md`, a live region and a real exit, and the
loading state draws its skeleton in the proportions of the card it is waiting
for. Byte size turned out to measure how much an error state has to say, which
is not much. What the measurement found instead:

| # | Weight | Where | Found by | Was | Became |
|---|--------|-------|----------|-----|--------|
| 1 | High | `_wf.css:35`, all 49 app pages | Claude | `.stage` carried `padding: 28px 16px 56px` at every width. At a 360 viewport the app was **313px** wide, with 16px of grey down each side, so the 360 check of conventions section 9, which every screen of this stage passed, was really being run at 313. A row that breaks at a true 360 had 13 per cent of slack and passed anyway. The gutter also stood against section 1: a screen that does not reach the edge of the viewport is the mockup frame this stage removed, drawn in padding instead of in a border | `@media (max-width: 460px) { .stage.stage-app { padding: 0 } }`. The screen fills the viewport at every phone width. Measured after the fix: **345 of 345 on all 49 app pages**, no horizontal scroll, no element crossing the edge. The landing keeps `.stage-flush`, which is the same decision taken for it alone when it was built, and the desktop layout is untouched: at 1280 the app starts at 266 and is 983 wide, before and after |

The breakpoint is not a new number. 460px is where the app already switches to
phone chrome (the fixed tab bar, `_wf.css:493`), so the gutter goes at exactly
the width where the stage stops being a window and starts being a device.

**Why it was worth a round.** The check was not wrong on any one page. It was
wrong on all of them by the same 47px, which is the one thing a page by page
review cannot see: every screen agreed with every other, and the agreement was
the defect. It was also one stage from travelling, because stage 07 copies these
screens into colour and the copies would have inherited a 360 that is not 360.

**Re-verified after the fix.** All 50 pages loaded at a 360 viewport with the
stylesheet re-fetched, measured against `documentElement.clientWidth` rather
than against the window (the window includes the scrollbar, and that 15px is the
same class of error as the one being fixed): 49 of 49 app pages fill the layout
viewport exactly, zero pages with horizontal scroll, zero elements crossing the
right edge. `index.html` is the landing and has no `.app`, which is why the
count is 49 and not 50.

## Four gates the stage had skipped, closed the same day

Run against the stage contract as a checklist rather than against the screens,
four items came back unmet. None of them was a defect on a page, which is why
neither instrument had seen any of them: an instrument pointed at the screens
cannot find a step that never ran.

| # | What the stage owed | State on 2026-08-10 | Closed by |
|---|---|---|---|
| 1 | Before and after screens from the consolidation: three screens, two widths, a pair at 360 among them | Never taken. `screens/` did not exist and the hub carried no image at all, so the delta of the whole stage was argued in prose | Six pairs in `wireframes/screens/`, shot from the pre-rebuild commit `8fffb37` in a git worktree, beside today's pages at the same widths. Recoverable only because the July frame is still in git |
| 2 | `wireframes/CLAUDE.md` correct, and under ten lines | Said Welcome is "node 0.0". That is the exact defect Codex found in the ROOT `CLAUDE.md` at Step 9 (finding 8), fixed there and missed in the nested copy. Twelve lines | Node 1.1, ten lines, and the 360 rule folded into rule 5 |
| 3 | The strategic dimension of `benchmark.md` named against a concrete element of the etalon, or its absence written down | Neither happened. The words benchmark, strategic, trust and clarity appeared nowhere in this folder | Named in `screens.md` under the etalon line and on the hub: GC4 carries mechanism 3 (real names, real context), the attention row carries mechanism 1 (active voice), GC6 carries mechanism 2 (the declaration that names the fear). All three were built from day one; only the paragraph was missing |
| 4 | Why this etalon and why this first flow, in `docs/decisions.md` | Not recorded anywhere. `screens.md` said only that both were "confirmed rather than re-derived" | Recorded with the count that makes it checkable: five of the seven globals stand on Home at once, and the two that do not are absent by rule (GC5 belongs to Alerts, GC7 is barred from the calm view by D3) |

**What this says about the two instruments.** Both were pointed at the artifact,
Codex at the source and Claude at the render. Neither was pointed at the process,
and all four of these are steps that did not happen rather than things that came
out wrong. Reading the stage contract as a checklist is a third instrument, and
it costs one pass.

