Files
scadaproj/impeccable/OtOpcUa.md
T
2026-11-13 14:40:42 -05:00

252 lines
28 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# OtOpcUa — Technical-Light UI recommendations
Reviewed 2026-11-12. Mode: **Operate**. Source baseline: `412da27c5fb7e440e4a75d4f8e82e0e5b9832411` in `/Users/dohertj2/Desktop/OtOpcUa`. This is a source-backed audit and design proposal, not a claim of a live-browser usability or WCAG certification.
[Review gallery](index.html) · [Fleet mockup](mockups/OtOpcUa.html) · [Raw workspace mockup](mockups/OtOpcUa-Raw.html) · [Detector evidence](evidence/OtOpcUa-detector.json)
## Recommendation
Keep the shared Technical-Light identity. OtOpcUa already delegates its shell to the common theme; its largest gaps are in the task surfaces inside that shell. Give operators a useful landing page, a clearly grouped rail, accessible trees and dialogs, a consistent page/table grammar, and readable failure details. Retain the Raw/UNS distinction and the fact that editing configuration does not deploy it.
The mockups propose the same neutral paper, IBM Plex, restrained blue action accent, hairline tables and compact rail used by the family. Green, amber and red carry operational meaning, always accompanied by words. They introduce no alternate brand, dark theme or new product claims.
## Scope and evidence
The shipped web surface is `src/Server/ZB.MOM.WW.OtOpcUa.AdminUI`, hosted by `src/Server/ZB.MOM.WW.OtOpcUa.Host`. `src/Client/...Client.UI` is an Avalonia desktop client and is outside this web review. The inspected `docs/images/connection-panel.png` depicts that desktop client; it is not a current web screenshot and was excluded as visual authority. No current web regression golden was identified. Runtime services and authenticated pages were not started or exercised.
Package `ZB.MOM.WW.Theme` is centrally pinned at **0.5.0** in `Directory.Packages.props:244`. `Components/App.razor:16–23` loads Bootstrap, then `ThemeHead`, then the small local stylesheet and scoped styles. `Components/Layout/MainLayout.razor:15` uses `ThemeShell`; there is no hardcoded shell accent overriding future theme-level changes. The current theme source in scadaproj can be newer than this consumed package; validate the release actually consumed before removing any compatibility rule.
Paths below are relative to `/Users/dohertj2/Desktop/OtOpcUa/`. For brevity, **UI** means `src/Server/ZB.MOM.WW.OtOpcUa.AdminUI/`.
### Coverage inventory
All 32 page files (31 routable components plus one development sample) were inventoried for route, policy, title/heading, table, state and component patterns. The 31 routable components declare **36 route templates**, because create/edit aliases share components; the machine inventory counts these templates separately. The development sample has no `@page` directive. Deep means markup and relevant behavior were read; scan means route/markup/state signatures were checked, with representative components reviewed. This is not a promise that every driver protocol's full runtime workflow was tested.
