GUI workspace design

Status: proposed target arrangement, 1 October 2026. This is a design authority for future changes, not a statement that the redesign has shipped. The appendix records the changes needed from the reviewed frontend. Keep release and user documentation tied to implemented behavior.

This document governs the shared application shell, page navigation, subtabs, actions, project objects and Advisor context. It applies to the browser and desktop application, including pages contributed by the managed server. Product capabilities and authorization remain deployment-specific.

The Scientific Result Surface Contract continues to govern scientific meaning. A cleaner page must still disclose data, executed parameters, population, units, exclusions and provenance.

1. The arrangement

Use the same order, vocabulary and interaction rules everywhere.

Main page	Page subtabs, in order	Purpose
Dashboard	Your Project, Archive, Storage	Choose/manage projects by default; restore archived projects; inspect and reclaim storage across projects.
Project	None	Understand and manage the active project and its objects.
Data	Import, Synthesis, Upload, Library, Multi-well, My Dataset	Acquire, prepare and inspect project data. Preserve existing capability restrictions.
Workflow	None	Build and execute workflows. Keep the existing sheet tabs inside the canvas.
Runs	History, Inspect, Compare, Artifacts, Batch	Open a run name in Inspect, compare saved executions, inspect their artifacts and apply fitted models.
Optimize	Campaigns, New round, Results	Browse project campaigns, prepare a round and evaluate/inspect results.
Deploy	Application, Folder Watches, Prediction History	Select a fitted application, operate watches and inspect predictions.
Report	Setup, Preview, ISO Validation	Configure a report, inspect/export it and follow the validation-evidence walkthrough.

Use Workflow as the main-page label. Workflows → on Project describes its collection. Existing /workflow URLs and internal component names need not be renamed to achieve this vocabulary.

Settings, Documentation, Logs, Audit, account/billing, administration and authentication keep their appropriate global or contextual entry points. They are not extra scientific main pages. Managed-only features remain contributed through the existing extension boundary; OSS must not import server UI code.

2. Shared page anatomy

GLOBAL    Active project                         status · jobs · Advisor · account
HEADER    Page title                              essential actions for this view
CONTEXT   Selected object · relevant facts · scope/status               summary
SUBTABS   First    Second    Third ...                  (left aligned, stable order)
CONTENT   One full-width active panel

Global state: the topbar owns the active project selector. Its current behavior of hiding the selector while Advisor is open is acceptable; this redesign does not require a persistent duplicate. Keep status indicators and other global controls at the right. Hiding the name must not change the underlying project binding or Advisor context. Dashboard can have no active project. Other project pages show a useful selection state when none is selected.
Header: left-aligned stable page title, right-aligned essential actions. Campaign names, run names and modes belong in the context row. No generic subtitle beneath Workflow, Runs, Deploy or Report. Project's name and description remain useful project content.
Context: describe the active subtab and selected resource. Do not repeat a Project tile. Show the facts needed to understand the current operation, not a wall of badges. Use human-readable names in headers and the context row; technical identifiers belong behind Details or a copy button, with their authoritative full values accessible.
Subtabs: one horizontal, left-aligned strip below the context row. The active panel spans the available workspace width. “Full width” refers to the panel, not equal-width stretching of every tab label. Tables and plot/detail splits can still use multiple columns inside a panel.
Content: adopt Data's spacing, tab treatment, surface colors and typography as the shared baseline, expressed through shared theme tokens/components. Report's iframe and generated preview must use a compatible document theme; removing only the surrounding black background is insufficient.

Use common building blocks for the page header, context strip, subtab strip, empty/error/loading state and action presentation. Reuse existing responsive header and theme infrastructure where possible. Do not create an independent shell in Optimize. Shared navigation and command definitions should also be usable by future Windows/macOS menus, with the same resource and capability checks as their browser equivalents.