---

# The rebuild critique (2026-08-05, Step 9)

The stage that rebuilt the prototype against the upgraded IA (steps 3 to 8, 46 of
49 pages) closed with a critique taken on **two instruments, separately, and
merged afterwards**:

- **Codex**, read-only, on the source: contradictions between files, orphans, a
  state that is not in the code, a broken link, a violated rule. It read
  `CLAUDE.md`, `docs/decisions.md`, both wireframe docs, all 47 html files,
  `_wf.css`, `_nav.js`, the sitemap, the flows, the 16 node files and the line
  inventory. Its brief is `AGENTS.md`.
- **Claude**, in a browser, on the render: every built page at 360 and at 1280,
  measured for overflow, clipped text, tap targets and elements a fixed tab bar
  covers, then read as screens.

Neither saw the other's list until both were finished. Eight findings held, four
did not. The instruments overlapped on nothing, which is the point: the source
pass cannot see a name breaking into two lines, and the browser pass cannot see a
document describing a defect that no longer exists.

## What held

| # | Weight | Where | Found by | Was | Became |
|---|--------|-------|----------|-----|--------|
| 1 | High | `home.html` and every screen with GC4 | Claude | The status badge took its width from the name beside it. At 360 "Amazon Prime" broke into two lines and "in 19 days · Aug 20" broke after "Aug"; at 1280 in the two-column dashboard the name column fell to 99px. On the etalon screen, at the etalon width | The name and its date have a floor of 160px in `_wf.css`, and below it the badge drops to its own line, right-aligned, instead of squeezing them |
| 2 | Medium | `add-subscription-loading.html` | Claude | The preset tiles ran 15px past the edge of a 360 screen and the page gained a horizontal scrollbar, because each placeholder line carried a fixed pixel width (up to 110px) inside a two-column grid | Placeholder widths are a percentage scale in `_wf.css` (`w40` to `w90`), so they narrow with the column, and the text half of a tile can shrink |
| 3 | Medium | `guided-reveal.html` | Claude | Two counters said "of 3" on one screen and meant different things: the header said step 3 of the onboarding chain, the first line under it said step 1 of the reveal | The chain counter is gone from that one screen. The chain ends when the reveal starts, so the counter that stays is the reveal's own. `guided-reveal-empty` keeps its header counter: it has no reveal to count |
| 4 | Medium | `cancel-guide.html`, `cancel-guide-no-guide.html` | Claude | Three sentences ("Aug 3, the end of the month you paid for") were set as data values, right-aligned against their labels, so each wrapped into a ragged column at 360. The block that answers "what happens if I cancel" is the anxiety block of the screen | `.facts.sentences`: when the answers are sentences the pair stacks and reads left to right |
| 5 | Medium | `overview.html:371` | Codex | The visible hub named **Peloton** as the failed payment, against `conventions.md` and against the state page that renders it (`<h1>Amazon Prime</h1>`) | Amazon Prime, Jul 20, $14.99, in the canonical set on the hub |
| 6 | Medium | 19 rebuilt pages | Codex | Twenty-nine literal CSS values still sat in `style=` attributes, four of them the same rule written out on four screens, against both criteria of conventions section 3 | Named classes in `_wf.css`. **Zero `style` attributes** on the 39 MVP pages and on the landing, which is one grep rather than a judgement |
| 7 | Low | `conventions.md`, `voice/docs/microcopy.md` | Codex | Both still described "alerts dates the failure Jul 2" as an open carried fix, though the fan-out had already dated it Jul 20 on the screen. A live document describing a defect that no longer exists | Both entries closed, with the date that is on the screen |
| 8 | Low | `CLAUDE.md:96` | Codex | The rule sent the product home page to "node 0.0 from the flows", and no node 0.0 exists in the sitemap, the flows or `screens.md`. Welcome is node 1.1 | The rule names node 1.1 |

## Dropped at verification

| Where | Claimed | Why it does not hold |
|-------|---------|----------------------|
| `alerts.html` | The last alert sits behind the fixed tab bar with the page fully scrolled | The rect came from inside a closed `<details>`. Chrome reports a laid-out box for content under `content-visibility: hidden`, and on screen the Older section is collapsed and nothing is covered. Measured, then looked at |
| `settings.html`, `data-privacy.html` | The notification switches are 20px tap targets | The `<input>` is 20px, but the whole `<label class="switch">` is the hit area and clears 55px |
| Every screen with a tab bar | The tab icon clips its content | The new-alerts dot is deliberately positioned outside the icon box, and nothing on the row has `overflow: hidden`. The check was wrong, not the screen |
| `index.html` | The landing's section links are 26px tap targets at 360 | Those are the footer links, and the top nav hides its section links below 760. The one real case, "Sign in" at 20px, was fixed: the landing is opened on the same phone as the app, so its nav clears the same 44px |

