Chinh (lelouvincx) / Weekly Report - 2026-W17

Weekly Report — Apr 20 – Apr 26, 2026 (friday mode)

Work

Internal

• Following the merge of dbt #853 (approved last week), captured the full reporting-layer plan in journals/2026_03_19.md covering Source / Staging / Domain / Mart / Deprecated / Fixed layers.
• New canonical mart: mart_growth__calendly_bookings (view → fct_calendly_bookings, grain = event × attendee × rep) and mart_growth__calendly_events (table → dim_calendly_events).
• Deprecated (enabled: false): mart_growth__demo_calls, mart_growth__onboarding_calls, mart_core_business__demo_call_bookings, dom_calendly__demo_call_bookings, dom_calendly__onboarding_call_bookings.
• Fixed: mart_core_business__sales_leads now joins dom_calendly__bookings directly on attendee_email_domain (resolves the 88% drop in call-booking leads found last week).
• Next: update Holistics BI Model / Dataset / Dashboard to align with the new layer (Ampcode thread).

• Round 1 review done. Used Ampcode (code review) and Ampcode (PR 858) to scaffold review.
• Now waiting on anh Dong’s response. Notion DE-214.

• Failing test: unique_map_identities_zendesk_hubspot_zendesk_organization_id. Zendesk org 900032617243 (1win.pro) maps to 2 HubSpot companies because 1win.pro appears in all_domain_names of both HubSpot 5855457913 and 18891852445.
• Root cause: IN UNNEST(all_domain_names) join in itg_mappings__zendesk_tenant.sql (and similar integration-layer joins) creates fan-out whenever a domain appears in multiple HubSpot companies. Originated from DE-187 phase 1 (PR #812).
• Impact: 16 tenants across ~10 companies; affected customers: Correlation Risk Partners, Cruisebound, 1win.
• Decision: leave the test failing for now and address as part of the broader DE-187 resolution (canonical customer_skey work). Not yet actioned due to sick leave.

• Phase 1: deprecate in code. Phase 2: remove the .sql file. Phase 3: drop the table.
• Don’t sit between phase 2 and 3 — AML models reference BQ physical names; removing the .sql only stops refresh, leaving the table with stale data and silently misleading consumers. Either keep the wrapper alive or drop the table entirely.

Presales

• anh Huy posted the consolidated workaround to Taha (Basata) on Apr 21:
  • Gap 1 (no theme context in embed payload): Basata duplicates the dashboard into 2 versions (light + dark); Basata app reads the user’s theme preference and renders the matching embed.
  • Gap 2 (Sankey + dynamic markdown don’t adapt): standard charts via per-theme color palettes; Sankey via custom CSS; dynamic markdown gets per-theme CSS in each dashboard. anh Huy offered to impersonate and apply the Sankey fix.
• Caveat communicated: dashboard header + zoom section can’t be customized via CSS yet (rendered at the Holistics layer); use case forwarded to PMs for the longer-term embed-payload theme support.

• Showbie (Slack thread): Harsha asked about parent-child filtering, Param Fields, and pivot-by-question for a Looker → Holistics rebuild. I confirmed all three are supported, sent doc links, and offered to log in and build the pivot example for him. No reply.
• Basata (Slack thread): anh Huy proposed the dual-dashboard workaround and offered to apply the Sankey CSS fix via impersonation. No reply yet.
• I don’t know how to best communicate with these kinds of users — possibly the offer of “I’ll do it for you” reads as too forward, or they need time to absorb. Worth raising in next 1-on-1.

Duty Support

• Confirmed Hieu’s dashboard is feasible to share as spreadsheet but not yet ready to consume — needs polishing before delivery.
• Aligned with the commercial decision: Vincent (CEO) clarified the in-app AI usage dashboard is free for all customers; what’s being sold is data sync/export to Aurora’s own systems (S3 preferred, Google Sheets secondary). Targeting enterprise compliance, security auditing, 7+ year retention.
• Final delivery format pending Nam’s email to Aurora. AI team (Phuong/Tien/Dat) owns separate in-app monitoring; data team scope is bounded to current ClickHouse data.
• Prior art: SweetSpot Slack thread — same pattern.
• UPDATE — deal deferred Email thread between Nam (Holistics) and Weiping (Aurora), Apr 20–23 (Slack file):
  • Nam pitched the $10,200/year “Usage and AI Data Monitoring Transfer Sync” (S3 delivery, scope: conversation time/ID, user ID, prompts + AI replies, token usage per user).
  • Aurora pushed back: hard to justify $10,200/year given the in-app AI usage analytics is on the Holistics roadmap for the next 1–2 quarters. Counter-proposals: monthly manual data dumps OR a read-only user on a read-only replica of the backend DB.
  • Both sides agreed to revisit when Holistics has more concrete AI usage data.
  • Implication for data team: no immediate delivery work needed. The Aurora-specific AI conversation dashboard (Hieu’s) can stay on hold; focus shifts to AI team’s in-app analytics rollout. Keep the dashboard ready in case the conversation reopens.

