ZeroBalance — Budget Planner

A production-grade budget planner built around one rule — Income = Savings + Expenses — and a self-hosted, superset clone of the reference ZeroBudget app (zero-balance-4885a8f3.base44.app), rebuilt as a single Next.js application with cookie-session auth, a Prisma/SQLite store, and a full three-tier test suite.

Overview

ZeroBalance tracks income, savings, and expenses as budget items, computes your net position against the net-zero goal in real time, and breaks your balance sheet into assets and liabilities for net-worth tracking. The dashboard aggregates everything into allocation percentages, a needs/wants/savings donut, and 50/30/20 guideline rows; every expense category can be decomposed into line items through a built-in calculator that recalculates the category total server-side on every change. The clone keeps pixel-level visual parity with the reference while fixing its two mobile navigation bugs (see Superset fixes).

Key Features

Superset fixes over the reference

The live reference app has quirks and bugs, measured against the clone:

1. Hamburger click-block — the reference’s empty toast container (pointer-events: auto, full-width, top 32px, z-100) covers the top half of the mobile hamburger button. The clone’s toast viewport is pointer-events: none (the sonner pattern); toasts re-enable pointer events individually. Pinned by tests/e2e/mobile-navigation.spec.ts.
2. Menu stays open after nav — tapping a nav link in the reference’s sheet changes the route but leaves the sheet + overlay trapping the user. The clone closes the sheet on every navigation. Pinned by the same spec.
3. Root URL nav highlight — the reference marks NO nav item active when you sit on / after login (its active check compares the pathname to /dashboard). The clone highlights Dashboard on the root route — the page you are actually viewing.
4. Mobile horizontal overflow — the reference’s own pages scroll sideways on a 390px phone: the dashboard measures 395px and the net-worth page 464px (its fixed text-5xl net figure + fixed 2-col summary grid stretch main past the viewport). The clone fits exactly (390px) via min-w-0 on main plus a responsive summary card (text-2xl on phones, sm:grid-cols-2). Pinned by tests/e2e/mobile-layout.spec.ts + the net-worth mobile spec.
5. Dialogs ignore Escape AND outside-click — the reference’s modal overlays are plain fixed divs with no keyboard dismissal and no overlay-click dismissal (verified on its budget-item and line-item dialogs: Escape with focus inside and real clicks on the overlay corner both leave it up; only X/Cancel close it). The clone’s Radix dialogs close on Escape and overlay click. Pinned by tests/e2e/dialog-buttons.spec.ts.
6. Delete without confirmation — the reference’s card menu Delete destroys the item immediately (verified live — the same for its calculator line-item rows). The clone’s delete paths confirm first (the edit dialog’s inline confirm / the alert dialog).

The mobile layout itself (top bar stacked inside main at the reference’s 61px geometry, no horizontal overflow) is pinned by tests/e2e/mobile-layout.spec.ts.

Architecture

flowchart TB
    subgraph Client["Browser"]
        UI["App Router pages(login · dashboard · income ·expenses · savings · networth)"]
        Store["Zustand storeitems · assets · liabilities ·lineItems · modals"]
    end
    subgraph Server["Next.js (standalone)"]
        API["JSON API routes/api/auth/* · /api/budget-items/api/line-items · /api/assets/api/liabilities"]
        Val["zod validation"]
        Auth["scrypt + HMACcookie sessions"]
    end
    DB[("SQLitedb/custom.db")]
    UI --> Store --> API --> Val --> Auth --> DB

File Hierarchy

📂 zero-balance/
├── 📂 prisma/
│   └── 📄 schema.prisma           # User, BudgetItem, ExpenseLineItem, Asset, Liability
│   └── 📄 seed.ts                 # Idempotent demo seed ([email protected] / Demo1234!)
├── 📂 src/
│   ├── 📂 app/                    # App Router: 7 routes + 23 API handlers (15 files)
│   │   ├── 📄 login/page.tsx      # Three-state auth card
│   │   └── 📄 [dashboard·income·expenses·savings·networth]/page.tsx
│   ├── 📂 components/
│   │   ├── 📂 budget/             # Views, dialogs, sidebar, store (the app)
│   │   └── 📂 ui/                 # shadcn-style primitives (dialog, select, toast…)
│   └── 📂 lib/                    # Domain seams: money, dashboard, validation,
│                                  #   auth, rate-limit, serializers, db-path
├── 📂 tests/
│   ├── 📄 *.test.ts               # Vitest unit suites (108 tests)
│   └── 📂 e2e/                    # Playwright specs (282 tests) + global setup
├── 📂 scripts/
│   ├── 📄 smoke-test.sh           # 35-step production API smoke test
│   └── 📄 capture-screenshots.mjs # docs/screenshots generator
├── 📄 docs/                       # Tailwind v4 report, SSH push runbook, DEPLOYMENT.md, remediation plan, session log
└── 📄 db/custom.db                # SQLite database (gitignored)

Quick Start

Requires Node.js ≥ 20 (or bun ≥ 1.2) and npm.

# 1. install
npm install

# 2. database (SQLite file at db/custom.db — created by the push)
npm run db:push

# 3. seed the demo workspace
npm run db:seed

# 4. dev server → http://localhost:3000
npm run dev

Verify setup

curl -s http://localhost:3000/api/health          # {"ok":true,...}
# then log in with the seeded demo account:
#   email: [email protected]    password: Demo1234!

Production build

npm run build        # standalone output (pins DATABASE_URL for the build)
npm start            # boots .next/standalone/server.js

Environment Variables

.env (see .env.example; .env is gitignored):

Database-location gotcha: Prisma resolves a relative file: URL from a shell-inherited DATABASE_URL against the CWD, but from a .env-loaded value against the schema dir. The npm scripts pin DATABASE_URL (runtime) and use env -u (Prisma CLI) so both always land at /db/custom.db. Don’t run bare prisma db push from an arbitrary directory.

Testing

npm test            # Vitest unit suite (108 tests) — pure domain seams
npm run test:e2e    # Playwright e2e (282 tests) — needs `npm run build` first
bash scripts/smoke-test.sh   # 35-step production API smoke (own server, port 3210)
npm run typecheck   # tsc --noEmit
npm run lint        # eslint (react-hooks v6 rules enforced)

• The e2e suite boots the production standalone server on :3100 with its own db/e2e.db, deleted and re-seeded every run — specs assert the seed’s exact arithmetic (e.g. +$2065.00 net balance) and restore their fixtures after themselves. Close live browser sessions before running it (four agent-browser sessions + Playwright’s Chromium exceed the sandbox’s memory and crash the run — session-17/19 lesson).
• One worker: the specs share a single seeded SQLite file (playwright.config.ts).
• Auth is signed in once by the setup project and replayed via storageState — the auth endpoints are rate-limited (10/IP/15 min), so per-test logins would trip the limiter.
• Unit tests cover the money arithmetic (integer-cent sums — no IEEE-754 drift), dashboard aggregations, every zod schema (including empty-string optional dates and impossible calendar dates), the rate limiter, and the SQLite URL resolution contract.

Design System

Measured from the reference’s :root (computed styles as ground truth):

• Typography: the system font stack (ui-sans-serif, system-ui, …) — no webfonts.
• Money — two formatters (measured in the reference bundle): plain "$" + toFixed(2) without thousands separators on the dashboard / items views / calculator ($5550.00); toLocaleString grouping on net worth only ($25,000.00). Savings/expenses breakdown rows carry - prefixes, Net Balance +. Add buttons carry per-surface 135deg gradients (forest→lime, lime→limeLight, blue→blue, orange→orangeLight).
• Motion: sheet slide-in 500ms / slide-out 300ms; data-[state]-driven; no reduced-motion needs beyond Radix defaults.
• Tailwind v4 specifics (bare-HSL theme, oklch drift, shadow scale) are pinned in src/app/globals.css — the full trap report is docs/Tailwind-V4-Validation-Report.md.