## What both instruments found clean

State coverage against `screens.md` (46 built, 3 spec, no invented state, no
orphan file), the registry against the file system, labels and slugs against the
matrix, every local link, greyscale (no hex, no shadow, no font-family, no
`<img>`, no `<style>` block), no em dash, one `h1` per page, the correct current
tab on every app screen, the canonical numbers across neighbouring screens, and
the tier rules (no cap on Free, no paywall in front of basic visibility or at the
cancel moment). One screen has no visible exit, `connect-bank-loading.html`, and
that is the written exception: a wait the system finishes by itself.

## Verified after the fixes

47 pages, every local link resolving, no horizontal scroll and no clipped text at
360 or at 1280 on any built page, no `style` attribute anywhere in the rebuilt
set, and the four changed screens walked again in a browser at both widths.

## Round 2, the same day: the stage closed at 50 of 50

The critique closed the MVP set; round 2 closed the stage. The three LATER
screens (node 4.11, node 5.12, node 5.13) came off the July frame, three states
were drawn for the first time, and one more was numbered in the map on the way.
What is worth carrying forward from it:

- **The bank round ran before the screens were drawn, not after.** All three
  nodes carried the line "a bank round precedes the build", and it would have
  been easy to treat as a formality. It was not: types H, I and J in
  `ia/docs/blocks.md` produced 32 decisions, 12 of them refusals of patterns the
  category agrees on unanimously.
- **A map fix, not a wireframe fix.** The trends screen could not be drawn
  honestly: the canonical person is on Free everywhere else in the product, and
  node 5.13 says out loud that she reached the upgrade screen "from Your trends".
  The locked view stood in the node file as an unnumbered variant, and an
  unnumbered variant gets no page. It was numbered as **5.12.4** in
  `ia/docs/sitemap.md` and drawn, which is the rule working the way it should:
  the hole was patched in the owner, not papered over here.
- **The `.phone` block is gone from `_wf.css`.** Three hundred lines of July
  frame, its two-pane variants and its dashboard override, deleted the moment the
  last page left them. That is what "strictly additive" was for.
- **Two pages now have no way out and both are correct:**
  `connect-bank-loading` and `upgrade-processing`. A way out in the middle of a
  charge is the one thing that screen must not offer. The reachability rule in
  `conventions.md` names both, and names the two pages that are unreachable by
  click on purpose.

---

# The first build critique (July 2026, 41 pages)

Phase: Wireframes, Step 9 of the July build. A strict pass over all 40 wireframe
pages against wireframes/docs/conventions.md, ia/docs/sitemap.md, and
ia/docs/flows.md. Defects first as a table, then the fixes applied. Structure
only; no visual design.

## What was checked

Every page in wireframes/ (40 screens plus index) across these defect classes:
- Stray look: color, hex, shadow, font-family, icon fonts, real logos, images.
- Placeholder text: lorem ipsum or filler instead of real Tendd copy.
- Missing states: each screen's state pages vs the marks in screens.md.
- Dead ends: an empty or error page with no visible exit.
- Zone without a main action.
- Screen not on the map (sitemap.md / flows.md).
- Shell drift: layout, wireframe tree, wf-meta, wf-states, css and _nav.js.
- Chrome: onboarding has no tab bar; app screens have the right current tab.
- Broken links; em dashes.

Method: an automated audit script over all pages, plus full reads of
data-privacy, alerts, connections, share-snapshot, cancel-guide-blocked, and
history-trends-empty to judge content-level quality, and a container-query
desktop reflow check.