3. Subtab and action rules

A subtab is a view

Use short, stable nouns. Selecting a subtab changes what is visible; it never runs a workflow, starts billing, approves a proposal or reveals protected data.
Give each subtab a stable string ID, independent of its index or displayed label. Keep order stable within a deployment. Changing a label must not change stored state or Advisor memory identity.
Do not add a second page-level tab strip inside a panel. Local plot/table selectors and saved-run Summary/Results/Validation views remain legitimate controls for the selected object. Their visual hierarchy is subordinate.
Campaigns, New round and Results are views of a retained process, not a compulsory wizard. Seed, proposal and approval are sections of one New round form; live evaluation and inspection share Results. Moving between sections never restarts work or silently creates missing evidence.
Use native tab semantics, selected state, keyboard navigation and visible focus. At small widths, scroll the strip without clipping labels or mixing tabs into a “More actions” menu. Keep the active tab visible.

An action does something

Put the active view's essential actions in the shared header, normally with one visually primary next action. Less frequent page actions use a consistent overflow menu; row actions remain beside the row they operate on.
Filters, selection controls, parameter editors and contextual row actions stay in the panel. Moving an action to the header must preserve its selected object, effective parameters and disabled/reason state.
Use verbs for mutations or computation: Generate Report, Run Batch, Approve, Start Evaluation. Use destination arrows for navigation: Data →, Workflows →, Runs →, Campaigns →, Workflow →. An arrow never means “execute”.
Put Reclaim Storage on Dashboard, because it works across the user's projects. Storage has its own Dashboard subtab; Runs must not execute reclamation there. Preserve the scope, retention disclosures and recovery flow; following the link alone must not remove anything.
Remove routine Refresh/Reload buttons, including nested evidence panels. Load on entry, update after successful changes, and synchronize ongoing jobs. Show stale/disconnected state and a contextual Retry when retrieval fails. Never auto-retry a charge, execution or destructive action under this rule.
Do not turn design cleanup into new scientific gates. In particular, local folder watch does not require a successful dry run or analytical qualification. Existing integrity, authorization and incompatible-input handling still apply.

Every subtab has a contract

Before implementation, specify its purpose, selected object, context facts, actions, resulting records, empty/error states, persistence and Advisor focus. The page tables below supply these contracts. A new subtab requires the same description and a destination for every function it replaces.

Panel state	Required presentation
Nothing selected	Say which object to select and provide the selection action; do not imply a missing dataset when one is already selected.
Empty collection	Explain that this project/object has no records and offer the appropriate create/import/source action.
Loading or running	Identify the operation, preserve useful retained content, and show progress/Stop where supported. Loading must not look like an empty collection.
Partial or stale evidence	Identify what is retained, missing or outdated and its scope. Keep inspection/export where valid.
Read failure or disconnect	Keep context, show the actual failure and a scoped Retry.
Unsupported capability or lost access	Explain the actual deployment/access limitation and the available route forward; do not show a generic empty collection.

Counts distinguish selected, displayed/filtered and total records. Status badges describe retained state; they do not become navigation controls or substitute for approval. Selectors in the context row and selectors in the active panel must share the same authoritative state.

Copy and disclosure

Navigation, field and section labels use short nouns. Action buttons use short verbs that name the action; do not use sentences as labels.
At most one plain explanatory sentence per section is visible by default. Put longer guidance, limitations and implementation explanations behind a consistently labeled Details toggle. Do not keep several disclaimer cards to work around this limit.
Keep essential scientific scope, exclusions and irreversible-action consequences visible as concise labels/values or that one sentence. Details expands the explanation; it must not conceal which population or operation is involved.
No technical IDs, UUIDs, hashes or digests in page/section headers. Use names and human-readable revision labels. Full identity is available through a copy button or Details; shortening the display must not alter retained evidence.
Example approval summary: Rounds remaining: 3 · Trials: 9 (including seed). Follow with one sentence explaining what Start does; technical authority and execution-limit explanations belong in Details.

State and lifetime

Back/Forward, a deep link and reopening a window restore the project, object and subtab. Resolve ownership/access on the server. Never load an object into an unrelated active project just because its ID was in a URL.
Scope cached selection and draft state to actor, project and resource. Clear stale selections and discard late responses when any authority changes. Navigation can remember the last authorized view; it cannot expand access.
Keep unsaved work across ordinary subtab navigation. Warn before genuinely destructive navigation. Multiple browser/desktop windows must not silently retarget each other's active operation through shared local storage.
Navigating away from Results does not cancel a campaign. Display a global job indication with a return link; explicit Stop/Cancel owns interruption.
Preserve old route/query aliases, including Runs tab=models, and old memory scopes through explicit migration or aliases. A redirect must carry object identity and return context, not just land on the default page.