| Surface | Routes / files | Depth and next validation |
|---|---|---|
| Shell, auth and account | `/`, `/login`, `/account`; `App`, `Routes`, main/login layouts | Deep. Check signed-in/out, long user names, policy-specific navigation, SSR login on a running host. |
| Fleet and runtime triage | `/fleet`, `/fleet/diagnostics`, `/hosts` | Deep. Verify polling failure, stale readings, large driver fleets and refresh/focus behavior live. |
| Raw authoring | `/raw`; `RawTree`, context menu, name/confirm/browse/import/tag dialogs | Deep tree, common dialogs and driver modal; scan specialized pickers. Preserve virtualization and lazy loading. |
| UNS authoring | `/uns`, `/uns/equipment/new`, `/uns/equipment/{EquipmentId}` | Deep tree and page/tabs; scan per-protocol tag editors, virtual-tag and alarm dialogs. Validate many references and long equipment paths. |
| Cluster management | `/clusters`, `/clusters/new`, `/clusters/{ClusterId}`, `/edit`, `/nodes/new`, `/nodes/{NodeId}` | Markup/state scan; representative creation/navigation patterns checked. Validate field errors and role-specific actions. |
| Cluster access and history | `/clusters/{ClusterId}/acls`, `/acls/new`, `/acls/{NodeAclId}`, `/audit`, `/redundancy` | ACL/audit/navigation deep; edit and redundancy scanned. Test grant scope, revoked principal hints and detailed failures. |
| Retired namespace routes | `/clusters/{ClusterId}/namespaces`, `/namespaces/new`, `/namespaces/{NamespaceId}` | Deep; compatibility notice retained, ordinary navigation should stop advertising retired work. |
| Deployment | `/deployments` | Deep markup, query and action feedback; test zero records, in-flight and degraded seal live. |
| Security and IDs | `/apikeys`, `/role-grants`, `/certificates`, `/reservations` | API-key/role/certificate/reservation markup and state scan; shared form/dialog findings apply. Keep one-time key reveal and UTC expiry semantics. |
| Scripting | `/scripts`, `/scripts/new`, `/scripts/{ScriptId}`, `/script-log` | Deep list/log and editor scaffold; Monaco bridge scanned. Test keyboard/editor fallback on the running host. |
| Operational events | `/alerts`, `/alarms-historian` | Deep markup/state scan; live connection, retained stale reading and bounded tail behavior inspected. |
| Shared secrets | `/admin/secrets` | Wrapper deep; shared Secrets.Ui internals reviewed separately in the family report. Preserve this host's per-page interactive boundary. |
| Driver families | Modbus, S7, AbCip, AbLegacy, TwinCAT, Focas, Heidenhain, OpcUaClient, Galaxy, Sql, Mqtt, MTConnect, Calculation; driver/device forms, address pickers, tag editors | Common modal/form dispatch deep; Modbus representative form deep; other family markup/pattern scan. Protocol correctness outside scope. |
| Development sample | `Dev/ContextMenuDemo.razor` | Scanned; no routable `@page` was present. Do not mistake this sample for a shipped navigation destination. |
## Provisional audit score
Scores summarize static evidence and do not imply measured live compliance. Performance is limited to implementation patterns, not latency or memory profiling.
| Dimension | Score / 4 | Evidence |
|---|---:|---|
| Accessibility | 1 | Repeated unattached labels, glyph-only tree expanders, incomplete hand-rendered modal semantics. |
| Performance | 3 | Raw virtualization, lazy loading, capped/coalesced event tails, overlap-guarded polling and scalar deployment projections. Runtime budget unmeasured. |
| Responsive design | 2 | Shared responsive shell and many table wrappers; nowrap log toolbar and unwrapped deployment/equipment tables need correction. |
| Theming | 3 | Actual shared package adoption; local chip aliases and mixed Bootstrap/shared tables still diverge. Light-only is intentional, not a missing dark-mode defect. |
| Implementation integrity | 3 | Coherent operational product; home placeholder and retired namespace navigation weaken task continuity. Detector hits were vendor code only. |
| **Total** | **12/20** | **Significant targeted improvements; preserve the identity.** |
### Detector interpretation
The bundled detector was run once on the AdminUI source and emitted **14 warnings**, all under vendored `wwwroot/lib/bootstrap` or `wwwroot/lib/monaco`. Bootstrap caret border geometry, generic library collapse transitions, Monaco CSS-language examples and generated `<img>` strings are not evidence of a broken OtOpcUa screen. They are recorded in the raw evidence, not promoted to product defects. No app-authored deterministic detector warnings appeared. The manually verified issues below are separate findings; a quiet detector does not establish accessible interactions.
## Prioritized findings
No P0 task-blocking defect was proven. Findings: **3 P1, 8 P2, 1 P3**. Shared token contrast and shared field-component semantics also apply, and are owned by the family theme report rather than counted again here.