## Defects found

| # | Severity | Screen | Defect | Resolution |
|---|----------|--------|--------|------------|
| 1 | High | subscription-detail-unrecognized.html | Inline `font-family:monospace` (plus a font-size) on the raw charge descriptor. A typography decision, which the conventions defer to Design. | Fixed: removed the inline font; the raw descriptor still reads as raw from its content. |
| 2 | Low x15 | various | Heuristic "zone with no action". | Not defects: every flagged zone is a title, a trust statement, or a form/checkbox zone, and each page has a clear main action elsewhere (verified by reading the pages and their links). No change. |
| 3 | Info | 24 agent-built pages | Inline `style=` used a little more than "tiny one-offs" (font-size, margins). | Kept: all values are greyscale (var() tokens) and structural, not brand or type-scale decisions. Acceptable at wireframe fidelity. |

## Cross-cutting checks (all passed)

- 41 files present (40 screens + index); filenames match the screens.md File column.
- Zero broken links; every link target is one of the 40 valid pages.
- Zero dead ends; every empty and every error page has a real exit, cross-checked with flows.md (bank error keeps the manual fallback, could-not-cancel has an in-app next step, failed-payment has a next-step card, sync-error keeps the last known list).
- State coverage matches screens.md exactly (no missing state, no invented state).
- Greyscale only: no hex color, no shadow, no icon font, no real logo, no `<img>`. One color leak that grep cannot see, native radios and checkboxes rendering in the OS blue, was fixed at the CSS level with a grey `accent-color`.
- No em dashes. One `<h1>` per page. Correct onboarding (no tab bar) vs app (correct current tab) chrome.
- Desktop responsive: every screen reflows through the shared container-query CSS (bottom tab bar becomes a left rail, content widens; onboarding stays a centered column). Verified live on Home, Welcome, and Upgrade.
- Main-flow click-through passes end to end: welcome, path-choice, connect-bank, loading, guided-reveal, home, subscription-detail, with state transitions and both-way branches.

## Outcome

The wireframe set is complete and consistent: 16 screens, 40 pages, all four
states covered where the scenario produces them, no dead ends, greyscale and
semantic, with real Tendd copy and a live desktop reflow. One real defect was
found and fixed; the rest of the audit was clean. Ready for the Design phase.

---

## Deep audit (Step 9 follow-up, four independent reviewers)

A second, deeper pass ran four independent reviewers (onboarding+core,
alerts+cancel, pro+account, cross-cutting) plus a visual review. The system
layer was confirmed clean (shell, naming, nav tree, state coverage, chrome,
greyscale, responsive reflow, links, no em dashes). The issues were in content,
flow, and data consistency. All of the following were fixed.

### High
- **settings.html** advertised Free as "up to 10 subscriptions", contradicting
  the locked D3 (Free is unlimited on count, gated only by depth/history).
  Fixed: "Unlimited subscriptions. Up to 2 bank connections. History, trends
  and advanced alerts are Pro." (The connection clause was later dropped by a
  July 2026 product decision - Free is uncapped on connections; settings.html
  now reads "Unlimited subscriptions and bank connections." See account.md.)

### Medium
- **Save tab / "Review these 2" / cancel-candidates nudge** all pointed at
  home-empty.html (the zero-state), because the FLAG 1 save-focus state had no
  wireframe. Fixed: added **home-savefocus.html** (candidates pinned open,
  summary swaps to potential savings) and retargeted the Save tab (all app
  pages), the reveal CTA, and the nudge to it. This is the 41st page; _nav.js,
  screens.md, and index.html were updated.
- **Canonical example dataset** locked to remove number drift. Home now shows
  all 14 subscriptions grouped 4/4/2/2/2 (matching the Guided Reveal), summing
  to the displayed monthly total **$192.90**. Propagated to Guided Reveal,
  home-error, History and Trends. Cancel candidates (Peloton App, The New York
  Times, save $29.99) are now visible rows. share-snapshot fixed to the
  post-cancel figures (13 subscriptions, $174.91, freed up $17.99), removing
  the Netflix double-count. Alerts reconciled to the dataset (Netflix next Aug
  3; payment-failed on Amazon Prime not the Peloton trial; trial-ending on
  Peloton).
- **data-privacy.html** delete-everything control moved to position 3, directly
  under the data explanation (one-tap reachable, per account.md), instead of
  last. Export copy now separates the free full download from the Pro CSV.
- **connections.html** dropped the invented Data-and-privacy zone (the title
  already links it) and removed "Reconnect" from the healthy source (it belongs
  to needs-reauth). Source counts reconciled to 14 (11 bank + 3 manual).
- **cancel-guide-blocked.html** re-weighted so the free next step ("I cancelled
  it") is the primary action and the Pro guide is secondary; D3 keeps the
  could-not-cancel moment a trust moment, not an upsell.
- **share-snapshot.html** card dropped its inline custom type scale
  (font-size:34px, weight, tracking) in favour of the shared .count/.total
  classes.

