feat: event-app-space-report skill (per-event completion report)

.claude/skills/event-app-space-report/SKILL.md

@@ -0,0 +1,92 @@
+---
+name: event-app-space-report
+description: >-
+  Use when asked for completion statistics / a report for an SAP event (TechEd,
+  ASUG Tech Connect, DSAG, etc.) on the tutorials-ims / developers.sap.com
+  platform, given an app-space URL (…/app-space/?eventId=NNN) or a bare eventId.
+  Produces the four-section report: tutorial completions by Track, abandonment +
+  step-drop histogram, completer email-domain breakdown, and SAP-vs-external.
+  Triggers on "report/statistics from the event <url|eventId>", "app-space stats
+  for …", "event completion numbers".
+---
+
+# Event App-Space Report
+
+Produces a point-in-time, read-only completion report for one event, scoped by
+the event's **date window** to the tutorials linked via its mission's
+CompletionPaths (Tracks). Reuses the server helper `srv/lib/event-path-buckets.js`
+so Track numbers match the live app-space page.
+
+## Input
+
+An app-space URL (`https://developers.sap.com/app-space/?eventId=10000003&theme=…`)
+or a bare `eventId` (= `Events.legacyId`). The script parses either.
+
+## Steps
+
+1. **Parse the eventId** from the URL (or accept the bare id).
+2. **Target PROD.** The report runs against PROD HANA. Confirm you are logged in
+   and targeted at the **prod** space first:
+   ```bash
+   cf target        # org/space must be prod
+   cf target -s prod
+   ```
+   The script enforces this (`assertCfTargetProd`) and refuses to run on any
+   other space — it is the only env-separation safeguard (binding follows the
+   current `cf` login; there is no prod-vs-dev named binding).
+3. **Bind PROD HANA** if not already bound in this worktree:
+   ```bash
+   cds bind -2 tutorials-hana
+   ```
+   **Do NOT use the `tutorials-hana:tutorials-hana-key` service-key** — per
+   `docs/developers/operations/gameboard-prod-data-troubleshooting.md` it maps to
+   a near-empty decoy HDI container, not the live app container. Use the plain
+   instance binding.
+4. **Run the report** (read-only; SELECT/CQL only, no writes):
+   ```bash
+   cds bind --exec -- node scripts/event-report.mjs <eventId> [--json]
+   ```
+   `--json` additionally prints the raw event + bucket aggregates.
+5. **Present** the emitted markdown to the user (the four sections).
+
+## Verification
+
+Cross-check the **Track** totals against the live app-space page or the admin
+`getEventBuckets` OData call for that event when the user wants assurance. The
+script reuses the same `computePathBuckets` helper, so a mismatch means a
+binding/environment problem (wrong space, decoy container), not a logic
+difference.
+
+## Report sections
+
+1. **Tutorial completions by track** — total + per-Track count and share
+   (`CompletionPaths.name`; FK-first tutorial resolution; windowed COMPLETED/
+   SUPERSEDED completions).
+2. **Abandoned tutorials** — a "Top abandonees by count" list, a computed
+   stop-pattern summary (minimum steps done before stopping; % of abandoners
+   who'd done 3+ steps), then per-tutorial median stop-step with "% through"
+   framing and a per-step drop histogram (stop-step = COMPLETED STEP records at
+   the user's current attempt vs `Tutorials.stepCount`).
+3. **Completer email domains** — distinct completers per domain (with known
+   multi-domain orgs like NTT Data / EY rolled up via `ORG_DOMAIN_MAP`),
+   completions, completions/user; a most-engaged table + a prose callout naming
+   the standout high-engagement companies.
+4. **SAP employees vs external** — `@sap.com` vs everyone else.
+
+## Notes
+
+- Read-only and prod-guarded. No writes, no deploy.
+- `scripts/event-report.mjs` lives in the main repo, so it's runnable outside
+  this skill too (`cds bind --exec -- node scripts/event-report.mjs <eventId>`).
+- Scoping is by **date window**, never `TaskRecords.event_ID` (NULL on normal
+  completions). Track = `CompletionPaths.name`, not Mission title.
+- **Sections 3–4 reconcile with the §1 total**: domain and SAP-vs-external
+  counts use the same all-windowed TUTORIAL completion set as the "Total
+  completions" headline, not only track-member tutorials. Anonymized/null-email
+  completers show as `(unknown)` in the domain table and are excluded from the
+  SAP-vs-external split.
+- **Abandonment is an approximation, not an exact count.** Stop-step counts a
+  user's COMPLETED STEP records at their current `attemptNumber` with no
+  per-step date filter, so for returning/multi-attempt users (steps completed
+  before the event, or under a different attempt) the step-drop histogram and
+  median are estimates. Treat them as directional.