DIREKT Responsive Operations and Admin Design

Purpose

Define the world-class privileged operations experience for verification, evidence, field work, complaints, incidents, reviews, commercial exceptions, audit and system health.

The operations portal is not a generic SaaS dashboard and must not imitate the customer marketplace UI. It shares DIREKT brand/trust semantics while optimizing for secure evidence review, decision quality, speed, keyboard efficiency and auditability.

Read with:

  • docs/product/WORLD_CLASS_PRODUCT_AND_AI_PLAN.md;
  • docs/design/VISUAL_COMPLETION_PLAN.md;
  • docs/design/DESIGN_SYSTEM.md;
  • docs/architecture/AI_PRODUCT_ARCHITECTURE.md.

Primary devices

Desktop/laptop is primary for evidence review and consequential decisions.

Tablet supports selected queue/case/field tasks.

Mobile supports field operations, incident acknowledgement and bounded urgent triage, not every administrative function.

Never squeeze a desktop evidence table/viewer into a narrow mobile canvas.

Core desktop composition

Primary verification workflow:

┌────────────────┬────────────────────────┬─────────────────────────────┐
│ Queue / Saved  │ Case / Check Context   │ Evidence + Decision         │
│ Views          │                        │                             │
│                │ Provider facts         │ Secure evidence viewer      │
│ Priority/SLA   │ Requirement/policy     │ Metadata/history            │
│ Assignment     │ Trust/publication      │ Checklist/reasons            │
│ Filters        │ Timeline/audit          │ Decision/escalation          │
└────────────────┴────────────────────────┴─────────────────────────────┘

Pane widths may adapt, but the reviewer must retain context between queue, canonical facts, evidence and action.

Primary groups may include, according to role:

  • Mission Control;
  • Verification;
  • Field;
  • Providers;
  • Complaints/Incidents;
  • Reviews/Appeals;
  • Commercial/Reconciliation;
  • Reporting;
  • Configuration;
  • Audit;
  • System Health.

Navigation visibility is convenience only. Server-side authorization remains authoritative.

Do not expose development stage/workstream/API route labels in production-facing navigation or case content.

Responsive layout

Desktop/expanded

  • persistent navigation sidebar;
  • queue + case + evidence/decision multi-pane where task benefits;
  • resizable/controlled panes only if accessibility and persistence are tested;
  • keyboard shortcuts limited to safe, discoverable actions;
  • high-risk action cannot be triggered by a single accidental shortcut.

Tablet/medium

  • collapsible navigation;
  • queue → detail with optional secondary evidence pane;
  • preserve case identity/context when switching panes;
  • touch targets remain 44–48 CSS px class.

Mobile/compact

Task-focused single column for:

  • field assignments;
  • visit/checklist tasks;
  • urgent incident acknowledgement;
  • bounded triage/status review;
  • safe escalation/contact actions.

Final evidence decisions, dense reconciliation and broad configuration remain desktop/tablet tasks unless explicitly redesigned and authorized.

Mission Control

Dashboard prioritizes operational action, not decorative metrics.

Show, role-permitting:

  • overdue/SLA-risk verification cases;
  • action-required/resubmission backlog;
  • expiring claims/evidence;
  • serious complaints/incidents;
  • field assignments requiring attention;
  • review/appeal backlog;
  • payment/reconciliation exceptions only when active;
  • integration/system queue health;
  • clear ownership and next action.

Metrics link to the exact filtered work queue.

Queue design

Columns/fields as applicable:

  • case/reference;
  • provider/check;
  • priority/severity/risk;
  • age/SLA;
  • evidence completeness;
  • assignment/owner;
  • status;
  • updated time.

Requirements:

  • fast filtering/search/saved views;
  • overdue/critical items cannot be accidentally hidden by default views;
  • bulk actions are narrowly scoped and confirmation-protected;
  • row selection and keyboard focus remain distinct;
  • queue data never exposes evidence contents.

Case workspace

Case header provides:

  • provider/check identity;
  • pathway/category/requirement version;
  • current canonical state;
  • publication impact where relevant;
  • assignment/SLA;
  • escalation status.

Case body provides:

  • canonical provider/check facts;
  • timeline/history;
  • current and prior evidence versions as authorized;
  • related field/registry context where authorized;
  • prior immutable decisions;
  • linked complaint/review/incident context only when role/purpose permits.