### OT-01 · P1 · Programmatically label forms and filters
**Evidence:** `UI/Components/Shared/Drivers/DriverConfigModal.razor:28–30` renders the driver name label and input without association. `.../Drivers/Forms/ModbusDriverForm.razor:17–23` repeats this for Timeout and PLC family, and subsequent numeric fields follow the pattern. `UI/Components/Pages/ScriptLog.razor:16–25` has an unnamed select and a placeholder-only search field. `.../Pages/Uns/GlobalUns.razor:21–23` has a placeholder-only filter. Some fields already do this correctly, e.g. API key fields at `.../Pages/ApiKeys.razor:35–46`.
**Impact:** Screen reader users cannot reliably determine the input purpose; placeholder text disappears while typing. WCAG 1.3.1 / 3.3.2 and, where no accessible name exists, 4.1.2.
**Ownership:** Local instances; shared Theme should provide a robust field ID/hint/error association contract.
**Recommendation:** Add stable unique `id`/`for`, visible filter labels, `aria-describedby` for help/errors and `aria-invalid` for invalid values. Prefer the enhanced shared field wrapper once available; do not assume visually adjacent text labels an input.
**Acceptance:** Accessible-name inspection identifies every input/select by purpose across all protocol forms, including two instances of a reusable form. Clicking a label focuses its control; field errors remain linked after failed save. `$impeccable harden`.
### OT-02 · P1 · Give every dialog one accessible focus lifecycle
**Evidence:** `UI/Components/Shared/Raw/RawConfirmDialog.razor:8–30` and `.../Shared/Drivers/DriverConfigModal.razor:11–18` create visible Bootstrap-shaped divs using `display:block` and `role="dialog"`, without `aria-labelledby`, `aria-modal`, focus initialization/trap/return or Escape handling in their component code. The UNS inline delete at `UI/Components/Pages/Uns/GlobalUns.razor:76–105` repeats the pattern. Loading Bootstrap JS does not activate its modal manager for a div shown manually.
**Impact:** Keyboard users can remain behind the modal, lose context or reach background controls while a destructive decision is visible. Potential WCAG 2.4.3 / 4.1.2 failures require runtime confirmation.
**Ownership:** Shared visual/accessibility primitive in Theme, adopted locally; domain delete and validation behavior remain local.
**Recommendation:** Standardize a dialog primitive with labelled title/description, modal background isolation, appropriate initial focus, Escape/close rules, focus containment and restoration. Preserve busy safeguards and validation content. Full driver configuration should graduate to a routed editor when users need prolonged work; keep short confirmation dialogs.
**Acceptance:** Open by keyboard, cycle forward/backward, cancel, and reopen; background controls remain unreachable, focus returns to trigger, title is announced, errors preserve input, and confirmation focus defaults safely. Test with nested Monaco suggestions and narrow scrolling. `$impeccable harden`.
### OT-03 · P1 · Name and expose tree expansion state
**Evidence:** `UI/Components/Shared/Raw/RawTree.razor:210–216` and `.../Shared/Uns/UnsTree.razor:44–50` render triangle-only buttons without a node-specific accessible name or `aria-expanded`. The raw picker checkbox at `RawTree.razor:242–244` has no label association.
**Impact:** Assistive technology announces ambiguous glyph controls and cannot discover which branch is expanded or which tag is selected. WCAG 4.1.2; picker naming also affects 1.3.1.
**Ownership:** Local tree semantics with a shared expand-control visual pattern. Theme's `TreeTable` is a candidate pattern, not a drop-in replacement for the virtualized Raw tree.
**Recommendation:** Give each toggle a node-specific name, accurate `aria-expanded` and appropriate controlled-region semantics; label every tag checkbox. Either implement a full keyboard tree pattern or retain semantic nested lists with real buttons; do not add `role="tree"` without its keyboard contract. Keep virtualization and lazy loading.