4. Page contracts

Dashboard and Project

Dashboard opens on Your Project, followed by Archive and Storage. Place Project → immediately to the right of New Analysis, leaving the project's relative update time right-aligned in its context row. Preserve project selection, archive, restore, permanent deletion and any profile-specific removal semantics.

Dashboard’s Storage subtab owns account-wide storage and Reclaim Storage. Existing /dashboard#storage links select that subtab. Expose its across-project scope, retained-output protection and outcome here. Opening Storage never performs reclamation; the explicit action remains inside that subtab.

Runs opens a run name in Inspect, within the Runs workspace. Existing /runs/:runId links retain their run, project and result context. Show the run identity first, a spaced action row next, and incomplete durable evidence after the retained details. Compare retains its comparison matrix and evidence-based ranking restrictions; it does not expose a Result pairing section.

Project has no subtabs. Give its name/description most of the available width. Move Edit and Export to the header. Keep New Project, Import, Audit, Memory Map and deletion/removal discoverable through primary or secondary actions. Memory Map must include campaigns and their relationships when the managed memory capability is available.

The project overview presents Data, Workflows, Runs, Artifacts and Campaigns as distinct collections with truthful counts. Use Data →, Workflows →, Runs → and Campaigns →. An artifact link goes directly to Runs → Artifacts. Keep the useful timestamps and short descriptions beside collection entries. Do not rename an artifact count as a run count: a run may produce zero, one or several artifacts. Campaign trial counts are not saved-run counts.

Preserve imported-project/package verification, exact local data binding, application workflow links, campaign reproduction and reproduction-report download. A concise imported-application section can link to Deploy → Application without hiding the steps required to reproduce the imported result.

Data

Keep the current subtab order and functional contents. Remove the duplicate Project context cell and left-align the tab strip. The context row follows the active acquisition/inspection view.

Subtab	Context and retained functions
Import	Provider/source and selected dataset; catalogs, availability explanations, previews, local/reference imports and Add to My Dataset. Respect download/egress policy.
Synthesis	Recipe, selected components and generated population; composition/curve editing, load/save curves, preview, bundle operations and guide.
Upload	Destination dataset and selected files; file-format handling, validation/diagnostics, scientific-asset selection and upload outcome.
Library	Provider/query, selection count and job state; search, preview, basket and import.
Multi-well	Selected dataset/acquisition plan and save state; samples, mixtures, factors, wells, run order and file assignments.
My Dataset	Dataset/file/asset identity, dimensions and selection; preview plots, data contents, metadata, quality/statistics, roles, rename/delete and source-file selection.

The Workflow → Data selection journey retains its explicit target sheet/source node, selected files or assets, target/group roles and Apply to this sheet / Cancel actions. Page cleanup must not replace an exact source binding with the most recently viewed dataset.

Workflow

No page subtabs. Preserve the present sheet tabs inside the canvas and all their operations. The context row starts with Active Sheet, then Active Data, then a right-aligned Canvas summary. Allocate about twice the width to each of the first two cells relative to the summary, adapting to long names and narrow windows. Remove the duplicate Project cell and generic subtitle.

Keep the settings/gear control as tall as adjacent header controls, even if it is narrower. Preserve Analysis Starter, Run, Export, Audit, autosave, version history, run-all settings, templates, sheet editing, catalog/graph operations, Inspector, expanded node detail and typed result views. Imported campaign trial views keep their source/evidence identity. Existing sheet tabs are neither replaced nor restyled into the page-level subtab mechanism.

