feat: captureDashboard screenshot helper + fix 16 Phase-1011 regressions surfaced by visual audit

[HanSur94]

Summary

Two intertwined workstreams in one branch:

1. Add captureDashboard screenshot helper — a pure-MATLAB primitive that renders a DashboardEngine (or a single widget) to a PNG. Designed for agentic UI testing via matlab-mcp: build a dashboard → capture → Read the PNG back into the agent for visual verification. Covers MATLAB R2024a+ (exportapp, with crop-to-panel fallback for uicontrol-only widgets), MATLAB R2020a–R2023b (exportgraphics), and Octave (print + stub-axes).

2. Fix 16 broken examples that visual audit caught. Phase 1011 (Tag-based domain model) migrated Sensor → SensorTag but left the widget/example layer partially updated. All caught by running each example through captureDashboard and inspecting the PNG.

What's new

libs/Dashboard/captureDashboard.m (+ test + example)

• captureDashboard(engine, path) — full dashboard
• captureDashboard(engine, path, 'Widget', titleOrHandle) — single widget (embedded or detached)
• Accepts figure handles directly as well
• Backend dispatch mirrors DashboardEngine.exportImage: exportapp → exportgraphics → Octave print+stub-axes
• tests/test_capture_dashboard.m — 4 function-style Octave-safe tests
• examples/03-dashboard/example_capture_dashboard.m — documents the matlab-mcp workflow

libs/SensorThreshold/SensorTag.m

• Added default Thresholds = {} public property as a compat shim. GaugeWidget, StatusWidget, ChipBarWidget, MultiStatusWidget, and IconCardWidget still probe sensor.Thresholds as a fallback colour/range source — the property was removed in Phase 1011 but the widget probes weren't. Now the guard ~isempty(sensor.Thresholds) cleanly evaluates to false instead of raising.

libs/Dashboard/DashboardSerializer.m

• configToWidgets now honours the SensorResolver callback for v2.0 layouts. Before: matched only source.type='sensor' (v1.x shape). After: also matches source.type='tag' with source.key (current FastSenseWidget output). Fixes example_dashboard_engine which silently reloaded dashboards with widget.Tag = [] and then failed render() with "Add at least one line before render()".

12 example scripts migrated off legacy Sensor.X/.Y setters

Since SensorTag.X and .Y are read-only dependent properties in the v2.0 model, assignment like sTag.Y(end) = 76 raised. Migrated to the constructor-inline pattern:

% Before
sTemp = SensorTag('T-401', ...);
sTemp.X = t;
sTemp.Y(end) = 76;

% After
yTemp = 70 + 4*sin(2*pi*t/3600) + randn(1,N)*0.8;
yTemp(end) = 76;
sTemp = SensorTag('T-401', 'X', t, 'Y', yTemp);

Files: example_widget_{fastsense,gauge,group,heatmap,histogram,iconcard,multistatus,scatter,sparkline,status}.m, plus example_widget_table.m and example_dashboard_all_widgets.m (replaced ResolvedViolations / countViolations callers with peak-sample / min-max summaries since those APIs no longer exist in the Tag model — MonitorTag owns threshold state now).

Why

Without captureDashboard, visual-regression checking on MATLAB dashboards was manual. With it, the agentic loop becomes: build → render headless → PNG → Read → inspect — one turn, no human in the loop.

The Phase-1011 regressions had been sitting in main without anyone noticing because the existing test suite mostly covered library internals, not end-to-end example rendering. The very first audit run caught them all.

Test plan

• tests/test_capture_dashboard.m → 4/4 pass
• tests/test_dashboard_toolbar_image_export.m → 4/4 pass (regression)
• tests/test_sensortag.m, test_monitortag.m, test_compositetag.m, test_fastsense_widget_tag.m → pass
• All 16 previously-broken examples render via captureDashboard end-to-end through matlab-mcp
• Spot-checked rendered PNGs for visual correctness (gauge, scatter, dashboard_info, etc.)
• Pre-existing failures in test_multistatus_widget_tag and test_icon_card_widget_tag — both reference a Threshold(...) constructor that doesn't exist in the codebase. These failures predate this branch (Phase 1011 test debt, tracked separately).

Notes for reviewers

• SensorTag.Thresholds = {} is intentionally a shim, not a new supported API. The right long-term fix is to migrate the five dashboard widgets off sensor.Thresholds and onto explicit MonitorTag bindings — but that's a bigger refactor and out of scope here.
• The FastSenseWidget.fromStruct path emits TagRegistry key not found warnings during reload when a SensorResolver is provided. The warning is cosmetic (the resolver fixes up obj.Tag on the next step), but could be silenced in a follow-up.
• example_dashboard_live.m uses sTag.X(end+1) = ... streaming-append in its live-timer callback. Not migrated in this PR since the timer path needs dedicated work; captured as an outstanding item.

🤖 Generated with Claude Code

[github-actions]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'FastSense Performance'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.10.

Benchmark suite	Current: 2517e61	Previous: e09e7fa	Ratio
Instantiation mean std(1M)	2.365 ms	0.999 ms	2.37
Render mean std(1M)	3.323 ms	2.917 ms	1.14
Render mean std50M)	3.611 ms	0.608 ms	5.94
Downsample mean ( std00M)	0.85 ms	0.049 ms	17.35
Downsample mean ( std00M)	34.576 ms	0.049 ms	705.63
Instantiation mean ( std00M)	9845.746 ms	40.042 ms	245.89
Render mean ( std00M)	479.231 ms	2.905 ms	164.97
Dashboard broadcastTimeRange mean	0.227 ms	0.11 ms	2.06
Dashboard broadcastTimeRange stdmean	0.157 ms	0.025 ms	6.28

This comment was automatically generated by workflow using github-action-benchmark.

CC: @HanSur94

[HanSur94]

Closing in favor of a narrower PR containing just the captureDashboard helper.

This PR originally bundled two things:

1. The new captureDashboard screenshot helper for agentic UI testing
2. Fixes to 16 examples caught by the very first visual audit using that helper

While this PR was open, PR #57 landed on main with equivalent fixes for the example regressions (commit 30927c7 "fix(examples,style): migrate examples to post-SensorTag API + clear lint") — so the second track is now redundant. This PR went dirty with merge conflicts as a result.

I'll reopen with a clean branch containing only the captureDashboard helper + its test + its example. That's the unique value that isn't already on main.