Engineering References

| docs/remediation-plan-v41.md | The net-worth summary card’s viewport census + the vertical-rhythm fix — the savings-class viewports queue (the FIRST production-code change since v38). The surface (the session-86/87 queue’s first item — the unmeasured MID viewport class): the summary card measured live on BOTH sites at 390×844, 640×960 AND 1280×800 (nw-summary-v41.mjs + nw-yoffset-v41.mjs + nw-below-v41.mjs). At 640 (the mid viewport): FULL geometry parity — the 608×333 card (p-8, radius 16, the forest→lime gradient, the rgba(26,58,46,0.3) 0 20 60 shadow), the 48px/700 figure (sm:text-5xl), the 64×64 white/20% gradient chip at x=528 (9999px radius, the 32px icon), the 264×2 stats grid (16px gap, 84px cells), the 18px ratio, the 48@x=16 header chip, the 256/0.1 overlay, sw=vw. THE v41 PARITY FIX (F1): the net-worth page’s section rhythm ran mb-6 (24px) where the reference runs mb-8 (32px) at BOTH gaps (the header row→card and the card→tablist — 8px offsets at every viewport; the stack above/below matches byte-for-byte, the y-offset probe isolated the two margins) — the items-view pages carry mb-8 since v20 G1; the net-worth view was the one page left at the pre-v20 rhythm (items-rhythm-v41.mjs ran the no-drift control on /income: identical both sites). The MOBILE superset pair (390): the reference’s card is 432px wide at the 390 viewport (74px page overflow, sw=464, its chip 26px offscreen at x=352, its stats grid a fixed 2-col, its figure a fixed 48px); the clone’s fits (358 card, the 24px text-2xl figure — superset fix #6, the 1-col grid, the chip fully inside, sw=390) with byte-identical card chrome. F2: the 640 census e2e (the full band census + the two rhythm pins — card.y − chipRow.bottom === 32, tablist.y − card.bottom === 32); F3: the 390 superset census e2e; F4: the a11y sheet-open structural pins extended to the remaining five routes (/dashboard /income /expenses /savings /networth — the v39 sheet-open test covered / only; the same dialog-role/labelledby/nav-landmark/five-links/Escape-close contract, the full axe scan stays on / since the sheet’s DOM contribution is route-independent). FOUR pin-sanity mutations verified to FAIL: the reverted mb-8 → the rhythm pin fails (24 ≠ 32); the stripped sm:text-5xl → the 48px pin fails; the fixed 2-col grid → the 1-col pin fails; the stripped DialogTitle → all five route tests fail. Standing checks clean (35th): mobile-nav R1–R4 both sites (Tailwind v4 pins hold, the mobile menu works), data drift (reference read-only, before + after), the SEO pair. — 108/203/35 green (7 new e2e, 2-line production fix, TDD) | | docs/remediation-plan-v44.md | The drill-down expansion ANIMATION + the chevron GLYPH swap + the 390/1280 census + the donut-legend census — the session-95/96/97 queue. The surface: the drill-down’s expanded state at the REMAINING viewports (390 mobile / 1280 desktop) + the donut legend’s geometry/interaction, measured live on BOTH sites (exp-vp-v44.mjs + legend-v44.mjs + collapse-ref-v44.mjs + band-mount-v44.mjs) — TWO v44 parity fixes (the first production-code change since v43): (F1) the expansion ANIMATES on the reference (~200ms height+opacity tween, both levels, both directions — mid-collapse its band reads inline height: 25.5455px, a per-frame JS tween; the section band unmounts after the collapse animation) while the clone mounted/unmounted instantly — fixed with the Collapse wrapper (the grid 0fr↔1fr technique + @starting-style entry + the onTransitionEnd delayed unmount; the collapsed DOM returns to the pre-fix node tree — the accordion count-0 pins hold); (F4) the chevrons SWAP GLYPHS — lucide-chevron-down collapsed ↔ lucide-chevron-up expanded, both levels, rotate: none at every state (a computed-style census cannot see a glyph swap — read the svg’s class; the standing VLM pair flagged it, the DOM verified it — the v43 “static chevron” finding was correct but incomplete). Pins: the 390 census (the chrome byte-identical, the mobile indentation, the clone’s sw 390 fully expanded — the superset; the reference reads 395, its own overflow), the 1280 census (the 304px narrow column, the section label’s 2-line wrap — labelH 40 on BOTH sites, the v42 56px row explained), the donut legend census (the geometry byte-identical: h 64/pad 12/radius 8/the label-amount-pct fonts/the 16×16 dot/gap 12/mt 24 — and the hover INERT on both sites: the queue’s legend-interaction item resolved, no interaction states exist). FIVE pin-sanity mutations verified to FAIL (the zeroed duration, the removed @starting-style, the legend p-3→p-4, the sub-row py-1→py-2, the reverted glyph swap). Standing checks clean (38th): mobile-nav R1–R4 both sites (Tailwind v4 pins hold, the mobile menu works), data drift (reference read-only, before + after), the SEO pair, the VLM pair (the F4 glyph flag was REAL — the VLM screening earns its keep). — 108/214/35 green (5 new e2e, TDD) | | docs/remediation-plan-v45.md | The shadow-scale sweep + the items-view card grid census + the calculator row hover — a PIN-ONLY iteration (no production-code change; the measured chrome was already at parity, the seven new pins lock it). The v45 discovery: a SYSTEMIC v3↔v4 divergence in the computed box-shadow composition — every live shadow-* utility’s VISIBLE layers match on BOTH sites (the --shadow-sm theme pin is load-bearing for the small end: v4 RENAMED the scale, its native shadow-sm = v3’s “shadow”; md/lg/xl/2xl kept v3’s numerics) while the LEAD layers diverge by construction (v4’s 5-slot composition emits 4 transparent leads vs v3’s 2 — invisible, the accepted v10/v25 class). The md/lg/sm/2xl sites are now PINNED (the calculator row’s hover shadow-md, the quick-action card’s hover shadow-lg, the action buttons’ rest shadow-md, the slate/Google buttons’ shadow-sm, the login panel’s shadow-2xl) — a future Tailwind upgrade that changes the scale numerics fails the suite. Pins: the items-view card census at 1280 (the grid 960 = the rail 256 + the content column 1024 − the p-8 64 — the SAME arithmetic on both sites, 3 cols × 309.328, gap 16; the card chrome; the REAL hover INERT on both sites — the inline boxShadow beats the hover:shadow-lg utility, the v29 family — while the expense action row REVEALS), the calculator row census + REAL hover (the panel 768@1280/608@640, the row 720/560 — byte-identical; the hover’s visible shadow-md layers + the action reveal), the category-level band animation pin (the v44 F1 extension — in-flight frames at the nested level). FIVE pin-sanity mutations verified to FAIL (the card inline 0.04→0.08; the grid gap-4→gap-6; the row hover:shadow-md removed; the quick-action hover:shadow-lg removed; the slate shadow-sm removed). Standing checks clean (39th) + the VLM pair on the new surface: IDENTICAL, zero flags. — 108/221/35 green (7 new e2e, pin-only TDD) | | docs/remediation-plan-v46.md | The net-worth card census + THE VALUE→FACTS SPACING FIX + the confirm-bar white-pill regression pin — the session-101 queue’s first and third candidates. The surface: the net-worth ASSET/LIABILITY cards at 1280 (the grid pitch, the card chrome, the hover state, the kebab reveal), measured live on BOTH sites (nw-card-v46.mjs + the REAL-hover driver steps) — the grid md:grid-cols-2 lg:grid-cols-3 gap-4 960 wide / 3 cols × 309.328 / gap 16 (the SAME family on both sites), the card chrome byte-identical (309/pad 20/radius 12/border 1px rgb(229,231,227)/rest shadow rgba(0,0,0,0.04) 0px 2px 8px 0px/transition all 0.2s; the title row gap 8/dot 8×8 rgb(143,188,63)/16-600; the type badge; the facts 12px rgb(107,114,128); the kebab 36×36 opacity 0), the card’s REAL hover INERT on BOTH sites (the inline boxShadow beats the hover:shadow-lg utility, the v29 family) while the kebab REVEALS. THE v46 PARITY FIX (F1): the value→facts spacing — the reference wraps its value in mb-4 (16px) while the clone’s bare mb-2 P rendered 8px (the card h 186 vs the reference’s 194; the item-card’s own mb-4 wrap the in-repo precedent — the net-worth cards were the one oversight): mb-2→mb-4 on both cards’ value P. F3: the confirm-bar white-pill pin (the clone’s superset chrome — the reference’s net-worth DELETE is DIRECT, no confirm bars, the v18 superset-fix family): the bar chrome + the Cancel pill’s shadow-sm VISIBLE layer rgba(0,0,0,0.05) 0px 1px 2px 0px (the --shadow-sm theme pin’s regression net, the v45 net’s fifth site). FOUR pin-sanity mutations verified to FAIL (the reverted mb-4; the pill’s shadow-sm removed; the card inline 0.04→0.08; the kebab’s group-hover reveal removed). Standing checks clean (40th) + the VLM pair on the /networth surface: IDENTICAL, zero flags. — 108/223/35 green (2 new e2e, 2-line production fix, TDD) | | docs/remediation-plan-v47.md | The dialog-input focus composite + the net-worth card MOBILE census + the mobile sheet’s shadow-lg — a PIN-ONLY iteration (3 new e2e, TDD with four pin-sanity mutations; NO production-code change — the measured chrome was already at parity on every VISIBLE layer, the pins lock it). The surface (the session-101 queue’s second item + the v46 follow-up + the session-104 sheet item): the dialog inputs’ focus-state shadow composition — the REST shadow-sm visible layer + under a REAL focus the 1px rgb(10,10,10) ring with the rest shadow persisting beneath it (the reference’s 3-layer v3 composite vs the clone’s 5-layer v4 construction — the same two VISIBLE layers, the leads paint nothing, the accepted v10/v25/v45 class); the net-worth card’s MOBILE census at 390 — the card internal rhythm at BYTE-PARITY post-v46-fix (358×194, the value mb 16, the facts 12px rgb(107,114,128) — the fix’s rhythm survives the mobile reflow; the reference’s own card is 194 tall inside its 432-wide OVERFLOWING grid, its documented bug); the mobile sheet’s shadow-lg census — both v3 pair layers (rgba(0,0,0,0.1) 0px 10px 15px -3px + 0px 4px 6px -4px) + the bg/geometry/overlay re-verified (the shadow-scale regression net’s SIXTH site). FOUR pin-sanity mutations verified to FAIL (the input’s ring-1 removed; the input’s shadow-sm removed; the value mb-4 reverted; the sheet’s shadow-lg removed — each needs the standalone REBUILD first, the e2e-runs-against-the-build lesson). Standing checks clean (41st) + the VLM pair on the mobile /income surface at 390: 3 flags, all DOM-verified as data/platform classes (the richer seed’s 4th Recurring badge wrapping in the byte-identical flex-wrap container + the reference’s own Base44 FAB). — 108/226/35 green (3 new e2e, pin-only, TDD) | | docs/remediation-plan-v48.md | The item-details sheet’s shadow-2xl + the dialog-input focus composite at 390 + the empty-state full census — a PIN-ONLY iteration (3 new e2e, TDD with four pin-sanity mutations; NO production-code change; the v45-family’s fourth pin-only iteration). The surface (the session-108 queue #1 + #4 + #3): the details panel’s shadow census at 1280 — the 512×680 (85vh-capped) panel at radius 16 with the v3 shadow-2xl VISIBLE layer rgba(0,0,0,0.25) 0px 25px 50px -12px (the reference’s 3-slot vs the