Runs

Subtab	Context	Primary contents and actions
History	Filtered/total saved runs, run kind, selected count	Sort/filter/page, inspect, select for comparison, save/label/delete where supported. Clicking a name inspects; checking a row selects.
Compare	Exact selected run identities and compatible metric/population scope	Existing comparisons, result pairing/Check evaluation and report/export paths; retain compatibility explanations and selection limits.
Artifacts	Artifact count, selected artifact, source run and model/feature authority	Existing Models catalog under its corrected label; rename, inspect, readiness annotations, application history, Batch and Deploy links.
Batch	Selected fitted artifacts, input dataset/asset and cohort	Existing durable-dataset and authorized private-upload prediction paths; explicit Run Batch and partial/failure results. Completed saved runs link to History.

Header: Export on the same line as Runs, with no storage link. The reclaim action lives only on Dashboard. Remove the generic subtitle and Project cell; the context row identifies History, Compare, Artifacts or Batch. Saved-run detail retains Summary, Results and Validation, node outputs, effective parameters and links to Report/Optimize. These detail controls do not add main Runs subtabs.

Artifact existence, a human deployment-readiness annotation and analytical qualification are distinct facts. Preserve their labels and evidence. The Application tab is the home for operational application selection; Runs remains the home for the artifact's recorded source and execution history.

Optimize

Use the shared Data-style shell and the stable title Optimize. Campaigns belong to the active project. Account and organization administration remain outside this page. Round allowance and trial count are visible in New round's approval section because they affect the decision to execute.

Subtab	Context	Contents and action ownership
Campaigns	Project campaign count; selected campaign and lifecycle status	Project-only campaign list and lineage, including draft, running and terminal work as applicable; select/resume a round, inspect its summary or start a new campaign. Lifecycle labels reflect server state.
New round	Campaign, seed revision, source workflow/run and validation scope	One page containing seed selection, proposal and approval. Preserve saved-run/starter/promoted seeds, source recovery, all three search methods, grid/recipes, questions/hypotheses, user selection and workflow differences. Show rounds remaining and trial count before explicit Start.
Results	Selected campaign/round/trial, progress and scientific scope	Live evaluation and inspection together: worker recovery, stop/cancel, provisional/final rankings, partial results, metrics/plots/refusals, keep/promote, confirmation, verification report, package/key export and Workflow →.

Campaigns places list/lineage on the left and selected-campaign context on the right. Results combines the live dashboard and selected-trial inspection rather than requiring another stage change. New round is one continuous form; do not recreate Seed/Proposal/Approval as nested tabs or a mandatory stepper.

Iteration: selecting a retained round opens Results directly. Keep/promote can open a prefilled New round with the chosen seed and retained context; the scientist reviews the proposal and starts from that same page. Finishing a round retains its evidence under the campaign and never automatically adds every trial to Runs. Draft inputs and the selected campaign/round survive tab switches.

Approval is usage-based, not a cost forecast. Show server-reported rounds remaining and the selected trial count, explicitly including the unchanged seed when counted. When the server reports no weekly round limit, say so; unavailable allowance must not be displayed as unlimited or zero. Pro has no monetary cost cap. Do not invent a predicted price or budget to approve. The server rechecks allowance and recorded usage at Start. Allow users to deselect proposed trials before starting; any changed selection requires a new approval. Retain the exact proposal/selection approval receipt, expiry, idempotency and operational safety limits; changes invalidate the old approval. Technical receipt and limit details go in Details.

Independent confirmation remains in Results. Select/freeze the candidate and protocol, review authorization and the irreversible reveal through explicit inline actions, then show progress and the terminal record in the same view. Preserve governed recovery and development-versus-confirmation distinctions. Opening Results must never consume authorization or reveal protected data.

Handoff to Runs: the agreed simple implementation is Workflow → from the selected result in Results, using the existing workflow-opening path. It is a destination action, not a fourth subtab. The scientist deliberately executes and saves that workflow to create a new Run. Preserve campaign/trial/seed links, source data binding and validation scope. A new run is not the original campaign evaluation and does not automatically prove equivalent reproduction.

Add to Runs remains an optional later shortcut, not a prerequisite for this redesign. If implemented, it must create an explicit, idempotent retained record with original evidence and scope; it must not copy metrics onto an unrelated execution or remove the original campaign record. The Workflow → route must remain available, including when direct materialization is unsupported.

