[prod] App Profiling Insights: dashboard reference + Comparison tab docs

[hardikshah197]

Rewrites the App Profiling Insights page as a reference for what the dashboard actually does, and documents the Comparison tab.

The previous version was advisory rather than descriptive — Understanding Performance Metrics Widgets, Advanced Analysis Techniques, Correlation Analysis, Performance Budgeting and Continuous Monitoring explained performance practice in general but never told a reader what any control on the page does. Replaced with a walkthrough of the real UI.

What the page now covers

• How to access, and the Trends / Comparison tabs.
• Filter bar — every filter with its values and default, and the fact that dropdowns only list values present in the selected tests.
• Saved filters — filters, date range and widget layout auto-save at org level; no Save button.
• Average and p90 — what each aggregation answers, a table of when to reach for which, and how p90 is actually derived: the mean of each session's own pre-computed 90th percentile, not the percentile of the whole window. Notes which metrics have no p90 and render as —, and that the toggle is Trends-only.
• Performance Overview, Performance Trends, Device Performance Matrix, Label Page Load Time, the eight per-metric trend widgets, and Compare mode.
• SLA thresholds — default bands, where they render, org-level and admin-owned.
• Threshold breach alerts — the email and Slack notifications.

Comparison tab

Documents the session-level overlay view, placed after Compare mode so the two "compare" concepts can be told apart — one overlays dimensions on a single widget, the other overlays whole test sessions.

• Complete flow — six steps from opening the tab to zooming the timeline.
• Test Sessions rail — every control: search, Test Name, date range, the More menu (OS, App Build Version, Device, Status), Clear Filters, the selection checkbox and Set as Baseline.
• How the graphs are drawn — one series per session with colours stable across cards, the shared elapsed-time axis and drag-to-zoom, legend toggles, per-metric variant toggles, SLA bands, and the Avg/Min/Max/P90 stats table.
• Baseline and Baseline Diff — how the deltas read, and that the direction counting as "better" follows the metric.
• Saved comparisons — rail filters, selected sessions, baseline, Baseline Diff and hidden series persist per test at org level. The date range deliberately does not: it belongs to the shared page picker, so both tabs always report over the same window.

The framing the section leads with: Trends charts calendar time, Comparison charts elapsed time within each run — every session starts at 0:00. That is what lets runs of different lengths and start times be read against one another.

Screenshots and assets

Three new Comparison captures from a live Android profiling test, plus the existing alert images converted to WebP. Every image the page references is now WebP and under 100 KB:

Image	Size
app-profiling-comparison.webp	76 KB
app-profiling-comparison-filters.webp	80 KB
app-profiling-comparison-startup.webp	64 KB
app-profiling-email-alert.webp	80 KB (was 264 KB PNG)
app-profiling-slack-alert.webp	60 KB (was 192 KB PNG)
app-profiling-screenshot.webp	84 KB (was 120 KB)

app-profiling-screenshot was 1825 px wide but rendered at 768, so it is downscaled to 1200 px. Its height attribute said 373 for a 1.29 aspect ratio — it never matched and would shift the layout as the image loads; corrected to 594.

Meta description rewritten to 159 characters.

Notes for review

• The heading anchor for the #comparison cross-link uses Docusaurus's auto-generated slug rather than an explicit {#comparison} id, because no other doc in this repo uses that syntax and onBrokenAnchors is not configured.
• assets/images/analytics/app-profiling-screenshot2.png is left in place but is no longer referenced by this page after the rewrite.
• The dashboard screenshot still shows "Network Upload (KB)" / "Network Download (KB)" in its legend. Those labels changed to MB in a recent analytics release, so that image is due a recapture — tracked separately, not blocking this page.
• No local Docusaurus build was run; relying on CI.

🤖 Generated with Claude Code