clone’s 5-slot — the accepted lead-divergence class; the shadow-scale net’s SEVENTH site) + the forest-scrim overlay rgba(26,58,46,0.5)/z 50/blur(8px) + the 36×36 sr-only-named X; the dialog input’s focus composite at 390 — the v47 F1’s MOBILE follow-up: the input w 310 on BOTH sites with the same visible layers (the rest shadow-sm + under REAL focus the 1px ring with the rest shadow persisting, the border unchanged, the 5-slot counts) — the family is viewport-independent; the empty-state FULL census via filter-to-zero at 390 — the bare block at BYTE-PARITY (the 358×328 py-16 text-center container, the 64px icon at opacity 0.2/mb 16, the H3 20px/600/forest/mb 8, the P 16px/gray/mb 24, the 36px gradient Add button + v3’s BARE two-layer shadow). FOUR pin-sanity mutations verified to FAIL (the panel’s shadow-2xl removed; the input’s ring-1 removed — shared class, also v47’s F1; the icon mb-4 reverted — the container h 328→320; the .zb-btn-add shadow removed). Standing checks clean (42nd) + the VLM pair on /savings at 1280: 4 flags, all DOM-verified as data classes (the richer seed’s 2-card grid + the Recurring/payment-method rows + the user’s avatar initial; the badge chrome byte-identical — need’s red chip = the clone’s same map). — 108/229/35 green (3 new e2e, pin-only, TDD) | | docs/remediation-plan-v49.md | The login panel’s mobile-only   spacer (THE FIX — the first production-code change since v46) + the donut legend’s mobile reflow + the calculator dialog’s 640 chrome (two pin-only surfaces) (3 new e2e, TDD with six pin-sanity mutations). THE v49 FIX (F1): the reference’s login wrapper carries a SECOND child at mobile — (h 16, mt 32, #94a3b8) below the panel — lifting its card stack 24px above the geometric center at 390 (the panel y 57; the pre-fix clone centered alone at y 81, its wrapper's ONLY child the panel). The v12/v13 login pins were desktop-measured where `sm:hidden` hides the spacer — first-time measurement at 390 caught it (verified on the reference's sign-in AND sign-up states; the 640 contract re-verified live: display none, the panel re-centers to y 111). The fix pins the color as the `text-[#94a3b8]` hex (the v4 named-slate oklab trap — the v7 login-family rule; the   never renders it). F2 — the donut legend's MOBILE reflow at 390 (the v44 F3 measured both viewports but pinned only 1280): the container flex/column/gap 12/mt 24, the row 64/radius 8/pad 12/rgb(245,248,245), the dot 16/radius 4/classification color, the leaves 16-500-forest / 16-700-classified / 12-400-gray, the right-aligned values + the 16px icon — byte-identical (the reference's row w 313 rides its own 395 docSW overflow — the documented superset-fix-#4; the clone's 308 fits at 390). F3 — the calculator panel's CHROME census at 640 (the read-only open flow: hover → Calculate → census → Close): w 608/x 16/radius 16/maxW 768px/maxH 816px (85vh)/bg white/overflow hidden, the 2xl VISIBLE layer `rgba(0,0,0,0.25) 0px 25px 50px -12px` (the clone's `.zb-modal-panel` emits the visible layer directly — no leads, the accepted lead-divergence class), the 36×36 close, the header border-b 1px rgb(229,231,227), the centering invariant, sw 640. SIX pin-sanity mutations verified to FAIL (the spacer removed [the TDD RED run]; mt-8→mt-4 [wrapper 730→722]; sm:hidden removed [the 640 sub-check]; the legend row p-3→p-2 [pad 12→8, h 64→56]; the zb-modal-panel shadow removed; the calculator's 85vh→90vh [maxH 816→864] — each with the standalone REBUILD). Standing checks clean (43rd — the sheet-trap re-verified with the precise overlay census) + the VLM pair on the LOGIN page at 390 (THE FIXED SURFACE): **IDENTICAL** — the only flag the reference's own "Edit with Base44" FAB. — 108/232/35 green (3 new e2e, 1 production fix, TDD) | | [`docs/remediation-plan-v50.md`](docs/remediation-plan-v50.md) | **The auth states' panel heights (THE FIX — the second production-code change since v46) + the networth tablist's 640 chrome + the dashboard hero's 390 reflow (two pin-only surfaces)** (4 new e2e, TDD with five pin-sanity mutations). THE v50 FIX (F1, `tests/e2e/login-parity.spec.ts`): the clone's sign-up and forgot panels were 16px SHORT at every viewport (406 vs the reference's 422; 294 vs 310) — the v4 space-y trap-log rule 4 violation: the back button's OWN `-mb-2` (the reference's idiom) OVERRIDES v4's zero-specificity `:where(& > :not(:last-child))` margin-block-end where the reference's v3 build STACKS them (the collapsed 16 + (-8) = the 8px net gap — the pre-fix heading overlapped the button by 8px). The fix: the scoped `.zb-auth-mode` restore (the SPACE-Y V3 SEMANTICS family's third member, v9 G3 / v24 G2) — the container DROPS the space-y class (no zeroing rule: a `> :not(:last-child)` zero at 0-2-0 specificity would beat the -mb-2 utility's 0-1-0 and kill the 8px collapse — the first attempt's measured 430) + the forgot's `.zb-sm6` sm tier (the reference's `sm:space-y-6`, 24px at >=640 — read from the reference's own DOM class). The pins: the sign-up 422/wrapper 470/y 187 + the forgot 310/358/y 243 + the button→heading net gap 8 + the 640 sm-tier sub-check. F2 — the networth tablist's CHROME census at 640: byte-parity (the list 448/36/grid 220px 220px/#f5f5f5/8px/4px; the triggers 220×28/6px/14px/500; the active green pair with the shadow-sm VISIBLE pair 0.1 1px 3px / 0.1 1px 2px -1px — the clone's 4 transparent leads vs the reference's 2, the accepted lead-divergence class; the inactive #737373 pair; the 16×16 currentColor icons). F3 — the dashboard hero's MOBILE reflow at 390: byte-parity (the 1-col/gap-24 grid; the goal card 358×316@y 253 + the breakdown 358×324@y 593 — the 24px rhythm; the icon circle 48/rgba(143,188,63,0.2) + the 24px target icon; the H3 18/700/white/lh 28; the sublabel/allocation/pct families; the track 12px/0.1/9999px; the fill's 90deg orange status gradient; the balance box 12px/24px/0.05/blur(10px); the figure 30/700/white; the chip's flex gap-8 row with the 16/600 rgb(245,169,98) pair — BOTH sites' 20px svgs squish identically under the row's flex-shrink, a byte-shared data-pressure artifact; the reference's card 363 rides its own 395 overflow, the clone's 358 fits). FIVE pin-sanity mutations verified to FAIL (the zb-auth-mode removed [the TDD RED]; the -mb-2 removed [wrapper 478 = +8]; the zb-sm6 removed [mt 24→16]; the active-tab bg removed [F2]; the hero track's h-3→h-2 [F3]) — each with the standalone REBUILD. Standing checks clean (44th — the sheet-trap re-verified with the precise overlay census) + the VLM pair on the login SIGN-UP state at 390 (THE FIXED SURFACE): the 4 flagged diffs all DOM-verified as noise (the panel y 187 / h 422 / the button 294@x 48 IDENTICAL on both sites — the v43 misattribution lesson's fifth application). — 108/236/35 green (4 new e2e, 1 production fix, TDD) | | [`docs/remediation-plan-v51.md`](docs/remediation-plan-v51.md) | **The verify-email mode container's missing `.zb-auth-mode` conversion (THE FIX — the third production-code change since v46, completing the v50 F1's claimed-but-never-applied third site) + four 390 pin surfaces (the stats row, the savings goal card, the Budget Guidelines card, the quick-actions row)** (5 new e2e, TDD with five pin-sanity mutations). THE v51 FIX (F1, `tests/e2e/verify-email.spec.ts`): the v50 plan/commit/docs all claimed the verify-email branch's container was converted — the diff landed the comment but NOT the line-608 class swap; the verify panel was 16px short (492 vs 508) with the junction net gap -8 (the centered inner stack overlapping the back button by 8px — the same v4 space-y × `-mb-2` override). The one-class fix (`space-y-4` → `zb-auth-mode`, no sm tier); the pins: the panel 508/wrapper 556/y 144 + the net gap 8 (the panel height is email-length-dependent — a fixed 21-char address keeps the pins deterministic). F2 — the stats row at 390: byte-parity (the 1-col/24px-gap grid; the three 210-tall cards @ y 949/1183/1417 — the rhythm byte-identical on both sites; the radius-16/pad-24 card; the 48px icon circle; the 30/700/mb-16 amount). F3 — the savings view at 390: byte-parity (the goal card radius 12/pad 20/the rgba(0,0,0,0.04) shadow; the H4 16/600; the amount 24/700 in the savings blue; the 8×8 dot; the H1 30/700 + the filter card + the search 308×36 — the card height is a data class, never pinned: the richer seed's 4-chip wrap). F4 — the Budget Guidelines card at 390: byte-parity (the 380-tall card at radius 12; the BLOCK tiles stack; the three tiles h 94/94/78 with the byte-identical tinted pairs). F5 — the quick-actions row at 390: byte-parity (the three 90-tall cards @ y 2751/2865/2979 — the 114px rhythm; the 40px icon circle; the trailing plus at opacity 0.5 — the reference's own class string read from its DOM). FIVE pin-sanity mutations verified to FAIL (the zb-auth-mode removed [the TDD RED]; the rounded-2xl→rounded-xl; the goal card's shadow removed; the Needs tile's bg removed; the circle's w-10 h-10→w-8 h-8) — each with the standalone REBUILD. Standing checks clean (45th) + the VLM pair on the /savings view at 390 (the F3 surface): the 3 flagged diffs all DOM-verified as the documented seed-data classes (the v48 precedent). — 108/241/35 green (5 new e2e, 1 production fix, TDD) | | [`docs/remediation-plan-v52.md`](docs/remediation-plan-v52.md) | **The details sheet's missing subcategory line (THE FIX — the fourth production-code change since v46) + the /income and /expenses 390 pins + the details-sheet internals pin + the standing script's burger-selector probe-infra fix** (4 new e2e, TDD with five pin-sanity mutations). THE v52 FIX (F1, `tests/e2e/item-details.spec.ts`): the reference's details-sheet summary renders FIVE elements — the chip row → the H3 name → **the SUBCATEGORY P (`text-lg`, 18/400/rgb(107,114,128))** → the amount (measured live: Rent → "Apartment", Salary → "Main Job") — while the clone rendered four (the H3 → the amount directly): the `subcategory` field was always on the model (the card renders it, the edit dialog edits it) but the details view never wired it. The one-insert fix (between the H3 and the amount, conditional on the non-empty value); the panel stays capped at its 85vh max-height — no geometry pin affected. F2 — the /income view at 390: byte-parity (the header row 358×112 flex-col/gap 16/mb 32; the 48×48 gradient chip; the H1 30/700; the Add button 147×36; the filter card + the 1-col 3-child grid; the search 308×36; the first card's chrome; docSW 390). F3 — the /expenses view at 390: byte-parity (the 155×36 orange-gradient Add button; the 4-child/3-trigger filter grid; the first card's chrome; docSW 390). F4 — the details sheet's INTERNALS at 390: byte-parity (the sticky header h 69; the title 18/700; the 36×36 close; the chip row gap 8/mb 12; the classification block; the fact rows gap 12/py 8; the meta footer grid-cols-2). P1 — the standing script's R1 burger selector fixed (the reference's `aria-label` idiom silently failed on the clone's sr-only-named burger — R1 read sheetLinks:0, a false red on a working surface; the fixed