### Low
- Error states unified to a calm `role="status" aria-live="polite"` (was a mix
  of alertdialog+assertive and alert).
- upgrade.html pay button now states the amount ("Start Tendd Pro - $69 a
  year"); a lifetime option is noted as deferred [?] (D4).
- history-trends.html time-range control is now real buttons with aria-pressed
  (was anchors cosplaying as tabs).
- home cancel-candidates nudge dropped the misleading aria-expanded (it
  navigates rather than expanding in place).
- cancel-win.html wf-states label standardized to "success (base)".

### Verified after fixes
42 html files (41 screens + index), zero broken links, zero dead ends, no
stray color/font/shadow/img, no em dashes, shell intact on every page, and the
example dataset is arithmetically consistent across Home, Guided Reveal,
save-focus, alerts, cancel, share, and history. Known-open [?] items remain
(auth target for "Sign in", lifetime price) and are annotated, not invented.

---

## Declutter pass (post-audit, July 2026)

Founder feedback: the frames carried too much reviewer scaffolding and the
desktop layout was not actually reflowing well. Applied across all 41 frames:

- Removed the reviewer chrome from every frame: the left wireframe tree, the
  meta/annotation bar (cluster, job, flow, tap depth), the "this screen:
  states / overview" strip, and the per-page desktop-reflow note. _nav.js
  (which rendered the tree) is deleted; nothing references it now. Each frame
  is now only the product screen.
- The overview (index.html) is unchanged and remains the single place that
  lists every page and state, grouped by flow. Navigation between frames is via
  the overview plus the product's own in-screen links.
- Desktop responsive made real. With the 250px tree gone, the stage is
  full-width, so the container query on `.stage` fires and the wide-screen
  layout shows: bottom tab bar becomes a left rail, the app shell centers at
  960px, and the content column widens with a capped reading measure.
  Onboarding stays a single centered column. Verified live at 390px and 1200px
  on Home (app) and Welcome (onboarding).
- Kept at wireframe fidelity: zone borders and zone labels remain (they are the
  structure being communicated, not reviewer chrome). Greyscale, semantic HTML,
  real copy, and the four-states-as-pages rule are all unchanged.

## Welcome rebuilt as the public landing (founder decision, July 2026)

Founder feedback: on desktop the Welcome screen was too narrow (capped at 560px
as an onboarding step) and too thin on content for a page that sells the
product. Decision (of three options offered): make it a full-width marketing
landing rather than a minimal onboarding step.

- welcome.html is now a full-width landing wireframe: top nav (wordmark, section
  links, Sign in, Get started), a two-column hero (value copy + a calm app
  preview using the canonical 14 subs / $192.90 dataset), three benefit cards
  (one per core job: J1 see all, J2 calm not judged, J4 no surprises), a
  three-step "How Tendd works", a trust and security block (read-only, Plaid,
  delete anytime, never sell), a three-quote social-proof band in the persona
  voices (Emma, Ravi, Claudia), a final CTA, and a marketing footer.
- Still a wireframe: greyscale, semantic HTML, real calm copy, no color or type
  decisions. New landing classes (`.landing`, `.lp-*`) added to _wf.css,
  mobile-first with a >=760px breakpoint (hero and card grids go multi-column).
- It is the only page that uses full width and the only one with a footer (it
  is a marketing page, not an app screen). All CTAs lead into the onboarding
  chain at path-choice.html; Sign in goes to settings.html. Footer links to
  unbuilt marketing pages are `#` stubs; Privacy and Data and privacy point to
  data-privacy.html. Verified at 390px and 1280px.

## Side menu restored + full-width + two-pane (founder decision, July 2026)

Founder asked to bring back the side menu (the wireframe tree) and to let the
app screens use the full width like the landing, with a real two-pane where it
helps. Chosen option: tree back + full-width content + two-pane for Home and
Subscription Detail.