Deploy

Subtab	Context	Contents and actions
Application	Selected application name, origin, source and evidence status	Full application selector; model/package information; source workflow/run, local reproduction/import links and optional qualification/evidence. Digests/technical IDs are in Details or copy controls.
Folder Watches	Selected application or explicit all-applications scope; watch/job state	New Watch, existing model binding/settings, enable/disable, optional Try a file, delete, uncertainty record and QC/maintenance controls.
Prediction History	Application/watch filters and visible/total records	Existing saved predictions, per-file status/errors, labels, artifact identity, QC snapshots and exports.

Moving the full application selector to the first tab must not hide which application a watch targets. Keep a compact selected-application indicator/link in the context row and explicit bindings on watch rows. Changing selection must not silently rebind existing watches. Disclose whether a list shows all project applications or only the selected one; don't silently narrow existing lists.

Keep the campaign-evidence yellow star with its exact scope: recorded campaign validation evidence does not certify predictions on new samples. Local dry runs and qualification remain optional. In hosted profiles without folder-watch execution, preserve the honest Workbench handoff and available export/batch paths; don't render active controls that call refused endpoints.

Report

Subtab	Context	Contents and actions
Setup	Workflow and selected run names; report length and data-disclosure choices	Workflow/run filters, selected-run chips, short/detailed option, section choices and explicit row-level plot/data opt-in. Full IDs remain available in Details/copy controls. Generate Report belongs to this view's header.
Preview	Generated report's source identities, scope and generation/staleness status	The actual report, AI narrative attribution where available and Export. Preserve supported PDF/HTML/Markdown/JSON formats and scoped current-workflow Python/notebook export.
ISO Validation	Selected workflow/run, audit availability and walkthrough scope	Existing ISO 17025 walkthrough, assumptions, limitations, Audit link and evidence-pack route. The title is not a compliance or accreditation claim.

Setup describes the next generation request. Preview describes the last generated result. Changing Setup keeps the previous preview with a visible “settings changed” indication until explicit regeneration; exports must identify the preview they export. Keep selected IDs/section choices and row-level disclosure consistent across preview and export. Enabling an AI narrative remains an explicit action with existing privacy/usage behavior, not a side effect of tab navigation. A generated report need not become a new backend model for this layout; a retained generation snapshot is the minimum authority.

Use the shared Data-style page background in all three tabs and a readable document surface in Preview. Preserve the explicit notice that exporting a current workflow's Python/notebook is not historical-run replay. ISO Validation organizes the existing walkthrough and Audit → evidence-pack journey; it must not silently replace analytical qualification or imply that a walkthrough validates the scientist's method.

5. Campaigns as canonical project objects

An optimization campaign is owned by its Project. Organization supplies accounting context, not scientific ownership or the history collection. Reuse the existing long-lived optimization campaign, seeds, examinations, iterations and promotion records. The bounded execution batch remains a distinct execution authority. Do not introduce another competing campaign identity.

Project
  Data · Workflows · saved Runs · Artifacts
  Optimization campaign
    Source workflow/run → seed → examination → iteration → trials/evidence
                                ↑ promotion to a new seed ← selected trial

Required implementation contracts:

Durable, unambiguous project binding and authorization through the project. Source run/workflow project and campaign project must agree. Preserve the existing access policy during migration; accounting membership alone must not expose another project's campaigns.
Project-filtered list/detail/count/search APIs, authorized on the server. A browser filter on an organization-wide response is insufficient.
Resolve billing account, round allowance and recorded usage separately on the server. Retain the account charged for each execution; a project transfer cannot rewrite historic charges. Organization controls do not belong on the scientific campaign page. This separates responsibilities without duplicating campaign records or removing commercial enforcement.
Canonical IDs exposed in Project, Optimize, Memory Map, exported evidence and workflow/run handoffs. Distinguish campaign, iteration, execution batch and trial IDs instead of overloading one campaign_id field.
Draft and running work must survive navigation and restart under the same identity. Idempotency and retained proposal/approval/execution links survive UI retries; a second click does not create another study or charge.
Define project archive, removal, export/import and authorized transfer behavior for campaigns and retained evidence. Never orphan lineage or silently move it to whichever project is active. OSS imports supported scientific packages and evidence; it need not acquire the managed campaign execution service.
Backfill older records only from unambiguous retained source authority. Expose unresolved records through an authorized recovery path. Do not guess ownership from titles or current browser state. Preserve existing access restrictions, including source-run ownership when opening a trial in Workflow.