Docs

• NOTE Follow-up DM with Khai To on Apr 23–24 (Slack DM):
  • Demo direction: lead with the outcome demo (like lightdash does — show power first); push step-by-step setup into the docs.
  • Khai’s MCP for local metric queries is on staging13, not yet prod. He’ll demo it (skill + MCP combo) once I’m back from holiday.
  • Estimated cadence: “probably 2 weeks after holiday” — aligns with the demo-video scoping push to mid-May.

Personal / Tooling

• Triet announced the Vanta → Drata migration to all engineering on Apr 23. Required steps: acknowledge all policies, install + register Drata agent, complete Security Awareness Training and HIPAA Training. Keep Vanta installed until further notice. Two-week deadline.

• HR (Nhu Nhu) asked everyone to install Etax Mobile and check 2025 PIT finalization status before her leave. Verified own status: single-source income at Holistics → finalization successful, no follow-up needed.

Learning & Notes

• Instead of “Hey boss, can we install action X?”, say “Hey boss, I’m going to install action X to solve XYZ. Will take care of it Monday unless I hear differently.”
• Shifts cognitive load: gives the manager an opt-out window without blocking your progress, while you still own the framing of the problem and the solution.
• Tone matters — needs to be calibrated to the person and what they actually want to know (e.g., size of the problem, downstream risks). Worth practicing on smaller decisions first.
• The tips category on mooreds.com reads like a senior engineer sharing small craft-level tips that LLM searches don’t surface — reminder to keep cultivating non-LLM reading sources.

Next Week

• Internal: Start MRR exchange rate work (DAT-576) — build dim_exchange_rates with live rate ingestion. Carry-over, manager P1.
• Internal: Pick up DAT-555 follow-up captured in journals/2026_04_27.md.
• Internal: Update Holistics BI Model / Dataset / Dashboard for the new Calendly mart layer (DAT-283).
• Internal: Resolve DAT-556 dbt test failure (root cause already analysed Apr 22).
• Internal: Close out CircleCI PR review feedback loop with anh Dong (prefect #397 / DAT-579).

• Internal: Lead Funnel by Sales Motion (DAT-560) — unblock once Calendly BI layer is rolled out. Carry-over since W14 (4 weeks).
• Internal: Clean up BI layer for job queue dashboards (delete WIP 3.0, migrate v2 / Report Monitoring to 4.0).
• Internal: Increase data freshness — impersonation lookup more frequent than daily (DAT-499). Manager P3.
• Docs: Draft scenario for local development demo video, then sync with anh Huy Vu / Dat. Carry-over.
• Teaching: Find a football dataset for Duc Anh.

• Internal: Contribute to Holistics skills and internal-skills repos.
• Presales: Read Modeling Patterns docs.
• Personal: Self-reflection + CV update. Research alternatives to 1Long for emergency savings.

Career & Personal Consulting

• Keep: When energy was low, you still chose the highest-leverage tasks (Basata follow-up, Aurora dashboard review, CircleCI review round 1) and let the rest slip rather than dragging everything to half-done. That’s healthy triage.
• Keep: Capturing the dbt deprecation policy as a reusable note (3-phase rule). These small principles compound — they prevent future “stale table” landmines and give you something concrete to teach Thuan / juniors.
• Start: Treat illness as a hard stop, not a soft pause. You felt off from Monday and pushed until Wednesday before resting. Sleep + early stop on Mon would likely have salvaged Tue/Wed. Build a rule: if you feel off two days in a row, stop on day two.
• Stop: Late-night gaming on weeknights — you already self-identified this as the root cause. The LEARNING note is good; pair it with a concrete trigger (e.g., laptop closes at 22:00 on weeknights).

• This was a low-output work week by design (illness + trip), and that’s fine. The bigger risk is that two of last week’s manager-priority items (MRR exchange rate, BI cleanup) are now zero-progress carry-overs. Next week is also short (2 holidays); be realistic about scope.
• The “Ask for no, don’t ask for yes” article is timely — your carry-over list is growing, and proactive framing (“I’ll start MRR Mon, will course-correct based on Friday review”) would reduce review-cycle latency.

• Trip to Bàu Sấu was a real reset — physical movement (6km/day) + no signal + nature is the right antidote to context-switching fatigue you flagged in W16.
• Health: the cold is a signal. Sleep > productivity. Seven days of recovery cost roughly the same time as four nights of better sleep would have prevented.

• Carry-over is becoming a pattern (MRR, Demo video, DAT-560, DAT-524). Consider a hard rule next week: any item carried over 3+ weeks gets a forcing function — either start it on a specific day, or formally backlog it. Avoid the “always-on but never-done” zone.

• Both threads end with us volunteering hands-on work (“I can build the example”, “we’ll apply the CSS via impersonation”) and the customer going quiet. Two non-mutually-exclusive hypotheses worth testing:
  • (a) Decision delay — they need time to socialize internally; silence is “not yet”, not “no”. Action: send a soft check-in 3–5 days later instead of waiting indefinitely.
  • (b) Perceived push — “I’ll do it for you” can feel like over-reach for users who want autonomy. Action: try framing as a self-serve doc + offer (“here’s the recipe; happy to pair if helpful”) rather than “I’ll do it”.
• Either way, a follow-up cadence rule (e.g., 3-day silent → ping; 7-day silent → close loop with PM) prevents these from becoming permanent open loops.

Recommended resources to learn