Secure evidence viewer

  • private-access banner;
  • synthetic-only committed visual fixtures;
  • watermark/reference if required;
  • zoom/rotate without default download;
  • metadata in a separate structured panel;
  • previous versions with explicit version labels;
  • decision/checklist context accessible without losing evidence position;
  • access audit;
  • short-lived/revocable access contract;
  • automatic session timeout appropriate to risk;
  • no persistent public URL;
  • no evidence copied into AI prompts/models unless that exact use case is separately approved.

Decision design

  • decision buttons separated from navigation/low-risk controls;
  • reason code required where policy requires;
  • no free-text-only final decision;
  • confirmation for high-impact actions;
  • four-eyes/supervisor flow where policy requires;
  • provider-safe message preview;
  • internal notes clearly marked and access-controlled;
  • current evidence/checklist status visible before final action;
  • decisions remain human/deterministic and auditable.

AI may draft provider-safe wording or highlight missing/conflicting items, but cannot enable or select a final outcome automatically.

AI operations copilot

AI is a secondary assistive layer embedded beside canonical case data.

Allowed patterns, when the specific use case is activated:

  • case-history summary;
  • evidence metadata/field candidate summary;
  • missing-checklist suggestions;
  • inconsistency highlights;
  • draft action-required explanation;
  • complaint/review history summary;
  • risk/anomaly advisory signals;
  • audit/search assistance.

Visual rules:

  • clearly label AI-generated synthesis;
  • show canonical/source facts separately;
  • make uncertainty/conflict visible;
  • provide refresh/retry/manual fallback;
  • allow human correction/editing;
  • never style an AI suggestion like an approved decision;
  • never hide source evidence behind an AI summary.

Authority rules:

AI cannot autonomously approve, reject, revoke, suspend, decide appeals/serious complaints, alter roles, move money or execute high-risk overrides.

Every tool/action is re-authorized by the backend and requires human confirmation where consequential.

Complaints, incidents and appeals

Design for seriousness and procedural clarity:

  • severity/SLA/owner prominent;
  • immediate-danger guidance distinct from platform case handling;
  • chronology and evidence separated from allegations/conclusions;
  • provider/customer-safe communications controlled;
  • interim restrictions clearly distinguished from final decisions;
  • appeal history visible to authorized reviewers;
  • AI summaries never become factual findings.

Field workflow

Responsive field flow includes:

  • assignment list;
  • provider identity confirmation;
  • safety details;
  • approved navigation link;
  • appointment/arrival/departure states;
  • offline draft;
  • photo/evidence guidance;
  • structured checklist;
  • explicit submit;
  • sync/retry/conflict state.

Field observations are advisory evidence and cannot create public trust claims directly.

Commercial, reporting and configuration

  • commercial/payment state uses a visual family distinct from trust;
  • reconciliation views display ledger/provider facts without implying trust;
  • configuration changes are versioned/audited;
  • taxonomy/evidence-rule changes require explicit governance;
  • AI may propose trends or taxonomy improvements but cannot mutate configuration automatically;
  • exports remain purpose/role restricted.

Accessibility and efficiency

Required:

  • complete keyboard access for desktop work;
  • visible focus and logical order;
  • screen-reader semantics;
  • 200% zoom/reflow without losing critical actions;
  • status not colour-only;
  • 44–48 CSS px-class targets for touch-capable layouts;
  • non-hover alternatives;
  • no auto-advancing evidence/decision state;
  • reduced-motion support;
  • clear destructive/high-impact confirmations.

Privacy and security presentation

  • display only role/purpose-authorized data;
  • mask/minimize sensitive identifiers unless operationally necessary;
  • exact private coordinates appear only where the authorized task requires them;
  • evidence and private notes never appear in public screenshots/fixtures;
  • environment status is present where necessary but visually secondary;
  • no direct privileged database/Supabase path;
  • no client-held model-provider secrets/prompts.

Visual completion evidence

For VC6/VC8 record synthetic-safe references for:

  • mission control;
  • verification queue;
  • case/evidence/decision workspace;
  • action-required/resubmission;
  • complaint/incident;
  • field compact flow;
  • audit/history;
  • loading/empty/error/session-expired states;
  • AI active/fallback state where applicable;
  • keyboard/focus and large-text/zoom behavior.

Completion gate

An operations slice is complete only when:

  • it follows approved Design DNA with appropriate operations density;
  • reviewer context is clearer and safer than the current prototype/test-harness presentation;
  • evidence access/authz/audit boundaries remain intact;
  • AI remains advisory and human accountability is explicit;
  • applicable portal/backend/database/OpenAPI/security/accessibility/AI-evaluation tests pass on the exact head;
  • synthetic-safe visual-reference evidence is recorded.