pr-analytics-review

PR Analytics Review

Constants

Constant	Value
Amplitude project ID	241072
Analytics import	import zenalytics from 'common/lib/zenalytics'
Taxonomy properties	kingdom, phylum, class — always include all three
Fabric Surfaces values	GROWTH_KINGDOM, EXPANSION_PHYLUM from app/constants/AnalyticsConstants

Common mistakes:

Anti-pattern	What to do instead
Assuming an event name is what it sounds like	Always call get_events first; read the description field
Reporting event volume without verifying source	Grep the actual tracking call; verify source constant value
Using an event name from get_events without codebase grep	Grep to confirm what code fires it before using it in any analysis
Treating File Uploaded as a CSV import event	That event is hardcoded to file attachments on person/company ID pages

---

Workflow

Copy and track progress:

PR Analytics Review:
- [ ] Step 1: PR read — feature + success moment identified
- [ ] Step 2: Existing tracking audited (zenalytics grep)
- [ ] Step 3: Related Amplitude events found + codebase-verified
- [ ] Step 4: Enterpret qualitative signal checked
- [ ] Step 5: Recommendations written (events + Amplitude + Enterpret)
- [ ] Step 6: Success definition + schedule check-in offered

---

Step 1 — Read the PR

gh pr view <number> --json title,body,files
gh pr diff <number>

Identify: the user action being enabled, the success moment, and which files changed.

---

Step 2 — Audit Existing Tracking

grep -r "zenalytics\|\.track(" <changed-files> --include="*.tsx" --include="*.ts"

Document: what fires today, what fires on the success moment, what's untracked.

---

Step 3 — Find Related Amplitude Events

mcp__claude_ai_Amplitude__get_events({ projectId: 241072 })

Read the description field for each candidate.

⛔ Do not skip: grep the codebase to confirm what code fires each candidate event before using it. Event names are often misleading — see common mistakes above.

---

Step 4 — Enterpret Qualitative Signal

mcp__claude_ai_Enterpret_Wisdom__search_knowledge_graph({ query: "<feature area>" })

Note frequency scores, sentiment, verbatim quotes. If users complain about something with no Amplitude event, that gap is highest priority.

---

Step 5 — Recommendations

Output three sections:

A. Events to add — for each gap: event name (follow [Area] Action Noun convention), exact file + handler to fire from, full properties including taxonomy, ready-to-paste snippet:

zenalytics.track('Object Sharing People Added', {
  kingdom: GROWTH_KINGDOM,
  phylum: EXPANSION_PHYLUM,
  class: 'Object Sharing Modal',
  mode: 'multi' | 'single',
  selection_count: number,
  select_all_used: boolean,
});

Priority-order by impact — success moment event first.

B. Amplitude monitoring — specify the exact funnel/charts to create: entry event → success event, segmented by plan tier. State what "good" and "bad" look like numerically.

C. Enterpret monitoring — 2–3 queries to run 2–4 weeks post-ship. State what sentiment shift to watch for.

---

Step 6 — Define Success

End with:

2-week check-in target: <metric> should <direction> by <amount> vs pre-ship baseline. If it doesn't move, suspect <hypothesis>.

Offer to schedule a follow-up check-in in 2 weeks to query Amplitude and report whether the metric moved.