**Acceptance:** A screen reader can identify the node and collapsed/expanded state, select a named tag, and recover from a lazy-load error. Focus survives a visible-row refresh and Show more. `$impeccable harden`.
### OT-04 · P2 · Replace the placeholder entry page with task orientation
**Evidence:** `UI/Components/Pages/Home.razor:6–7` is just “OtOpcUa Admin” and “Use the nav above…” although navigation is a side rail.
**Impact:** Every operator starts with an empty page and must infer whether to check fleet, hosts or deployments. New users receive an inaccurate spatial instruction.
**Ownership:** Local composition using shared page-header, notice, table and status components.
**Recommendation:** Start with an operational overview of known fleet/deployment context and links to existing tasks; show time and source of data. Prefer an exception summary and compact fleet table over decorative metric tiles. Gate authoring actions to the user's policy. When no data is available, provide a precise next step such as open Fleet or create a cluster according to role.
**Acceptance:** An operator can reach fleet diagnostics, a configuration author can reach Raw/UNS, and a deployer can inspect drift within one action; no unavailable or stale value appears healthy. The Fleet mockup illustrates the triage pattern, not an existing aggregate endpoint. `$impeccable onboard`.
### OT-05 · P2 · Group navigation by task and honor page policies
**Evidence:** `UI/Components/Layout/MainLayout.razor:17–40` puts 11 links into one Navigation section; Role grants and API keys are visible to roles their page policy rejects, while Secrets is policy-gated. The source itself documents that inconsistency at lines 33–37. `UI/Components/Shared/ClusterNav.razor:13` still advertises Namespaces; `.../Pages/Clusters/NamespaceEdit.razor:14–19` says it is retired in v3.
**Impact:** High scan cost, advertised dead ends and misleading legacy authoring destinations.
**Ownership:** Local navigation taxonomy, shared rail mechanics. Hiding a link supplements server authorization; it never replaces it.
**Recommendation:** Use shared sections for Operations (overview/fleet/hosts/live), Configuration (clusters/Raw/UNS/reservations), Scripting, and Administration (keys/roles/certificates/secrets); keep Deployments close to configuration outcomes. Gate links with their actual policies. Remove Namespaces from ordinary tabs while retaining compatibility URLs with working Raw/UNS links. Use `ScrollNav` after validating the consumed Theme release supports it; handle long rails without hiding the session block.
**Acceptance:** Operator, Designer and Administrator each see only meaningful destinations. Existing deep links remain explainable; active branch is evident; footer remains reachable on short screens. `$impeccable clarify`.
### OT-06 · P2 · Unify page headings, document titles and table anatomy
**Evidence:** `UI/Components/Pages/Deployments.razor:21,43–68` uses an h1 plus a bare Bootstrap striped table; Fleet uses h4 and `data-table` inside `table-wrap` at `.../Pages/Fleet.razor:21,67–108`. `.../Pages/Scripts.razor:29` and `.../Pages/Uns/EquipmentPage.razor:36–41` use other Bootstrap patterns. Fleet, Hosts and most admin pages do not set a page-specific `PageTitle`.
**Impact:** Similar pages have different density and weight, and browser tabs do not consistently identify the current task. Heading levels express visual size rather than document structure.
**Ownership:** Shared header/table grammar; local page adoption.
**Recommendation:** One semantic h1 at a shared compact size, consistent description/actions placement, h2 for sections, a product-qualified page title, labelled tables with header scopes and accessible row actions. Put deployment/script/equipment tables in the same overflow region/chassis as fleet. Tables may have domain-specific columns; they should not have separate visual systems.
**Acceptance:** All routes have one primary heading and a meaningful browser title. Keyboard and screen-reader users identify table purpose and action context. Side-by-side Fleet, Deployments and Scripts share typography, border, cell padding and action sizing. `$impeccable layout`.