- The review side menu (_nav.js tree) is restored on all 41 pages, including the
  landing (which sits in an edge-to-edge `.stage-flush` beside it). It collapses
  to a top "Wireframe map" bar below 820px. Only the annotation bar, the state
  strip, and the reflow note stay removed.
- App screens now fill the available width on desktop (the 960px cap and the
  620px zone cap are gone). The bottom tab bar still becomes a left rail.
- Home and Subscription Detail are a genuine two-pane master-detail at stage
  width >= 1040. Home: the list is the master, a selected-item detail pane
  (Netflix, mirroring the Subscription Detail blocks) opens on the right instead
  of navigating away. Subscription Detail: the grouped list returns as a master
  on the left with the current row marked, the detail sits on the right. Both
  panes are hidden on mobile and mid-desktop, so the phone experience is
  unchanged (a row still opens subscription-detail.html).
- Verified live at 1440px (Home two-pane, Detail two-pane, Settings full-width,
  landing with tree) and 390px (tree collapsed, single column, panes hidden).

## Post-hoc reconciliation (July 12 2026)

- **Upgrade annual-savings copy vs D4.** upgrade.html said the yearly plan was
  "two months cheaper than paying monthly", but D4 ($7.99/mo, $69/yr) makes the
  annual save $95.88 - $69 = $26.88, about 3.4 months, not two. Reworded to "it
  saves about $27 a year versus paying monthly" in upgrade.html and the two
  voice inventory mirrors (voice/docs/microcopy.md, voice/microcopy.html).
  Caught while adapting the Voice prompt to the current repo.

## Full-page dashboard shell (founder decision, July 14 2026)

Founder feedback: the landing (welcome.html) reads as a real full-width desktop
page, but every app screen still read as a phone card framed in a grey field,
and the bottom tab bar sat at the very bottom of a tall scrolling card rather
than behaving like app chrome. The critique, verified live at 1440px:

- Phone-frame-in-a-box: each app screen is a bordered `.phone` card floating in
  the grey `--page` stage; on desktop it reads as a phone mockup on a page, not
  a product. The landing has no such frame.
- Wasted width: Home two-pane put the list in a ~340px ribbon with a short
  detail and a large empty void bottom-right; single-column screens capped at
  820px and hugged the rail, leaving a wide empty field.
- The bottom tab bar was not chrome: `position: absolute; bottom: 0` inside a
  720px-min card, so on a long list it was stranded far below, not fixed.
- The desktop rail was a thin stub in a tall empty column.

Decision (of the options offered): turn the app shell into a real full-page
dashboard, like the landing, still at wireframe fidelity (greyscale, structure,
real copy). Onboarding and the landing are untouched.

Implementation (all in _wf.css, behind a new `.dash` class so the change is
opt-in per page):

- The old framed-phone desktop rules are fenced with `:not(.dash)`, so every
  non-dash page renders exactly as before.
- `.dash` desktop (stage >= 760): the stage goes edge-to-edge (`.stage-app`
  removes the gutter), the phone loses its border and fills the viewport height,
  the header folds into the top of a persistent full-height left sidebar, and
  the tab bar becomes that sidebar's vertical nav. The content area owns the
  width and scrolls independently of the sidebar.
- Two-pane (stage >= 1040): the detail pane (Home) and the master + detail
  (Subscription Detail) span full height beside the list, so a wide screen has
  no outside void.
- Single-column app screens: a comfortable measure (max 940px), left-aligned in
  the full-bleed content area, not a centered floating card.
- Mobile (<= 460): the bottom tab bar becomes `position: fixed`, real viewport
  chrome, instead of resting at the bottom of the card.

Scope and verification: `.dash` + `.stage-app` applied to all 29 app pages
(base screens and their empty/error/loading states, including the two-pane Home
and Subscription Detail). Excluded and unchanged: onboarding (Welcome landing,
Path Choice, Connect Bank, Add Subscription, Guided Reveal) and index.html.
Verified live across a two-pane page, a single-column page, a state page, and
mobile: full-height sidebar, content owns the width, empty-state exits intact,
mobile bar fixed. Cross-checked all 42 pages: 0 broken links, 0 em dashes;
onboarding, landing, and index not modified.

Known trait: on a very wide screen a two-pane detail pane shows whitespace below
a short detail. This is normal master-detail behaviour (a reading pane taller
than its content) and was left as is.

This revises the earlier "Side menu restored + full-width + two-pane" desktop
model: the review side menu, the two-pane structure, and mobile are all kept;
what changed is that on desktop the app is now edge-to-edge dashboard chrome
rather than a framed phone widened in place.