6. Advisor memory and attention

The durable memory anchor is the authorized project + object identity. Subtab/stage and active node/plot/table describe attention within that object. They must not create unrelated conversation buckets whenever the user moves between New round sections or from Campaigns to Results.

Pass project, page/subtab, resource type/ID, applicable seed/iteration/trial, workflow/run and focused node/plot IDs through the shared context mechanism. The server resolves their contents and checks authorization/disclosure; browser focus is a hint, never access authority. Preserve existing memory-topic choices, history and attribution through scope aliases/migration. Switching project, actor, object or focused plot clears stale context before another AI request.

Memory Map gains project campaign nodes and their source/derived links. Account settings do not become campaign memory. OSS BYOK chat and managed Advisor keep their existing feature/privacy boundaries while using consistent visible focus.

7. Acceptance and change discipline

The design is not complete merely because the pages look consistent. For every moved function record: old source, destination, object identity, action outcome, profile availability and an acceptance journey. Keep dated source inventories separate from this canonical document.

Minimum implementation checks:

All eight pages at wide/narrow sizes and with Advisor open; stable underlying project binding, keyboard navigation and no duplicate Project tiles. The topbar's existing hide-while-Advisor-is-open behavior is allowed. Check short labels, one explanatory sentence per section, working Details toggles and no technical IDs in headers.
Deep links, Back/Forward, reload and project/actor switching; no stale results, wrong-project actions, lost drafts or orphaned memory.
All Data sources and Workflow sheet operations; exact data/target/group binding survives navigation and the canvas sheet tabs remain unchanged.
All four Runs tabs, saved-run detail, comparison, artifact provenance, batch outcomes and direct Dashboard Storage navigation. Reclaim Storage executes only on Dashboard, with across-project scope and retained-output protection.
All three campaign search methods across the three tabs, single-page round creation, approval showing rounds/trials, stop/partial outcomes, confirmation and export. Keep → prefilled New round avoids a stage wizard; completion does not flood Runs. No cost forecast or Pro monetary cap is introduced.
Campaign → Workflow → explicit execution/save → Runs; verify linked identities and distinguish development evidence from a fresh execution.
Cloud package → local import/bind/reproduce → Application → Folder Watches → Prediction History, with optional qualification and no new local enable gate.
Report Setup → Generate → Preview → export, changed-Setup disclosure, row-data opt-in and ISO walkthrough → Audit → evidence pack.
OSS, Demo and Pro capability differences; navigation does not create new entitlements or hide promised features. Existing administrative/system pages remain reachable for authorized users.

Use the scientific contract's deliberately dissimilar data cases as well as these navigation journeys. A source audit establishes a destination for each function; only implementation tests and browser evidence establish it works.

Appendix A. Changes from the reviewed frontend

Baseline: monorepo 444232f781261f88b187a4acb42ceffc1007a017 (merged #1214). Paths below are relative to the repository; FE means packages/spectra-sherpa/frontend/src, Managed FE means packages/spectra-server/frontend/src, and Server means packages/spectra-server/src/spectrasherpa_server.