### OT-07 · P2 · Let filters wrap and keep touch targets usable
**Evidence:** `UI/Components/Pages/ScriptLog.razor:13–30` puts title plus an unwrapped cluster of select, 200px input, connection pill and button on a single flex row. `.../Shared/Raw/RawTree.razor:213` uses an 18px-wide twist column (`wwwroot/css/site.css:47–49`) and zero padding. `.../Shared/ContextMenu.razor.css:21–22` fixes ellipsis controls at 1.75rem.
**Impact:** The toolbar can overflow or compress unreadably; adjacent tree controls are difficult on touch. These source dimensions establish a risk, not a measured viewport failure. WCAG 1.4.10 and 2.5.8 require rendered checks; 44px is the family comfortable touch target, not the WCAG AA minimum.
**Ownership:** Shared responsive toolbar/action spacing; local tree density.
**Recommendation:** Use wrapping toolbar groups, flexible/min-width-zero fields and stacked small-screen headers. Keep horizontal scrolling inside labelled data regions. Enlarge interactive hit areas on coarse pointers, preserving compact desktop rows and virtualization height estimates; do not enlarge CSS row heights while retaining `ItemSize="30"` unchanged.
**Acceptance:** At 390px and 200% zoom the document itself has no horizontal overflow; controls remain labelled and reachable. On coarse pointers primary controls have 44px targets. Test touch expansion with adjacent siblings. `$impeccable adapt`.
### OT-08 · P2 · Make failure details available without hover
**Evidence:** `UI/Components/Pages/Fleet.razor:103` abbreviates failure text at 80 characters into a title tooltip. `.../Pages/Deployments.razor:61` puts the degraded reason only in `title`. `.../Pages/ScriptLog.razor:73` abbreviates messages into title; `.../Pages/Clusters/ClusterAudit.razor:56–58` truncates details JSON into a single line with hover title.
**Impact:** Touch and keyboard users cannot reliably read the reason or copy the complete diagnostic content. An operator may see “degraded” without a recovery path.
**Ownership:** Shared row-detail disclosure presentation; domain content and diagnostics links local.
**Recommendation:** Provide “View details” or expandable rows with full wrapping message, UTC timestamp, identity/correlation and copyable payload. Keep compact summaries in the table. Preserve the failure reason, and offer the existing diagnostics link where relevant. Long payloads should scroll within their own region.
**Acceptance:** Full error and audit content is reachable by keyboard and touch without hover, remains readable with a 2,000-character message, and can be copied without truncation. `$impeccable harden`.
### OT-09 · P2 · Distinguish accepted deployment from completed application
**Evidence:** `UI/Components/Pages/Deployments.razor:133–141` marks `Accepted` as success and announces “dispatched”, then reloads once; the page has no subsequent timer. Initial `_deployments` is empty (`:84`), so loading and genuinely empty history render the same table. Domain code already distinguishes degraded sealed results at `:60–61`.
**Impact:** Dispatch feedback is technically truthful but the UI does not help users follow the eventual apply/seal outcome; an empty-looking table can hide an in-flight initial read.
**Ownership:** Local deployment workflow; shared busy/status/empty state pattern.
**Recommendation:** Use explicit “Request accepted — deployment in progress” feedback, show the current request in a progress region, then refresh until a terminal state or offer a clear manual Refresh. Show initial loading, no deployments, rejected/no changes, dispatch error, applying, sealed and sealed-degraded distinctly. Preserve per-node apply semantics; never equate “sealed” with clean operation.
**Acceptance:** A slow acceptance shows progress, a no-change response is informational, and a degraded terminal result exposes per-node reason and diagnostics without a browser reload. Polling stops/backs off appropriately and is disposed on navigation. `$impeccable harden`.