text-walk reads 5). FIVE mutations verified (the P removed [the TDD RED]; the chip gradient; the filter p-6→p-4 [both F2+F3]; the header py-4→py-2; the H1 text-3xl→text-2xl [both]). Standing checks clean (46th) + the VLM pair on /expenses at 390 (the F3 surface): 4 flagged diffs all DOM-verified as the documented seed-data/platform classes. — 108/245/35 green (4 new e2e, 1 production fix, TDD) | | [`docs/remediation-plan-v53.md`](docs/remediation-plan-v53.md) | **The /networth tablist's mobile pin + the calculator dialog's mobile pin (PIN-ONLY — the second consecutive pin-only iteration; no production-code change)** (2 new e2e, TDD with five pin-sanity mutations; the session-124 v53 queue's read-only members: the /networth mobile tablist census + the calculator dialog 390 reflow). F1 (`tests/e2e/networth.spec.ts`) — the TABLIST at 390: byte-parity on every chrome axis (the list h 36/grid/bg #f5f5f5/radius 8/pad 4/maxW 448; the ACTIVE trigger 175×28/radius 6/14/500/#14532d on #dcfce7 + the shadow-sm VISIBLE pair + the 16×16 currentColor icon; the INACTIVE 175×28/#737373/transparent/none) — the WIDTH family rides the reference's own documented 464-overflow bug (its list renders 432 wide inside the overflowing page, overhanging the 390 viewport; the clone's superset fix keeps the page in-bounds at 358/pageSW 390 — the v47 net-worth card mobile-census divergence class). F2 (`tests/e2e/calculator.spec.ts`) — the CALCULATOR DIALOG at 390 (the v45 F2b add-then-cleanup flow): byte-parity (the panel 358 @ x 16/radius 16/maxW 768/maxH 717.4/white/overflow hidden/flex + the 2xl VISIBLE layer + the CENTERING invariant; the header h 101/pad 16 24/border-b; the chip 32×40 + the close 29×36 — the identical flex-shrink squish at 390 on BOTH sites; the total card 310/#fff7f5/radius 12/pad 16; the row 310/pad 16/white/border-t/radius 12; the Add Item button 110×32/12/500/gradient/white; docSW 390 — the panel h 413 with 1 row on BOTH sites, data-dependent, never pinned). FIVE mutations verified (the grid-cols-2→grid-cols-1; the active bg utility removed; the header py-4→py-2; the panel maxWidth 48rem→40rem; the total card p-4→p-2). Standing checks clean (47th) + the VLM pair on the calculator at 390 (the F2 surface): 5 flagged diffs all DOM-verified (the close-button "boxed X" = the probe's synthetic-click focus artifact — the outline:auto renders only under :focus-visible, the keyboard a11y superset; the reference's outline-style:none is its own a11y regression; no ring in real mouse flows; the rest = the seed-data/platform classes). — 108/247/35 green (2 new e2e, 0 production changes, TDD) | | [`docs/remediation-plan-v54.md`](docs/remediation-plan-v54.md) | **The savings goal-card's badge-row + footer internals at 390 (PIN-ONLY — the third consecutive pin-only iteration; no production-code change)** (1 new e2e, TDD with five pin-sanity mutations; the session-125 v54 queue's first read-only member: the goal-card internals BEYOND the v51 F3 chrome pin). F1 (`tests/e2e/items.spec.ts`) — the savings view's goal-card INTERNALS at 390: byte-parity on every chrome axis (the header row flex/gap 8/mb 4/center; the dot 8×8/9999/rgb(59,126,161) — the savings TYPE accent on BOTH sites, the reference colors by TYPE not classification; the BADGE ROW flex/wrap/gap 8/mb 12; every CHIP h 22/radius 6/pad 2px 10px/12/600 with the 12×12 icons at mr 4 — the "monthly" chip w 78 and the "active" chip w 63 on BOTH sites, the byte-parity width proof; the FOOTER flex/space-between/12/rgb(107,114,128)/h 16 + the date span flex/gap 4/12 with the 12×12 calendar icon) — the chip COUNT (the reference's 3 vs the seed's 4 — the Recurring chip), the row HEIGHT (22 vs 52, the wrap), and the payment span (the reference's goal carries none) are the documented DATA classes. FIVE mutations verified (the badge row gap-2→gap-4; the chip px-2.5→px-1.5; the chip text-xs→text-sm; the footer text-xs→text-sm; the date span gap-1→gap-2). Standing checks clean (48th) + the VLM pair on the /networth tablist (the v53 F1 surface — the session-125 standing-check suggestion) AND the savings goal card (the F1 surface): every flag DOM-verified (the active tab bg byte-identical rgb(220,252,231) — a screenshot-perception artifact; the Add button's full text on both sites, no truncation — a small-text misread; the trend icon = the documented 464-overflow width class; the rest = the seed-data/platform classes). The v54 probe lessons: a generic justify-between+svg footer finder matches the card's HEADER row first (the header also carries justify-between + the kebab svg) — the footer locator is "the LAST justify-between direct child WITHOUT an h4"; the chips' inline-flex SOURCE class computes to "flex" as flex items (CSS blockification) — the census finder must accept "flex". — 108/248/35 green (1 new e2e, 0 production changes, TDD) | | [`docs/remediation-plan-v55.md`](docs/remediation-plan-v55.md) | **The ICON-GLYPH swap (THE FIX — the fifth production-code change since v46, breaking the v53/v54 pin-only streak)** (6 new e2e, TDD RED-first with five pin-sanity mutations; the session-128 v55 queue's goal-card classification-icon glyph census, extended to a FULL icon-glyph audit). The audit (`glyph-audit-v55.sh` + `glyph-diff-v55.mjs`, 24 lucide glyphs across the five views + the login page, measured live on BOTH sites): 19 of 24 shared glyphs byte-identical, **4 DRIFTED** — lucide-react 0.525 redesigned **piggy-bank / trending-up / search / mail** since the reference's lucide version (the v44 chevron-swap class: parity is the SHAPE DATA, which a computed-style census cannot see). F1 — `src/components/ui/legacy-icons.tsx`: four components rendering the reference's EXACT svg (base attrs + path/polyline/rect/circle children byte-identical to the live reference DOM, the `lucide lucide-` class prefix kept — the e2e pins select `svg.lucide-*`), swapped into all eight usage files (item-card / items-view / dashboard-view / item-details-dialog / budget-item-dialog / sidebar / net-worth-view / login-card). The reference renders the piggy's OLD 3-path glyph on EVERY surface (piggy-sweep-v55.sh: the nav rail, the stat row, the donut legend, the quick actions, the hero, the chips) and the trending-up as TWO POLYLINES. The e2e pin (`tests/e2e/glyph-parity.spec.ts`): the goal card's piggy/calendar/repeat at 390, every dashboard + networth piggy + trending, the search handle (4.3 not 4.34), the login mail flap (a1.94 1.94) + lock, and the /expenses heart + circle-alert same-glyph control. SIX mutations verified (the body path, the polyline points, the search handle, the mail flap, a usage-site revert). Standing checks clean (49th) + the VLM pair (the calculator with-row — the v53 F2 surface + the goal card post-swap): CLEAN per the v43 rule (the X-close 29×36 transparent byte-identical, the amounts 24/700 on both; the rest = the documented seed-data classes). Probe lessons: a path-only glyph census reads ZERO paths on circle-alert (walk every shape tag, read `points` for polylines); the glyph audit's login can RACE the page mount — 'no inputs' is the RETRY signal + an explicit post-login auth check. — 108/254/35 green (6 new e2e, TDD) | | [`docs/remediation-plan-v56.md`](docs/remediation-plan-v56.md) | **The 1280 desktop VLM standing pair + the glyph-swap HOLD re-audit (PIN-ONLY — no production-code change, the v53/v54 line's return to pin-only after the v55 FIX)** (1 new e2e + three pin-sanity mutations; the session-131 v56 queue's own suggestion — a fresh VLM pair on the v55 F1 surfaces' desktop 1280 state). The glyph re-audit (`glyph-audit-v56.sh`, the polyline/rect-rx capture fix applied): 24 shared glyph names, **0 DIFFS among the shared set** — the v55 legacy-icon swap HOLDS byte-identical; no new drift. The desktop 1280x800 pair: PAIR 2 (the networth trending surfaces) VERDICT IDENTICAL; PAIR 1 (the dashboard) flagged 4, every one DOM-arbitrated clean (the v43 rule) — the nav highlight is the documented root-URL active-nav gap (a reference bug the clone fixes; at /networth BOTH sites render the byte-identical forest-gradient active style), the progress-bar length is data-driven (the track byte-identical 568x12 on both sites; the fill = pct x 568 per each site's data), the avatar the user-identity class, the FAB the platform chrome. The v56 pin (`tests/e2e/dashboard.spec.ts`, +1 → 255): the hero-card DESKTOP 1280 census (`hero-desk-v56.sh`, byte-identical on both sites) — the 304px triple-column hero grid + the [632x364, 304x364] children at y 136, the goal card 632x316/radius 16/pad 32/forest gradient + shadow, the 568x12 track, the fill's pct-proportional width arithmetic, the 48px stat tiles. Probe lessons: branch the glyph census per shape tag (the v55 polyline fix); the reference has no landmark — nav probes scan ALL anchors; split long shot scripts per-view when the timeout bites. — 108/255/35 green (1 new e2e, pin-only) | | [`docs/remediation-plan-v57.md`](docs/remediation-plan-v57.md) | **The items-view desktop 1280 VLM standing pair + the item-card footer + badge-row census (PIN-ONLY — no production-code change, the pin-only line's continuation) + the not-found G7 flake fix (TEST-ONLY)** (2 new e2e + three pin-sanity mutations; the session-134 v57 queue's own suggestion — a fresh VLM pair on the ITEMS view's desktop 1280 state, the v56 pair having covered the dashboard + networth). The pair (income + expenses): every flag DOM-arbitrated to the documented data classes — the avatar; the seed data; the payment-method chips (DATA-CONDITIONAL: the footer row's chrome byte-identical on both sites [flex justify-between 12px rgb(107,114,128) h 16, each span flex gap 4px with the 12×12 gray icon] — the reference's own income card renders the same two-child structure [the date + "Bank Account"] while its expense items carry no payment methods); the badge wrap (the row's chrome byte-identical [flex wrap 8px mb 12px, every badge flex center/6px/2px 10px/12px/600/h 22], the clone's Recurring superset badge wraps the row to h 52). The v57 pin (`tests/e2e/items.spec.ts`, +2 → 257): F1a the /income card's footer two-children conditional render + the badge chrome; F1b the /expenses grid census + the wrapped 4-badge row. THREE pin-sanity mutations verified to FAIL (the footer text-xs→text-sm, the badge-row gap-2→gap-4, the badge px-2.5→px-2). F3 (test-only): the not-found G7 "Go Home" flake fix — the logged-out opt-out raced the /login redirect's ~50-180ms transient dashboard render (a genuine ~1-in-8 failure by --repeat-each, not load); the test carries its own describe-scoped storageState now. The 51st standing checks clean (the mobile menu verified working; the reference's four bugs all live). — 108/257/35 green (2 new e2e, pin-only + 1 test-only fix) | | [`docs/remediation-plan-v59.md`](docs/remediation-plan-v59.md) | **The calculator-view desktop 1280 VLM standing pair + the calculator panel/total-card census (PIN-ONLY — no production-code change; the ninth iteration since v46, the pin-only line's continuation)** (1 new e2e + three pin-sanity