Area / current source	Current arrangement	Required change / preservation condition
`FE/components/Topbar.vue`	Project selector hidden when chat is open	Keep this accepted behavior; preserve the underlying project and Advisor scope when removing page duplicates.
`FE/layouts/MainLayout.vue`, `FE/components/Sidebar.vue`, server module navigation	Shared shell plus contributed pages	Use one page/subtab/action grammar; keep status, jobs, version, system pages and profile authority.
`FE/views/dashboard/SimplifiedDashboard.vue`	Current-project icon arrow; no storage action	Label Project →; preserve project management; host account-wide storage/reclaim under the Storage subtab.
`FE/views/project/ProjectContent.vue`	Narrow description; Edit/Export below; artifact list/count; no campaign collection	Wider identity; header actions; distinct Runs/Artifacts/Campaigns lists and counts; keep import/reproduction controls.
`FE/views/data/DataContent.vue`	Duplicate Project cell	Remove cell, align tabs; retain every data view and workflow-binding action.
`FE/views/workflow-builder/WorkflowBuilderContent.vue`	Project/Sheet/Data/Canvas context; canvas workbook tabs	Remove Project, enlarge Sheet/Data, right-align Canvas, normalize gear height; keep canvas tabs/operations.
`FE/views/models/ModelsContent.vue`	History/Compare/Models/Batch; storage/qualification above	Rename Models to Artifacts, keep Batch, move Reclaim Storage to Dashboard; remove the Runs storage link; use subtab context. Qualification remains accessible through Deploy.
`FE/views/models/RunDetailContent.vue`	Summary/Results/Validation detail views	Preserve under selected-run inspection, including exact saved evidence and Report/Optimize actions.
`Managed FE/campaign/views/CampaignsView.vue` and children	Launch, planning, allowance review, confirmation, history and monitor intermixed	Remap to Campaigns/New round/Results; one-page seed/proposal/approval, combined live/inspection dashboard. Show rounds/trials, not a cost forecast; preserve all search methods and reduce copy.
`Server/models/optimization_campaign.py`, `Server/routes/harness.py`	Persistent campaign stored with org ownership; list lacks project filter	Make Project the scientific owner/list/access scope; retain organization as accounting context, bounded execution batches and lineage.
`Server/models/advisor_memory.py`, `Server/memory_map.py`, `FE/lib/sherpaAttention.ts`	Existing project/tab/subscope memories; partial focus IDs	Evolve scopes with aliases; add campaign relationships and precise stage/resource focus without orphaning existing topics.
`FE/views/deploy/DeployContent.vue`	Application catalog above Folder Watches/Prediction History	Put catalog/qualification under first Application tab; keep compact context and per-watch binding elsewhere.
`FE/views/deploy/QualificationPanel.vue`, `InstrumentQCPanel.vue`, Data acquisition panel	Nested Refresh/Reload actions	Synchronize after changes/on entry; retain explicit error Retry and all evidence/export/edit functions.
`FE/views/report/ReportContent.vue`, `FE/utils/reportGenerator.ts`	Setup/preview/walkthrough on one page; dark iframe and generated HTML styling	Split Setup/Preview/ISO Validation; retain report-length/section/data options; align page and generated preview theme.
`FE/views/report/ValidationWalkthroughPanel.vue`	Walkthrough beneath report	Move to ISO Validation; preserve Audit/pack path and limitations, no certification claim.
Router, stores and managed navigation callbacks	Index/alias-based tabs and different object query conventions	Stable tab IDs; retain legacy links and object context; prevent late-response/project leakage.
Historical `docs/plan/runs-navigation.md`	Earlier two-tab plan	Keep as historical evidence and point to this four-tab authority.

Appendix B. Implementation order

Shared shell and stable project/state/command contracts.
Canonical campaign/project authority, source scoping and memory migration plan.
Data/Project/Workflow/Runs layout, preserving four Runs tabs and canvas tabs.
Optimize three-tab presentation over existing services, including all confirmation and workflow/export paths; no new scientific or commercial authority in UI.
Deploy and Report layouts, then the cross-page acceptance journeys above.

These are dependency boundaries, not authorization to implement or a prescribed PR count. Do not delete a current path until its mapped replacement is reachable.

2026-10-02 visual correction

Shared page headers and context rows own vertical spacing. Page containers do not add a second flex gap; use Workflow’s compact title/context/content rhythm.
Every subtab label has a visible rim on all sides. The selected tab uses the primary border and a light tint; retain keyboard focus and disabled states. Remove page-specific tab rules that override the shared component.
Dashboard has Your Project (default) and Storage subtabs; project archive/restore remains with projects. Reclamation remains account-wide and never runs merely by opening a tab.
Remove “Storage on Dashboard” from Runs.
Report uses the same workspace background as Data, Runs and Deploy. The report document itself retains its print-friendly presentation.