### OT-10 · P2 · Clarify tree/filter scope and empty-state recovery
**Evidence:** `UI/Components/Shared/Uns/UnsTree.razor:28–31` documents that the filter can only consider lazy tag children after loading. `.../Pages/Uns/GlobalUns.razor:21–23` calls it only “Filter by name”. Raw root errors at `.../Pages/Raw/GlobalRaw.razor:39–45` have no retry action; no-cluster states do not link to cluster creation.
**Impact:** Users may read no matching loaded item as proof that a tag does not exist, or be stranded after a transient load error.
**Ownership:** Local tree data behavior and recovery; shared filter/empty/error state visual grammar.
**Recommendation:** Label “Filter loaded items” until server-side whole-tree search exists; explain exactly what is searched and distinguish no matches from no data. Provide Clear filter and retry actions. Link the no-cluster state to a policy-appropriate next step. A proposed global search must be separately implemented and must not imply unloaded-tree coverage today.
**Acceptance:** Match count and scope are visible, Clear restores context, load error can be retried, and no-data text does not claim absence of unloaded records. `$impeccable clarify`.
### OT-11 · P2 · Standardize event-tail control and stale-data communication
**Evidence:** `UI/Components/Pages/Alerts.razor:22–30` and `.../Pages/ScriptLog.razor:26–29` expose live/disconnected and Clear, but no pause reading control. Fleet shows last refresh (`.../Pages/Fleet.razor:29–30`) while failures outside cancellation can escape to its error boundary; Alarms historian already preserves last status plus read time when unavailable (`.../Pages/AlarmsHistorian.razor:36–40`).
**Impact:** Incoming rows can move the content being read. Users need to distinguish disconnected transport, old retained information and a healthy empty window.
**Ownership:** Shared freshness/live-status pattern; local subscription and retention semantics.
**Recommendation:** Replicate the historian's honest stale-read behavior where feasible. Add a clearly labelled Pause updates/Resume control to live tails and define retention while paused. Announce connection/error summaries through a polite region rather than announcing every event row. Rename Clear to “Clear visible log” or equivalent so it cannot imply deleting historian records.
**Acceptance:** Reading can be paused without losing scroll/focus, Resume reconciles to the bounded tail, disconnected differs from zero events, and stale data shows last successful read. Stress-check the capped/coalesced buffer rather than replacing it with unbounded accumulation. `$impeccable harden`.
### OT-12 · P3 · Normalize technical iconography and status aliases
**Evidence:** `UI/Components/Shared/Raw/RawTree.razor:259–268` maps kinds to platform emoji; `.../Shared/Uns/UnsTree.razor:99` embeds a gear glyph in a link. `UI/wwwroot/css/site.css:24–25` maintains chip aliases alongside Theme's status variants. `.../Shared/ToastNotification.razor:15` has an unlabelled dismiss button; include that naming fix in OT-01 rather than treating it as decorative polish.
**Impact:** Emoji shape, color and alignment vary by OS, unlike the disciplined shared chrome. Aliases increase drift risk.
**Ownership:** Shared icon/control naming conventions, local migration.
**Recommendation:** Use one family icon set or simple shared SVG controls, hide decorative icons, and retain visible text. Converge local alias call sites on shared semantic variants after checking visual contrast in the theme release consumed. Do not invent alternate meaning for caution vs warning.
**Acceptance:** Same object/action has the same icon/stroke across sisters; no emoji substitutes for accessible state text; chips use canonical semantic variants. `$impeccable polish`.
## What to retain
- Genuine use of `ThemeHead`, `ThemeShell`, `ThemeScripts` and `LoginCard`; vendored fonts/Bootstrap/Monaco support disconnected industrial environments. Do not introduce CDN dependencies.
- Static SSR login with antiforgery token, deliberate per-page interactive boundaries and the `/admin/secrets` wrapper. A global interactive-router conversion would change behavior, not merely style.
- Raw `Virtualize` (`RawTree.razor:30`), lazy child loading, capped Show more and menu caching. Accessibility work should preserve their scale advantages.
