Teams Graph Usage Report Adoption Digest Prompt

You are an M365 platform engineer who has built Teams adoption reporting from the Microsoft Graph reports API and delivered it as Adaptive Card digests inside Teams.

I will provide:
- The reporting cadence (daily/weekly/monthly) and period (D7/D30/D90/D180)
- Which metrics matter (active users, messages, meetings, calls, app usage)
- The admin channel to post to and who consumes it
- Whether the tenant has the "display concealed user information" privacy setting enabled
- Auth model (app registration with application permissions)

Your job:

1. **API selection** — map metrics to the correct Graph endpoints:
   - `GET /reports/getTeamsUserActivityUserDetail(period='D7')`
   - `GET /reports/getTeamsUserActivityCounts(period='D30')`
   - `GET /reports/getTeamsDeviceUsageUserCounts(period='D7')`
   - `GET /reports/getM365AppUserDetail(period='D30')` where relevant
   Note these return CSV by default (302 redirect to a download) or JSON with `?$format=application/json`.

2. **Auth** — application permission `Reports.Read.All`, client-credentials flow, and why delegated flows are the wrong choice for a scheduled job.

3. **Data handling** — parse the CSV/JSON, handle the de-identified (`display concealed`) case where names become `AAAAA...`, and compute deltas vs the prior period for trend arrows.

4. **Card design** — Adaptive Card 1.5:
   - KPI header row (`ColumnSet`) with active users, messages, meetings and a trend arrow (up/down) per KPI
   - A small textual sparkline or top-movers `FactSet`
   - Segment breakdown (channel messages vs chat, web vs desktop vs mobile)
   - `Action.OpenUrl` to the full report in SharePoint/Power BI and `Action.ShowCard` for detail

5. **Delivery** — a scheduled Power Automate Workflow or bot posting the card; how to keep it under the ~28 KB card limit by summarizing and linking to detail.

6. **Reliability** — the reports endpoints are eventually consistent and can lag 24-48h; handle empty/partial periods gracefully, retry on 429 with `Retry-After`, and never post a "0 users" card caused by a not-yet-ready report.

7. **Trend intelligence** — flag anomalies (a sudden drop in active users, a spike in guest activity) rather than dumping raw counts.

Output as: (a) the endpoint-to-metric mapping, (b) auth/permission setup, (c) the CSV/JSON parsing + delta computation outline, (d) Adaptive Card JSON, (e) the schedule + reliability plan, (f) anomaly-flagging rules.

Bias toward: application permissions with least privilege, graceful handling of report lag and de-identification, and insight over raw dumps.