mutations; the session-139 v59 queue's first TWO items — the calculator dialog's 1280x800 pair [the panel + row surfaces; the v45 F2a pin covered the panelW + row census, the v49 F3 panel chrome at 640 only] and the login view's 1280 pair [the v55 glyph pin's surface]). The calculator pair: both VLM flags DOM-arbitrated — the total card's "Will update category total" span (DATA-CONDITIONAL: the lineTotal-vs-parentAmount-at-open contract — the reference's own Rent $25 === its line-item sum so hidden, the clone's $1850 parent showed it) and the row's hover-action pair (POINTER-POSITION artifact: both sites gate it at opacity 0 at rest — the reference's shot caught it mid-hover); the census byte-identical on every axis (the panel 768/393 @ 256,204 — the centering invariant / radius 16 / maxW 768px / maxH 680px [85vh of 800] / the 2xl VISIBLE layer / the total card 720×90 #fff7f5 radius 12 pad 16 / the close 36×36 / the header border-b / the row 720 pad 16 radius 12 — the v45 F2a values re-confirmed live). The login pair: IDENTICAL (the only flag the reference's Base44 platform badge). The v59 pin (`tests/e2e/calculator.spec.ts`, +1 → 259): the calculator panel + total-card census at 1280 — the read-only open state. THREE pin-sanity mutations verified to FAIL (the total card bg #fff7f5→#fff8f0, the panel maxWidth 48rem→42rem, the total card p-4→p-5). The 53rd standing checks clean (the mobile menu verified working — the 53rd time; the reference's four bugs all live). The 16 screenshots: 9 byte-identical + 7 diff the SEED-DATE class (isoDate(-N) anchors to the seed-run date — VLM-arbitrated DATE-TEXT-only, zero layout diffs). — 108/259/35 green | | [`docs/remediation-plan-v60.md`](docs/remediation-plan-v60.md) | **The networth-view desktop 1280 VLM standing pair + the dashboard drill-down EXPANDED-state pair (PIN-ONLY — no production-code change; the tenth iteration since v46, the pin-only line's continuation)** (2 new e2e + three pin-sanity mutations; the session-141 v60 queue's first TWO items — the NETWORTH view's dedicated 1280x800 pair [the standing-pair program's remaining major view — the v56 PAIR 2 was the glyph-swap audit pass; the v52-era pins cover the 390/640 chrome] and the DASHBOARD's drill-down EXPANDED surfaces at 1280 [the v43 F3 family pinned the sub/item rows at 640 only; the v44 F2b 1280 pin walked the section/category/footer heights] — both FULL-PAGE shots, the below-the-fold surfaces) + the 54th standing checks (the expenses census now in the standing script). The networth pair: **VERDICT IDENTICAL** (zero layout/style flags; every difference the expected data class — the figures, the groups, the ratio "∞:1" vs "0.21:1"); the census byte-identical on every axis (the summary card 960/32px/16px/the 135deg forest→lime gradient/the rgba(26,58,46,0.3) shadow; the figure 48px/700; the medallion 64×64; the stat grid 440px 440px gap 16; the tile 16px/12px/white-10%/blur(10px); the ratio value 18px/700; the tablist 448/36/"220px 220px"; the group header 18px/600/forest-medium; the grid 960/3×309.328/gap 16 + the card 309/20px/12px — the v46 F2 values re-confirmed live). The drill-down pair: FIVE flags, every one DOM-arbitrated — the nav highlight (the root-URL active-nav gap), the avatar (user-identity), the Base44 tooltip (platform chrome), the hero progress fill 30.5% vs 62.8% (the pct × 568 data class), and the donut percentage labels (a PERCEPTION ARTIFACT — the reference DOES render them: 3 texts/labelLineCount 0/16px/black, the v9 G1b measurement re-confirmed live); the drill census byte-identical on every axis (the card 304; collapsed 364 BOTH; the section/category/sub/item/footer chrome; the heights 364→445→499 [reference] vs 364→481→535 [clone] = the DATA class). The v60 pins: the networth summary-card + tablist census at 1280 (`tests/e2e/networth.spec.ts`) + the drill-down sub/item-row census at 1280 (`tests/e2e/dashboard.spec.ts` — the finder pins the REAL rows by chrome, never rows[last]: the itemdump lesson). THREE pin-sanity mutations verified to FAIL (the summary card p-8→p-10, the tablist max-w-md→max-w-lg [448→512], the item label max-w-[120px]→max-w-[140px]). The 54th standing checks clean (the mobile menu verified working — the 54th time; the reference's four bugs all live). The 16 screenshots: 15 byte-identical + 1 diff the DATE class (07-add-item-modal's Date input defaults to the capture-run day — VLM-arbitrated DATE-TEXT-only). — 108/261/35 green | | [`docs/remediation-plan-v61.md`](docs/remediation-plan-v61.md) | **The Budget Guidelines card's STRETCH fix — the FIRST production-code change since v55 (the pin-only line broken after six consecutive iterations) — + the quick-action row + the Budget Guidelines card's 1280 census pins** (2 new e2e + three pin-sanity mutations; the session-144 v61 queue's first TWO items — the dashboard's below-the-fold FULL-PAGE 1280x800 pair + the DOM census both sides). The VLM pair verdict IDENTICAL — but the DOM census caught what the VLM missed: the reference's Budget Guidelines card sits at NATURAL height (348) inside a stretched `space-y-4` wrapper (the grid child; the row 656 under the seed) — the pre-v61 clone's card ITSELF stretched to 656 with ~308px of empty white card body below the Savings tile. Arbitrated three ways (the census both sides; the REGION-CROP VLM re-arbitration — VERDICT DIFFERENT pre-fix / SAME post-fix; the reference's own row DOM measured live) and fixed by mirroring the reference's DOM (the card wrapped in). TDD: the pin RED (cardH Expected 348, Received 656) → the fix → GREEN. The quick-action row + the guidelines chrome censused byte-identical on every axis (the grid 960/3×304/gap 24; the cards 304×90/16px/24px; the tile 40×40/12px/type tints; the label 16px/500/forestDark via the LEAF-SPAN finder — the v61 label lesson; the three GUIDELINE_ROWS tiles byte-identical). The 55th standing checks clean (the mobile menu verified working — the 55th time; the reference’s four bugs all live). The 16 screenshots byte-identical (the fix’s effect lives below every viewport shot’s fold). — 108/263/35 green | | docs/remediation-plan-v62.md | The reference-side a11y surface re-audit (NO DRIFT since v38) + the mobile header-subtitle 390 census pin + the SKILL doc count-line sweep (PIN-ONLY — no production-code change; the pin-only line resumes) (1 new e2e + one pin-sanity mutation; the session-147 v62 queue first TWO items — the reference axe-tree re-audit deferred from v61, and the four stale SKILL count lines). The a11y re-audit (axe-detail-ref-v62.mjs, the full a11y.spec.ts surface: six authed routes at 1280 + the 404 + the login + the mobile 390 sheet-open state): NO DRIFT — the same two structural families (button-name 10 nodes + svg-img-alt 6), the identical 7-color contrast palette, the login still zero-violation; the clone’s a11y gates stay valid as pinned. The TWO-pair VLM program (desktop 1280 full-page + the NEW mobile 390 full-page twin): every flag DOM-arbitrated to documented classes — the mobile pair’s “subtitle wraps to a third line” flag was a PERCEPTION ARTIFACT (the leaf-text census: both sides 2 lines / h 48 / block h 92 / top 121 byte-identical; the v61 meta-lesson’s inverse — the census ACQUITTED what the VLM convicted). The v62 pin: the mobile header-subtitle block census at 390 (the P 48/2×24px lines/16px/400/gray-500 + the lineRhythm contract + the block 92 = h1 36 + mb-2 8 + p 48). The F3 docs sweep: the SKILL doc’s stale count lines 214/223-era corrected to the current counts. The 56th standing checks clean (the mobile menu verified working — the 56th time; the reference’s four bugs all live; the data census intact). The 16 screenshots: 10 byte-identical + 6 VLM-arbitrated (5 SEED-DATE + 1 sub-perceptual AA-timing). — 108/264/35 green | | docs/remediation-plan-v63.md | The /networth mobile 390 full-page pair (the standing-pair matrix COMPLETED) + the networth summary card’s stat-tile family census pin + the axe-core version-drift re-check (PIN-ONLY — no production-code change; the pin-only line continues) (1 new e2e + three pin-sanity mutations; the session-149/151 v63 queue’s two items — the last un-paired viewport × view combination, and the axe toolchain drift watch). The pair matrix is now COMPLETE — {dashboard, items, savings, calculator, networth} × {1280 desktop, 390 mobile} all paired. The /networth 390 pair: VERDICT DIFFERENT with THREE flags, ALL DOM-arbitrated via arbit-v63.sh to documented classes — the stat-tile stacking (the v41 mobile-superset class: the reference’s 432-wide card in its own 464 overflow renders 2-col 176×84 tiles; the clone’s 358-fit card stacks 1-col 294×84), the card height (the arithmetic consequence: 333 + 84 + 16 = 421), and the trend-icon “larger” (a PERCEPTION ARTIFACT of the v54 clipping class — the medallion 64×64 + icon 32×32 BYTE-IDENTICAL both sides; the reference’s clips the 390 viewport at right 416, the clone’s fits at 342). The axe-core version-drift re-check: NO DRIFT — @axe-core/[email protected] is still npm-latest and its NESTED engine [email protected] is byte-identical to the v38/v62 audits (the top-level 4.14.0 belongs to eslint-plugin-jsx-a11y — lint-time only, never scans pages). The v63 pin: the stat-tile family census at 390 (tests/e2e/networth.spec.ts — the stat grid {cols “294px”, gap “16px”}, the tiles 294×84 with the translucent-blur chrome, the card h 421 stacking arithmetic, the medallion’s inner 32×32 icon). THREE pin-sanity mutations verified to FAIL (the grid gap-4→gap-2; the tile p-4→p-2; the medallion svg h-8→h-6). The 57th standing checks clean (the mobile menu verified working — the 57th time; the reference’s four bugs all live; the data census intact). The 16 screenshots BYTE-IDENTICAL. — 108/265/35 green | | docs/remediation-plan-v64.md | The OLDEST desktop pair’s fresh standing sweep (the v57-era items pair re-run) + the axe-core 5.x wrapper watch (NO DRIFT) + the income filter-card computed census pin (PIN-ONLY — no production-code change; the pin-only line continues) (1 new e2e + three pin-sanity mutations; the session-153 v64 queue’s first two items — the re-run cadence on the OLDEST desktop pairs, and the axe 5.x wrapper watch). The fresh items pair (income + expenses, FULL-PAGE at 1280×800): PAIR 1’s flags the documented avatar/FAB/seed classes; PAIR 2’s three style flags ALL DOM-arbitrated via arbit-v64.sh to documented classes — the card footer content (the v57 DATA-CONDITIONAL class re-measured: the footer chrome BYTE-IDENTICAL [flex justify-between 12px rgb(107,114,128) h 16; every child span flex gap 4px + the 12×12 icon]; the reference’s expense footers render the date alone — its items carry no paymentMethod; the clone’s render the SAME two-child structure the reference’s own income card renders [“Oct 7, 2026” + “Bank Account”]), the badge wrap (the v57 class re-measured: the chrome BYTE-IDENTICAL; the clone’s Recurring superset badge wraps h 52 = 22+8+22), and the grid population (the seed-count class: the axes BYTE-IDENTICAL [960/16px/the template “309.328px 309.328px 309.344px” — the same fractional last track both sides]; the reference’s 4 cards vs the clone’s 3) — plus the card heights (200 vs 230 = the badge-wrap arithmetic data class). The pair is CLEAN. The axe-core 5.x wrapper watch: NO DRIFT — npm-latest is still 4.13.0 (no 5.x exists), the NESTED engine 4.13.0 byte-identical to v38/v62/v63. The v64 pin: the INCOME view’s filter-card COMPUTED census at 1280 (tests/e2e/items.spec.ts — the never-census’d sibling of the v38 S2 savings pin; the early-era test pins class strings only): the card 960×86/pad 24/radius 16/white/border rgb(229,231,227)/mb 24, the search 447×36 (the 2-col span; the asymmetric 40px left pad), the inner grid 4×215.5px/16px gap/3 children, the comboboxes 216×36 at 14px. THREE pin-sanity mutations verified to FAIL (the card p-6→p-4; the search wrapper md:col-span-2 removed; the grid gap-4→gap-2). The 58th standing checks clean (the mobile menu verified working — the 58th time; the reference’s four bugs all live; the data census intact). The 16 screenshots BYTE-IDENTICAL. — 108/266/35 green | | docs/remediation-plan-v65.md | The OLDEST remaining desktop pair’s fresh standing sweep (the v58-era savings pair re-run) + the axe-core 5.x wrapper watch (NO DRIFT) + the savings page-header row computed census pin (PIN-ONLY — no production-code change; the pin-only line continues) (1 new e2e + three pin-sanity mutations; the session-155 v65 queue’s two items — the re-run cadence’s other OLDEST candidate, and the axe watch). The fresh savings pair (FULL-PAGE at 1280×800): 2 style flags + data classes, every one DOM-arbitrated via arbit-v65.sh + arbit-v65b.sh — the avatar (the user-identity class re-measured with a NEW perception lesson: the chrome BYTE-IDENTICAL [36×36/radius 9999px/14px-600/white/bg rgb(143,188,63) on BOTH sides — the same lime]; only the INITIAL differs [“U” the reference’s nameless account vs “D” the clone’s demo user]; the VLM’s “green vs lime” was a PERCEPTION ARTIFACT — identical computed colors read as different hues), the grid layout (the seed-count class: the axes BYTE-IDENTICAL [960/16px/3 tracks]; the fractional last-track distribution is CONTENT-COUNT-CONDITIONAL — the reference’s 309.328×3 at ONE child vs the clone’s 309.328/309.328/309.344 at TWO; the v64 census measured the same 309.344 tail both sides at 4-vs-3 — the axes are the contract, the subpixel tail a rendering artifact of the data class), plus the footer (the v57 DATA-CONDITIONAL class re-measured: the chrome BYTE-IDENTICAL; the reference’s item carries no paymentMethod so its footer renders the date alone), the badge wrap (the chrome BYTE-IDENTICAL; the clone’s Recurring superset badge wraps h 52 = 22+8+22), the card heights (200 vs 230 = the wrap arithmetic), and the Add button’s shadow (the reference’s two LEADING zero-alpha layers over the clone’s .zb-btn-add visible layers — the v4 computed-form class; the visible geometry BYTE-IDENTICAL). The pair is CLEAN. A DOM-structure discovery (documented, not a gap): the reference’s savings page carries a HIDDEN zero-size h1 “ZeroBalance” FIRST in its DOM (its own SEO pattern; an 18px/700 h2 logo in its sidebar) before the VISIBLE 30px/700 h1 — the clone renders the visible h1 alone, BYTE-IDENTICAL chrome. The axe watch: NO DRIFT (npm-latest still 4.13.0, no 5.x exists; the NESTED engine byte-identical to v38/v62/v63/v64). The v65 pin: the SAVINGS view’s PAGE-HEADER ROW computed census at 1280 (tests/e2e/items.spec.ts — the never-census’d family: the v42 census pinned the MID-640 STACKED state only, the v63 pin the 390 networth block, the early-era tests pin CLASS STRINGS): the row flex space-between/center/gap 16/mb 32/h 60, the chip 48×48 radius 12/blue gradient + the white 24×24 icon, the h1 30/700/lh 36/forestDark, the subtitle 16/400/gray-500, the Add button 150×36/6px/8px 16px/14px-500/white + the visible shadow layers, the 16×16 plus. THREE pin-sanity mutations verified to FAIL (the row mb-8→mb-4; the chip h-12 w-12→h-10 w-10; the h1 text-3xl→text-2xl [the row h 60→56 — the text block governs the row height]). The 59th standing checks clean (the mobile menu verified working — the 59th time; the reference’s four bugs all live; the data census intact). The 16 screenshots BYTE-IDENTICAL. — 108/267/35 green | | docs/remediation-plan-v66.md | The v59-era pair’s fresh standing sweep (the calculator dialog + login view re-run at 1280) + the axe-core 5.x wrapper watch (an upstream 4.14.0 appears — local NO DRIFT) + the calculator dialog’s HEIGHT-family computed census pin (PIN-ONLY — the pin-only line continues) (1 new e2e + three pin-sanity mutations; the session-159/160 v66 queue’s two items + the 60th standing checks). The fresh calculator pair (the clone half adds the matching “Contents Insurance” $25 line item through the real UI then FULLY restores the fixture): 2 style flags BOTH DOM-arbitrated via arbit-v66.sh + arbit-v66b.sh (the computed census on both sides, the row-HEIGHT family added) to documented classes — the row actions (the POINTER-POSITION class: opacity 0 at rest on BOTH sides; the reference’s shot caught the pair mid-reveal) and the “• Will update category total” span (the DATA-CONDITIONAL class: the reference’s natively-$25 Rent hides it, the clone’s opened-at $1850 snapshot shows it — the conditional itself the pinned contract); the login pair VERDICT IDENTICAL (only the Base44 FAB — platform chrome, excluded by rule; the census byte-for-byte). The pair is CLEAN. The never-pinned HEIGHT family measured BYTE-IDENTICAL: the row 720×102 (32+4+28+32+2 — deterministic), the name h4 16/600/forestDark/h 24/mb 4, the badge pill row flex/8px/12px/h 20, the first pill 20/65/2px 8px/9999px/12px/400, the amount P 20/700/orangeDark/h 28, the totalCard h 90. The axe watch: local NO DRIFT (the lockfile holds 4.13.0; the nested engine byte-identical; the top-level 4.14.0 still lint-time only) — BUT npm-latest is now @axe-core/[email protected] (a MINOR bump inside 4.x, NOT the guarded 5.x) — recorded as the candidate for a future DEDICATED upgrade iteration, never a drift; the next re-check point unchanged. The v66 pin (+1 → 268): the height-family census at 1280 in tests/e2e/calculator.spec.ts (riding the add-then-restoreRent fixture flow). THREE mutations verified to FAIL (the row p-4→p-2 [6 failures — the v45/v49/v53/v59 pins + the v66]; the h4 mb-1→mb-2; the totalCard p-4→p-2). The 60th standing checks clean (the mobile menu verified working — the 60th time; the reference’s four bugs all live; the data census intact — the 60th clean drift check). The 16 screenshots: 15 byte-identical + 1 the documented dialog-default DATE class (07-add-item-modal — pixel-arbitrated: a 16×22 px region exactly the date input’s day digits “15”→“16”, the calendar rolled past midnight). — 108/268/35 green | | docs/remediation-plan-v67.md | The v60-era pair’s fresh standing sweep (the networth view + the dashboard drill-down EXPANDED re-run at 1280) + the axe-core 5.x wrapper watch (local NO DRIFT — the upgrade DECISION: v68 = the dedicated upgrade iteration) + the networth summary card’s HEIGHT-family computed census pin (PIN-ONLY — the pin-only line continues) (1 new e2e + three pin-sanity mutations; the session-161/162 v67 queue’s two items + the 61st standing checks). The v60-era re-run covers BOTH of that era’s pairs (the v66 precedent): the fresh networth pair: VERDICT IDENTICAL — the VLM found ZERO layout/style differences (“pixel-perfect matches” on the nav rail, the page header, the summary card geometry/gradient, the tab bar, the card layouts, the typography, the spacing); every difference the expected DATA class + the Base44 FAB (platform chrome, excluded by rule). The fresh drill-down EXPANDED pair: VERDICT DIFFERENT with 5 style flags, every one DOM-arbitrated via arbit-v67.sh + arbit-v67b.sh (the two-part computed census on both sides — the five flags + the networth summary-card HEIGHT family) to documented classes — the Dashboard nav highlight (the root-URL active-nav gap, re-confirmed live: the reference transparent/zinc-700/400 vs the clone transparent/white/500 — the superset contrast #3), the avatar (the user-identity class), the hero progress-bar fill (the pct × 568 data class re-measured live: the track 568×12 rgba(255,255,255,0.1) BYTE-IDENTICAL both sides, the fills 173/357 = each site’s own allocation pct), the donut percentage labels (the PERCEPTION ARTIFACT re-confirmed live: both sides render count 4 [1 group + 3 labels]/labelLineCount 0/16px/black/3 sectors — the v9 G1b contract holds), and the stat cards’ footer TREND ICONS (a NEW flag variant — “different paths/angles” — DOM-arbitrated to a PERCEPTION ARTIFACT via a GLYPH census: BYTE-IDENTICAL 16×16/rgb(143,188,63)/viewBox “0 0 24 24” with the same two-polyline Heroicons-v1 outline trending-up — POLYLINE-based, no `` element, which is why the census needs the innerHTML glyph signature; plus the hero-glyph census: the same 7 svgs in the same order on both sides). The pair is CLEAN. The never-pinned HEIGHT family measured BYTE-IDENTICAL: the card 960×333 (the deterministic arithmetic 32+76+32+84+24+24+1+28+32), the head row h 76 (the label 20 + mb 8 + the figure 48 govern; the 64px medallion rides), the figure h 48 (lh 48px), the stat grid rows “84px”/h 84 (ONE row at 1280 — the 390 twin’s two-row 184 is the v63 class), the stat tile 440×84, the ratio footer mt/pt 24/24 + border 1px + h 53. The axe watch: local NO DRIFT (the registry state unchanged from v66; the lockfile holds 4.13.0; the nested engine byte-identical) — and the upgrade DECISION: the 4.14.0 upgrade is DEFERRED TO A DEDICATED v68 ITERATION (never a drift, never a rider — the complete specification carried in the v67 plan’s v68 queue). The v67 pin (+1 → 269): the summary card’s height-family census at 1280 in tests/e2e/networth.spec.ts (a new describe after the v60 F1 block, the read-only logged-in state). THREE mutations verified to FAIL (the head row mb-8→mb-4; BOTH stat tiles p-4→p-2 — the shared-class regression: a FIRST-tile-only mutation does NOT move the axis, the tile height is STRETCH-GOVERNED, the grid row follows the taller p-4 sibling and stretches the p-2 tile back to 84, measured live; the ratio footer mt-6 pt-6→mt-3 pt-3). The 61st standing checks clean (the mobile menu verified working — the 61st time; the reference’s four bugs all live; the data census intact — the 61st clean drift check). The 16 screenshots BYTE-IDENTICAL. — 108/269/35 green | | docs/remediation-plan-v68.md | The DEDICATED axe-core wrapper upgrade (the deliberate ^4.13.0→^4.14.0 pin bump + the lockfile regeneration + the engine verify + the full a11y gate re-run + the rule-coverage re-audit: the activated rule sets byte-identical, 69 of 105 rules each, zero added/removed) + the v61-era pair’s fresh standing sweep (the dashboard full-page 1280 re-run — VERDICT IDENTICAL) + the a11y gate’s ENGINE-CONTRACT pin (TOOLCHAIN-ONLY — src/ untouched; +1 e2e = 270: a real scan’s testEngine.version asserted at 4.14.x, the “never a