- Fleet polling overlap guard, cancellation and disposal (`Fleet.razor:121–131,143–150`); the UI should build on these rather than launch new overlapping queries.
- Explicit sealed-degraded/applied-degraded semantics. Healthy-looking color must not erase partial address-space failures.
- Alarms historian availability distinctions, a retained last reading and its timestamp. This is a good pattern for the family.
- Scalar-only deployment projections and a 50-row limit (`Deployments.razor:98–110`); bounded 500-entry, coalesced script log rather than per-event unbounded rerendering.
## Mockup reading guide
**Fleet triage** keeps the known fleet fields, replaces aggregate tiles with a compact status summary and filters, and adds an expandable full failure explanation. The proposed rail groups existing routes; the local mockup filter searches only its illustrative rows. A Read details disclosure demonstrates keyboard/touch access without pretending to fetch a real snapshot.
**Raw workspace** keeps Enterprise → Cluster → Folder → Driver → Device → TagGroup → Tag and distinguishes authoring from deployment. It demonstrates a visible selected device path and a companion inspector with grouped connection information, while retaining native hierarchy disclosures. Its static selection is illustrative; it is not a working configuration editor. This is a proposed way to reduce context-menu discovery load, not an assertion that Raw currently has live tag values or a new search endpoint.
Both mockups use the exact same shared mockup stylesheet as the other family examples. Values, node names, revisions and times are illustrative. Only local navigation, disclosures and explicitly described sample filtering work; no mock control modifies a fleet.
## Rollout and acceptance plan
1. **Shared theme first:** settle contrast, field associations, page header, labelled status/freshness, responsive toolbar, dialog contract and row details with the family recommendations. Publish one reviewed compatible release; do not copy its CSS into each app.
2. **P1 local adoption:** label all protocol/device/tag fields and filters, fix dialog lifecycle, and name tree expanders/pickers. Validate keyboard and screen reader flows on one representative protocol, then verify every form family using a field/interaction inventory.
3. **Operational consistency:** apply page/table grammar to Fleet → Hosts → Diagnostics → Deployments → Alerts/Script log, then the remaining admin tables. Introduce deployment progress and readable error details with honest stale/empty states.
4. **Navigation and authoring:** replace placeholder home, gate/group rail links, remove retired Namespace tabs, then trial the Raw inspector with expert users. Keep existing URLs and wire contracts stable.
5. **Bounded review:** desktop 1440px, compact 1024px, mobile 390px, 200% zoom; keyboard-only and a screen reader; long node names, 2,000-character errors, no rows, filtered empty, initial loading, request rejection, disconnected/stale, and least-privilege roles. Include big Raw branches and rapid navigation during polling to catch performance/focus regressions.
6. Run `$impeccable audit` after implementation and `$impeccable polish` last. Compare the same Fleet and authoring scenarios against the other family screens before accepting release.
Recommendations are review artifacts only. No production OtOpcUa files were changed, no app was deployed, and source-only scores must be revised when runtime evidence is available.
## Rendered proposal previews
These are proposed UI artifacts with illustrative data, not screenshots of the current application. All use the same shared stylesheet. See [verification and limits](VERIFICATION.md).
### OtOpcUa
[Interactive HTML](mockups/OtOpcUa.html) · [Desktop PNG](evidence/OtOpcUa-desktop.png) · [Mobile PNG](evidence/OtOpcUa-mobile.png)
![Proposed OtOpcUa desktop UI](evidence/OtOpcUa-desktop.png)
### OtOpcUa-Raw
[Interactive HTML](mockups/OtOpcUa-Raw.html) · [Desktop PNG](evidence/OtOpcUa-Raw-desktop.png) · [Mobile PNG](evidence/OtOpcUa-Raw-mobile.png)
![Proposed OtOpcUa-Raw desktop UI](evidence/OtOpcUa-Raw-desktop.png)