# Builder PLAN review against the requested experience

Reviewed September 9, 2026. Builder source: `C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder`, revision `3f2991d63756d50e5c478e5b90f9f7eaec0d20f1`. QMM source and fixture inspected at `6decdf8`. This review reads the PLAN together with its adopted audits, the user's quoted statements, and the implementation. It does not change the candidate or its PLAN, run client services, inspect a live database or deploy anything.

The PLAN's central direction is right: distinct tools for distinct disciplines, sharing one project and a coherent application. The main improvement is to make complete customer-facing workflows the unit of delivery and authorize the supporting rewrites in the same run. Existing partial behavior measures the remaining distance; it does not satisfy the requested result or constrain how the new application may be built.

[BUILDER-OVERNIGHT-HANDOFF.md](BUILDER-OVERNIGHT-HANDOFF.md) is the ready-to-use execution supplement. It incorporates the findings below, substantial rewrite authority and continuation through the full assignment. The owner's accompanying prompt supplies the branding update.

## The quoted requirements as acceptance tests

| User's direction | What must be observable in the result |
|---|---|
| Intuitive tools for creating and modifying each process | The primary workflow completes inside the correct workspace, with real changes and persistence. |
| Different engines and toolsets that communicate smoothly | Each discipline has its own useful composition; a saved message, asset, list or pipeline is available where another tool needs it. |
| Full-width application; panels clearly distinct | Judge the populated application at 2560 × 1440. Space, separation and controls support work at that size. |
| Scrollbars beside unused space signal a problem | A catalog uses its allocated region instead of inheriting a small fixed-height viewport. |
| Background preparation so workspaces are ready | First opening exposes useful content promptly; switching preserves active work. |
| Lists can be created; people can belong to multiple lists | Empty lists exist independently, creation works, and memberships reflect identifiable customer activity. |
| Standard marketing tracking; reports that can be filtered | Cohorts, ranges and filters have consistent meanings; the rows explain the numbers. |

Source: [AUDITOR-BRIEF.md](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/docs/audits/2026-09-08/AUDITOR-BRIEF.md:11). The later request for Codex-style dimensions is additional direction from this conversation. No claim here depends on inspecting the Codex application's own browser chrome.

## 1. Make the assignment finish the whole application

PLAN has a phase order and verification gates, but the run needs an explicit instruction to continue through every phase. Branding is step 7; Patterns, Interactivity, Email, Search, Calendar and Creative are collected in step 10 behind substantial unrelated work. These tools should be complete named deliverables within the same run, with independent work developed concurrently when useful.

The interface does not have to wait for a completed people model. Missing models and write paths are part of the implementation, developed and connected with the tools that need them. Likewise, substantial existing code may need replacement. Preserve the user's work, correct behavior and real-data meaning; do not require the desired experience to fit the old architecture.

The overnight supplement makes checkpoints internal, supports continuation after context compaction, and requires the final packaged candidate. It retains PLAN's actual boundaries around the running production Builder, real-site development servers, deployment-triggering pushes and separately authorized shared-database changes. It does not manufacture a phase-by-phase permission process.

Source: [PLAN order](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/PLAN.md:274).

## 2. Finish creation, especially Email and Creative

**Creative has a concrete gap between its promised action and its writer.** PLAN calls for Approve and Place, and the Creative audit says saving through `assets.save` re-derives the image in its role. The fixture's writer saves image bytes and a provenance record under `src/assets/builder/`; it does not receive a destination role, change source binding or generate derivatives. The current create handler also expects an existing role/source.

The completed workflow must create or import, compare, assign an existing or new role, update the actual destination and all affected previews, persist, reopen and revert. Favicon creation must attach the usable output. A saved file alone is not placement.

Sources: [PLAN Creative](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/PLAN.md:263), [audit claim](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/docs/audits/2026-09-08/audit-creative.md:215), [asset writer](C:/ClaudeCode/projects/.mlm-work/qmm-next/tools/edit-server.mjs:1502), [create handler](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/integration/index.mjs:1818).

**Email needs a complete New message path.** PLAN specifies catalog, viewer and in-place editing. Its audit adds duplication and covers transactional-message editing, but does not finish new-message identity, composition, persistence and availability to Automations. Add creation from an empty library, purpose/layout choice, supported content arrangement, representative recipient preview, save/reopen and automation selection. This is a recommendation grounded in the user's broad create-and-modify requirement, beyond the specific complaint about Email's short catalog.

Sources: [PLAN Email](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/PLAN.md:216), [Email audit](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/docs/audits/2026-09-08/audit-email.md:145), [current viewer](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/panel/email-panels.js:695).

Pattern scaffolding, pipeline creation/writers and element-level Branding Link are already specified in the adopted audits. They are completion obligations, not missing-plan findings.

## 3. Shared components must support each tool's own workflow

PLAN's blanket empty-state instruction requires one action and “Ask Claude for it,” and prevents panels supplying their own empty language. This conflicts with the useful New pattern and New pipeline actions in its own adopted audits. Share the state presentation; let the workspace supply meaningful copy and creation actions.

The shared inspector similarly should be a component, not a mandatory composition for every discipline. Current layout variation does exist, but it does not establish the desired tool quality. User recognition, working actions and the task's use of space determine the layout.

The precise 6px gutter, 28px header, eight type sizes, two weights and one easing are authored design prescriptions, not the user's quoted requirements. Refine them as needed using the supplied brand system and rendered product. The quote about panel separation does not prohibit lighter panel grounds; that prohibition is the auditor's interpretation after the quote.