drift” discipline made executable) + the 62nd standing checks (the mobile menu verified working, the 62nd consecutive time). | | docs/remediation-plan-v69.md | The mobile 390 pair re-run (the networth full-page twin — the documented queue item — PLUS the dashboard mobile twin, the other never-re-run mobile pair, both in one two-pair iteration) + the axe-core watch at its re-check point (NO DRIFT — npm-latest 4.14.0, no 5.x, the engine-contract pin green) + the networth view’s PAGE-HEADER ROW computed census pin (PIN-ONLY for src/; +1 e2e = 271) + the F0 lint-gate repair (the v68 ship’s own miss — the seedcheck .cjs committed after v68’s final lint run, repaired with a single-file eslint-disable) (1 new e2e + three pin-sanity mutations + the 63rd standing checks). Both mobile pairs CLEAN, every flag DOM-arbitrated via arbit-v69.sh: the networth re-run’s three flags = the v63-era classes re-confirmed live with the exact v63 numbers (the stat-tile stacking [the v41 mobile-superset class], the trend icon [the v54 clipping perception artifact — the medallion 64×64 + icon 32×32 byte-identical both sides], the Base44 overlay [platform chrome]); the dashboard re-run’s flags = the Base44 tooltip [platform chrome] + the subtitle wrap [the documented v62 class — the chrome byte-identical, the reference’s own docSW 395 overflow shifting its wrap point] + a NEW flag variant DOM-DISPROVEN (“the reference’s breakdown card contains a donut” — ZERO donut on EITHER side, rechartsWrappers 0/pieSectors 0 both, the card h 324 both sides, the 7 svgs the byte-identical row icons — a perception artifact, the DOM census the ground truth). The v69 pin: the networth header row at 390 — the row flex/center/gap-3 12px/mb-8 32px/w 358/h 60 (the TEXT BLOCK governs, not the 48px chip — the v65 lesson), the chip 48×48/rounded-xl/forest→lime gradient + the 24×24 white icon, the h1 30/700/lh 36/forestDark, the block w 250 (the subtitle’s max-content governs)/h 60, the subtitle 16/400/gray-500 one line, the row→card gap 32, docSW 390. THREE mutations verified to FAIL (the row mb-8→mb-4; the chip h-12 w-12→h-10 w-10; the h1 text-3xl→text-2xl). The 63rd standing checks clean (the mobile menu verified working — the 63rd consecutive time; the reference’s four bugs all live; the data census intact — the 63rd clean drift check). The 16 screenshots 16/16 BYTE-IDENTICAL (the first fully-identical set since v65 — the same-day seed). — 108/271/35 green | | docs/remediation-plan-v70.md | The OLDEST re-run candidate’s fresh standing sweep (the v57-era items pair — income + expenses at 1280 full-page, last re-shot by v64 itself; the re-run cadence’s first COMPLETED cycle: v64→v70) + the axe-core watch at its re-check point (NO DRIFT — npm-latest 4.14.0, no 5.x, the engine-contract pin green) + the expenses view’s filter-card computed census pin (PIN-ONLY for src/; +1 e2e = 272) (1 new e2e + three pin-sanity mutations + the 64th standing checks). Both items pairs CLEAN, every flag DOM-arbitrated via arbit-v70.sh + arbit-v70b.sh: PAIR 1 (income) — the avatar (the user-identity class: chrome byte-identical 36×36/radius 9999px/14px-600/white, only “U” vs “D”), the card grid (the seed-count class: the axes byte-identical — w 960/gap 16px/the same fractional 309.344px last track), and the card heights (ref 200 vs clone 230) DECOMPOSED arithmetic-wise — the clone’s Salary card carries the 4th Recurring superset badge → the wrap h 52 (22+8+22) → +30px, and the sibling Freelance card (natural 200) STRETCHES to 230 with it (the grid stretch-governed class, the v67 lesson: the row = the taller sibling); PAIR 2 (expenses) — the footer (the v57 DATA-CONDITIONAL class: chrome byte-identical, the reference’s expense items carry no paymentMethod values), the badge arrangement (the v57 wrap class re-confirmed live: 4 badges → h 52 vs 3 → h 22, the card-height delta the same wrap arithmetic), the avatar (user-identity, again). The v70 pin: the expenses filter card at 1280 — the LAST never-census’d variant (v38 S2 = /savings, v64 = /income, v52 F3 = the mobile 390 state): the card 960×86/pad 24/radius 16, the search 216×36 (the 1-COL span — the search does NOT span 2 columns, the expenses variant’s defining difference — with the asymmetric 40px left pad), the 4-track 215.5px grid/4 children, the three comboboxes 216×36 (“All Categories”/“All Frequencies”/“All Payment Methods”). THREE mutations verified to FAIL (the expense-branch gap-4→gap-2; the card p-6→p-4; the search pl-10→pl-4). The 64th standing checks clean (the mobile menu verified working — the 64th consecutive time; the reference’s four bugs all live; the data census intact — the 64th clean drift check). The 16 screenshots 16/16 BYTE-IDENTICAL (the second consecutive fully-identical set — the same-day seed). — 108/272/35 green | | docs/remediation-plan-v71.md | The NEW-pair program opens (the items views’ mobile 390 full-page pairs — income + expenses + savings, the below-the-fold surfaces never individually shot) + the axe-core watch at its re-check point (NO DRIFT) + the item-card SUBTITLE-P conditional-render fix (THE FIX — the first production-code change since v61; +2 e2e = 274): the reference renders NO subtitle P for empty subcategories (titleRow 28, card 180) — the clone’s unconditional render left a 20px empty block on user-created empty-subcategory items; TDD (the F4 regression RED→GREEN + the F5 with-subtitle pin + the verified pin-sanity mutation); the 65th standing checks; the screenshots 16/16 byte-identical. | | docs/remediation-plan-v72.md | The NEW-pair program continues (the details-sheet mobile 390 twins — the WITH-subtitle Rent sheet + the EMPTY-subcategory sheet, the v71 fix’s sibling surface — + the drill-down EXPANDED mobile twin) + the axe watch (NO DRIFT) + the details-sheet EMPTY-subtitle pin (PIN-ONLY; +1 e2e = 275). The surface findings: the no-sub sheets BOTH render NO subtitle P (the v52 conditional’s empty side verified live against the reference’s own Miscellaneous sheet); every other flag DOM-arbitrated (the close-button “border” a perception artifact — the chrome byte-identical; the crop claim a perception artifact — the meta rows behind the reference’s own FAB; the payment-method row the v52 data-conditional class; the drill-down row inventory the seed/notes data classes; the 5px width delta the reference’s own docSW-395 overflow). The probe lessons: the reference’s sheet carries no role=dialog (the custom fixed-overlay detection), the card-open is the CARD ROOT click, the clone’s Edit-dialog cleanup reads the Category INPUT VALUE. | | docs/remediation-plan-v73.md | The NEW-pair program’s LAST never-shot surfaces (the login + calculator mobile 390 twins — the login pair VERDICT IDENTICAL, zero flags; the calculator pair CLEAN: the “scrim lighter” claim a PERCEPTION ARTIFACT (the scrim chrome byte-identical — bg rgba(26,58,46,0.5)/opacity 1/blur(8px)/390×844, the reference’s wrapper-scrim vs the clone’s Radix sibling the v72 structural class), the “Will update category total” span the v66 DATA-CONDITIONAL class) + the axe watch (NO DRIFT) + the calculator scrim + summary-text chrome pin (PIN-ONLY, +1 e2e = 276) | | docs/remediation-plan-v74.md | The re-run cycle’s OLDEST candidate (the v58-era savings pair’s second re-run — the only pre-v70 pair not re-shot since v65) — the savings view at desktop 1280 FULL-PAGE re-shot both sides, every VLM flag DOM-arbitrated via arbit-v74.sh (the grid chrome byte-identical — the ≈309.33px 3-col template/16px gap/960 w, the childCount the seed class; the footer the v58 DATA-CONDITIONAL class re-verified — the paymentMethod-presence childCount, the chrome byte-identical, a new micro-observation: the reference’s footer children DIVs vs the clone’s SPANs, the v72 structural class; the badges the classification-VALUE data class with the monthly/active byte-identical and the clone’s 4th Recurring superset badge wrapping to h 52) + the axe watch (NO DRIFT) + the savings goal-card’s interior-text chrome + the hidden app-shell h1 pin (PIN-ONLY, +1 e2e = 277 — the h4 16/600/lh 24/rgb(26,58,46), the amount 24/700/lh 32/rgb(59,126,161), the hidden md-hidden h1 “ZeroBalance” 20/700/rgb(26,58,46) ZERO-SIZE w 0/h 0) | | docs/remediation-plan-v75.md | The re-run cycle’s next-oldest candidate (the v59-era login + calculator pair’s second re-run — last re-shot v66): the LOGIN pair at desktop 1280 VERDICT IDENTICAL (zero style flags; the both-sides census byte-identical on every axis — the card 448/16px/white-95, the h1 30/700/lh 36, the inputs 48/12px/14px, the Sign in 368×48/12px/14px/500, the logo 96×96) + the CALCULATOR pair’s three flags all DOM-arbitrated via arbit-v75.sh to documented classes (the will-update span the v66 DATA-CONDITIONAL class; the row hover state the v66 pointer-position class — actionsAtRest opacity 0/btnCount 2 on BOTH sides; the amount alignment the pointer-position corollary — the amount chrome byte-identical 20/700/rgb(224,122,59)/h 28) + the axe watch (NO DRIFT) + the login card’s geometry + typography pin (PIN-ONLY, +1 e2e = 278 — the never-pinned axes of the VLM-identical surface: the card geometry, the h1’s TYPOGRAPHY family, the subtitle fs, the input pair’s radius/fs, the Sign in FULL chrome, the logo geometry) | | docs/remediation-plan-v76.md | The re-run cycle’s next-oldest candidate (the v60-era networth + drill-down pair’s second re-run — last re-shot v67): the NETWORTH pair’s summary-card HEIGHT family re-measured byte-identical with ZERO drift since the v67 pin (the card 960×333/32px/16px/the forest→lime gradient, the headRow 76, the figure 48/700, the medallion 64×64, the statGrid 440px×2/rows 84px, the statTile 440×84/blur(10px), the ratioFooter 53) + the DRILL-DOWN pair’s six flags all DOM-arbitrated via arbit-v76.sh to documented classes (the nav highlight the v56 root-URL class; the progress fill the pct × 568 data class; the donut labels the v60 perception class — the label chrome byte-identical count 4/16px/black/3 sectors; the extra rows the seed-data classes; the tooltip the Base44 chrome) + the axe watch (NO DRIFT) + the breakdown section-icon chip glyph + donut label chrome pin (PIN-ONLY, +1 e2e = 279 — the never-pinned axes: the income-section icon’s 16×16 twin-polyline glyphSig/rgb(143,188,63)/viewBox + its 32×32 chip + the donut label count/fs/fill/sectors) + the F0 lesson (the standalone static copy: npm run build, never a bare npx next build) | | docs/remediation-plan-v77.md | The re-run cycle’s next-oldest candidate (the v61-era dashboard pair’s second re-run — last re-shot v68; the session’s target-selection correction of the v76 plan’s queue suggestion): the dashboard FULL-PAGE at desktop 1280 — VERDICT IDENTICAL, ZERO style flags (matching the v68 re-run) + the both-sides census byte-identical on every axis with ZERO drift since v61 (the guidelines card 468×348 — the STRETCH-fix natural height holding; the header 16/600/forestDark/lh 16/tracking −0.4; the three tiles 418×78 with the tint+border pairs + label 16/500/type + pct 14/400/type + desc 12/gray/lh 16; the quick-action row byte-identical — the label delta the v61 wrapper-vs-leaf lesson re-learned, the grid mt the margin-structure class with the rendered cross-row gap 340 byte-identical) + the axe watch (NO DRIFT) + the Budget Guidelines interior-text family + quick-action icon viewBox pin (PIN-ONLY, +1 e2e = 280 — the never-pinned axes: the header’s lh + tracking, tiles 2/3 + all three tiles’ label/pct/desc text chrome at 1280 [the v51 F4 family’s desktop twin], the quick-action icons’ viewBox [the v76 glyph family’s extension]) | | docs/remediation-plan-v78.md | The re-run cycle’s next-oldest candidate (the mobile 390 twins’ second re-run — the networth + dashboard full-page pair, last re-shot v69; the ledger-verified queue — the v77 plan’s suggestion agreeing with the rotation ledger this time): BOTH mobile twins at 390 — verdicts DIFFERENT with every flag a documented class (the v41 mobile-superset stacking + seed data + the Base44 overlay + the avatar on the networth; the v62 subtitle wrap width-delta + the pct data class + the platform tooltip + the donut seed proportions on the dashboard) + the both-sides census: the networth medallion 64×64 + trend icon 32×32 + tiles h 84 + gap 16 BYTE-IDENTICAL, the dashboard subtitle chrome + the breakdown card’s h/svg families BYTE-IDENTICAL (h 324, svgCount 7, svgs [16,14,16,14,16,14,16], rechartsWrappers 0 + pieSectors 0 — no donut on EITHER side, the v69 disproof class; the 5px width delta the reference’s own docSW-395 overflow class) + the axe watch (NO DRIFT) + the breakdown card’s mobile svg family + chart-absence + width census pin (PIN-ONLY, +1 e2e = 281 — the never-pinned axes: the card w 358, the svg family [16,14,16,14,16,14,16], the chart-absence contract, the H3 308×20 + DIV 308×238 children) | | docs/remediation-plan-v79.md | The re-run cycle’s next-oldest candidate (the v57-era items pair’s third re-run — the income + expenses view pair at desktop 1280, last re-shot v70; the ledger-verified queue — the v78 plan’s suggestion agreeing with the rotation ledger): BOTH items views at 1280 — verdicts DIFFERENT with every flag a documented class (the avatar the user-identity class; the card counts the seed-count class; the footer chips the v57 DATA-CONDITIONAL class; the badge arrangement the v57 Recurring wrap class — matching the v70 re-run’s own flag set exactly) + the both-sides census (arbit-v79.sh): the AVATAR chrome BYTE-IDENTICAL (36×36/9999px/14px/600/white/rgb(143,188,63)), the GRID + CARD chrome BYTE-IDENTICAL (grid 960/16px/3-col; card 309/20px/12px — the h 200-vs-230 the v70b-decomposed Recurring-wrap + sibling-stretch class), the AMOUNT leaf chrome BYTE-IDENTICAL (24/700/lh 32 with the income lime + expenses orange accents), the FOOTER + BADGE chrome BYTE-IDENTICAL, the FILTER CARD BYTE-IDENTICAL on both variants (the v64/v70 pins’ own values, zero drift) + the axe watch (NO DRIFT) + the rail avatar + item-card amount-accent + wrap/stretch card-height pin (PIN-ONLY, +1 e2e = 282 — the never-pinned axes: the avatar’s full chrome [the tokens.spec pin holds the radius only], the amount’s per-type accents [the v74 savings twin completed to all three types], the income first card’s h 230 [the v70b wrap/stretch arithmetic regression guard]) | | docs/remediation-plan-v58.md | The savings-view desktop 1280 VLM standing pair + the savings goal-card’s footer + badge-row census (PIN-ONLY — no production-code change; the eighth iteration since v46, the pin-only line’s continuation) (1 new e2e + three pin-sanity mutations; the session-137 v58 queue’s own first suggestion — the items family’s remaining view, the v57 pair having covered income + expenses; the savings goal-card’s 1280 state was unpinned at the pair viewport, its internals pinned at 390 only [v54 F1]). The pair: every flag DOM-arbitrated to the documented data classes — the seed data (the reference’s own 1-item savings view [“1 items · $1000.00”, “Emergency Fund” $1000] vs the clone’s dev seed [“2 items · $1250.00”, Investments + Emergency Fund]); the savings card’s footer content (DATA-CONDITIONAL: the footer row’s chrome byte-identical on both sites [flex justify-between 12px rgb(107,114,128) h 16, each span flex gap 4px with the 12×12 gray icon] — the reference’s own savings item carries no paymentMethod so its footer renders one child [the date alone], the clone’s render the same two-child structure the reference’s own income card renders [the date + “Brokerage”] — the v57 conditional-footer contract measured on the third view); the badge family (DATA-CONDITIONAL: the row’s chrome byte-identical [flex wrap 8px mb 12px, every badge flex center/6px/2px 10px/12px/600/h 22], the monthly purple-50 + active slate-50 badges byte-identical — the reference’s savings item is classified “need” [red] while the clone’s are “savings” [green]: the classification VALUE is the data class, the color maps stay locked by badge-colors.spec.ts; the clone’s Recurring superset badge wraps the row to h 52 = 22+8+22); the grid (960/16px/3-col ≈309.33 — the v45 F1a pin’s own values). The v58 pin (tests/e2e/items.spec.ts, +1 → 258): the savings goal-card census at 1280 — the /savings grid (960/2 children/16px gap/the 3-col template), the Investments card’s two-children footer (the date span + the “Brokerage” chip), the wrapped 4-badge row (h 52, [savings, monthly, Recurring, active]) + every badge’s geometry. THREE pin-sanity mutations verified to FAIL (the footer text-xs→text-sm, the badge-row gap-2→gap-4, the classification badge py-0.5→py-1 — the fresh VERTICAL-pad axis proving the wrap arithmetic 22+8+22). The 52nd standing checks clean (the mobile menu verified working; the reference’s four bugs all live). — 108/258/35 green | | docs/remediation-plan-v43.md | The drill-down EXPANDED-state census + the static-chevron and hover-rgb fixes — the session-92/93 queue’s two items. The surface: the drill-down’s expanded bands (the categories/subcategories/items rows — the v42 census pinned the collapsed card only) + the summary card’s hover/interaction states, measured live on BOTH sites at 640×960 (exp-census-v43.mjs + exp-detail-v43.mjs + exp-rotate-v43.mjs + hover-v43.mjs) — the per-row chrome BYTE-IDENTICAL at every level (the section 48/“8px”/-8px/8px radius, the category 28/“6px 12px”/inline rgb(249,250,251)/6×6 dot, the sub 24/“4px 8px”/rgb(250,250,250)/4×4 dot at 0.6, the item 24/“4px 8px”/12px rgb(156,163,175)/2×2 dot at 0.4/120px truncate, the footer 29/“6px 12px”/border-t; the expansion arithmetic 324→441→495, sw 640 fully expanded — the heights differ only by DATA). TWO v43 PARITY FIXES (the first production-code change since v42): (F1) the chevrons — the clone’s v4 rotate-180 rendered rotate: 180deg (0.15s) where the reference’s are STATIC (rotate: none, transition “all” 0s at both levels — Tailwind v4’s rotate- emits the rotate PROPERTY, not transform; a transform: none reading is meaningless on v4*), both chevrons stripped, the aria-expanded superset KEPT; (F2) the section-row hover — the named palette hover:bg-gray-50 computed lab(98.2596 -0.247031 -0.706708) where the reference’s engaged v3 gray-50 computes rgb(249,250,251) — the named-palette trap LIVE in a hover state (the section button carries no inline bg) — fixed hover:bg-[#f9fafb]. The category rows’ hover is inert on BOTH sites (the inline bg overrides — v29); the queue’s summary-card item RESOLVED (static on both, no interaction states). THREE pin-sanity mutations verified to FAIL (the restored rotate-180, the reverted hover hex, the sub-row py-1→py-2). Standing checks clean (37th): mobile-nav R1–R4 both sites (Tailwind v4 pins hold, the mobile menu works), data drift (reference read-only, before + after), the SEO pair, the VLM pair (all flags in the known classes — the reference’s own Base44 badge was misattributed to the clone, DOM-verified). — 108/209/35 green (3 new e2e, 2-line production fix, TDD) | | docs/remediation-plan-v42.md | The mid-viewport census of the five routes + the breakdown card’s single-divider fix — the session-90/91 queue (the v41 lesson applied to the remaining route set). The surface: the five non-networth routes (/, /dashboard, /income, /expenses, /savings) measured live on BOTH sites at 640×960 (mid-census-v42.mjs) — the 32px rhythm HOLDS on all five routes (gapRowToFirst === 32 everywhere, capW 608, sw 640, the items-view structure identical — the v20 G1 pin’s class extended to the mid viewport). THE v42 PARITY FIX (F1, the first production-code change since v41): the NetZeroBreakdownCard ran +13px taller than the reference at BOTH viewports (640: 337 vs 324; 1280: 377 vs 364) — attributed band-by-band (hero-attrib-v42.mjs + bd-attrib-v42.mjs): the goalCard byte-identical (316, the bands 48/48/108, the donut 558×300), the row heights identical (56/64/64 at 1280, the wrap width-driven not data-driven) — the delta was ONE EXTRA DIVIDER: the clone rendered TWO (Income|Savings AND Savings|Expenses), the reference ONE (Income|Savings only — the app’s ONE rule grouped: [Income] | [Savings, Expenses] | border-t | [Net Balance]). The fix: {i === 1 && }; every band below the hero shifts back 13px at 640 (secondBand 925, thirdBand 1635, scrollH 3029 — the tail-v42.mjs check proved the delta was the divider alone, the lastBand mb=0 on both sites). F2: the breakdown census e2e at 1280 (the 5-band stack [56, 1, 64, 64, 45], the single divider, the 364px card) + the hero grid census at 640 (goalCard 316, breakdownCard 324@y=569, hero 664, secondBand 925, thirdBand 1635, scrollH 3029); F3: the five-route mid-viewport census e2e (the common rhythm + the route-class branches: the dashboard’s 1-col 24px-gap hero grid, the items’ block mb-24 filter + 1-col 16px-gap grid). THREE pin-sanity mutations verified to FAIL: the reverted divider condition → the band-count pin fails (6 ≠ 5); the items-view mb-8 → mb-6 → the /income rhythm pin fails (24 ≠ 32); the dashboard mb-8 → mb-6 → the / rhythm pin fails. Standing checks clean (36th): mobile-nav R1–R4 both sites (Tailwind v4 pins hold, the mobile menu works), data drift (reference read-only, before + after), the SEO pair, the VLM pair (all flags in the known classes). — 108/209/35 green (3 new e2e, 1-line production fix, TDD) |

| docs/session_1.md · docs/session_2.md · docs/session_3.md | Narrative logs of the build + re-verification sessions | | worklog.md | Rolling project worklog (all sessions, latest first) | | Project_Architecture_Document.md | 7 ADRs, topology, ER diagram, security model |

Deployment

next build emits a standalone server (docs/DEPLOYMENT.md has the full runbook):

npm run build
DATABASE_URL="file:/absolute/path/to/custom.db" AUTH_SECRET="$(openssl rand -hex 32)" \
  node .next/standalone/server.js

Use an absolute file: URL in production — a relative URL resolves against the standalone build’s own location.

Screenshots

Full 16-shot set in docs/screenshots/ — regenerate with node scripts/capture-screenshots.mjs.

Troubleshooting

License

Private project — all rights reserved.