Sources: [shell states](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/PLAN.md:154), [Patterns empty state](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/docs/audits/2026-09-08/audit-patterns.md:246), [pipeline creation](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/docs/audits/2026-09-08/audit-interactivity.md:135), [panel quote](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/docs/audits/2026-09-08/AUDITOR-BRIEF.md:22).

## 4. Add true Width × Height, with display scale separate

PLAN already adds laptop widths, typed width, Fit, a ruler and a second viewport. Extend this to explicit Width × Height in CSS pixels, presets/custom sizes, orientation swap and independent Fit/100% presentation. The full-width Builder window and the page's simulated device dimensions are different things.

The current implementation has phone presets and custom phone width, but clamps requested preview width to the available region and inherits its height. Fitting must preserve the requested page dimensions and responsive breakpoint. Verify pointer editing, selection and capture when scaled. Show the capture target independently of selected object and focused panel.

Sources: [PLAN Website](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/PLAN.md:190), [existing dimensions](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/app/main.mjs:1236), [capture selection](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/app/main.mjs:1080).

## 5. Lists must reflect the actual customer journey

PLAN's “no people-and-lists model” claim is too broad. QMM already stores people, tags and audience memberships, including join/leave behavior. The important missing result is an independently declared list catalog with creation, rules, empty-list identity, editing and automation relationships. Existing memberships do not deliver that result.

PLAN also assumes all calculator/quote runs already collect email. They do not: calculations submit tool inputs and a session; emailing results is a separate later action. Capture anonymous activity with precise entry-tool identity and connect it when the person supplies identity. A shared result kind such as `myga`, `fia` or `both` cannot establish which entry tool was used. A forced email gate would be an additional product change, not a fix to match existing behavior.

Time-based journey rules and score decay need time-based evaluation. “Not booked after N days” cannot depend solely on a second action arriving. Handle repeats, order/windows and late identity, keep contributions explainable, and give historical reports an explicit time meaning. Deliver scoring as planned without turning score setup into a prerequisite for an operator's New list action.

Sources: [existing people/membership tables](C:/ClaudeCode/projects/the-quiet-money-machine/migrations/0011_add_people.sql:11), [row-derived catalog](C:/ClaudeCode/projects/the-quiet-money-machine/functions/api/__lists.ts:56), [booking memberships](C:/ClaudeCode/projects/the-quiet-money-machine/functions/api/booking.ts:452), [calculation inputs](C:/ClaudeCode/projects/the-quiet-money-machine/src/components/QuoteCalculator.tsx:738), [later email capture](C:/ClaudeCode/projects/the-quiet-money-machine/src/components/QuoteCalculator.tsx:2652), [PLAN journey rules](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/PLAN.md:315).

## 6. A common insights UI also needs corrected measurements

PLAN correctly requests ordered cohorts, filters and shared definitions. Its assertion that fragmentation is Builder-side understates the underlying query work. Current funnel steps count sets independently; each unfiltered funnel receives all outcome money in the window; the Tracking journey query omits the selected range. The generic environment-scoping table list also omits outcomes, page events and journey links.

Deliver coherent source environment, event-time window, counting unit, cohort entry, ordered progression and attribution across the UI and queries. Prove summaries against drill-down rows. These corrections belong in the same feature work; they are not a reason to postpone designing useful Reports and Funnels.

Sources: [funnel counts](C:/ClaudeCode/projects/the-quiet-money-machine/functions/api/__funnels.ts:190), [money allocation](C:/ClaudeCode/projects/the-quiet-money-machine/functions/api/__funnels.ts:253), [journey query](C:/ClaudeCode/projects/the-quiet-money-machine/functions/api/__tracking.ts:226), [scoping helper](C:/ClaudeCode/projects/the-quiet-money-machine/functions/api/__scope.ts:23).

## 7. Verify continuity, graph meaning and first opening

Current automation draft/build behavior can flatten richer graph structure. The desired canvas must preserve branch, condition, action, merge and identity through the entire save/reopen/build path. A working visual graph or node palette alone does not complete an automation engine.

PLAN already promises state preservation and preloading; make the result concrete. Existing hidden workspace layout returns before mounting. Separate background preparation, mount and activation; deduplicate pending jobs, preserve drafts, refresh catalogs, clean up subscriptions and compute geometry on activation. Verify useful first interaction rather than relying on “mounted hidden” as evidence.

Extend the Notes conflict-guard principle to direct edits and undo throughout the application. The user and agent can change the same sources; ordinary editing should remain quick, but saving or undoing must not silently replace intervening work.

Sources: [draft/build conversion](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/panel/automation-panels.js:418), [hidden layout](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/panel/workspace-dom.js:129), [current document write](C:/ClaudeCode/projects/.mlm-work/max-lvl-website-builder/integration/index.mjs:2462).

## Accuracy notes for the opening inventory

Replace “none of them can change the thing it is about” with the particular unmet experience. Branding can save documents but lacks the intended element-by-element guide editing. Calendar has creation actions but lacks the requested finished planner treatment. Search has field writes but constructs unreliable addresses. Funnels and Automations have limited editing, which is materially short of the requested engines.

The inherited `30vh` and fixed-size catalog rules are useful concrete evidence of wasted space; the mere existence of a full-window shell is not evidence that the full-width experience is solved. Likewise, some keyboard navigation already exists, but a coherent application keyboard model is still required.

Do not treat this correction of the inventory as a reduction in scope. Judge the final experience against the user's words and the complete PLAN, including its audits. The known missing rebrand section is deliberately outside this review's findings.