Glean is $500k/yr.
Mulesoft is $80k/yr.
That's the integration tax.

Two of our top-10 accounts (Northwind, GlobalCo) just told me on the QBR that they 'didn't know v2 was being deprecated.' I've sent the v2-to-v3 guide and booked migration sessions, but flagging because if it's two of ten, there are likely more. We may need a louder push than the monthly email.

Spinning my wheels on the Clerk webhook signature verification — the docs say sha256(timestamp + '.' + body) but every payload from staging fails verify with that recipe. Filed a support ticket. ENG-503 stays blocked until I hear back. If it drags past Friday I'll switch to polling and we lose real-time, which OPS-88 needs.

Update from outside counsel: for the EU launch we have to keep customer event data inside the EU under GDPR. US-only deployment is not an option for the design partners we're onboarding. We need a residency story before we can sign the German DPA.

ClickHouse spike done. Same rollup query: 41 sec on Postgres -> 380 ms on ClickHouse with the same hardware. Disk usage is also ~9x smaller after compression. Writing it up in ADR-007.

Two more backend engineers won't help if production keeps paging Marcus at 2am. The bottleneck is operational — we have no on-call rotation, no runbooks, no observability beyond what Marcus carries in his head. The first hire has to be an SRE.

We're tracking against three deadlines and one staff eng (Marcus) is critical-path on all of them: ClickHouse migration, auth migration, EU launch. I want to open two senior backend reqs this week so we can parallelize.

Heads up: the daily usage rollup query just timed out for the third day running. We're at 14B events in the events table and Postgres OLAP is hitting a wall. I'm going to spike ClickHouse this week and post numbers.

Reminder for the team: v2 API sunset is locked for July 15, 2026. CSMs own the customer side. Engineering side: please don't accept any new v2 PRs after May 15 unless they're security fixes. Issue #987 is the source of truth.

Per the EU residency RFC: provision a fully isolated stack in eu-central-1 — VPC, RDS Postgres, ClickHouse, Clerk EU tenant, ingestion edge. No cross-region event replication. Blocked on: ENG-503 (Clerk EU tenant requires the new auth migration to be live). Acceptance: a customer signed up via app.eu.acme.com only sees their data resident in Frankfurt; US ops dashboard cannot query event data.

Postgres handles all writes today, including the analytics rollups read by the dashboard. Rollup queries now exceed 30s and time out under peak load. Plan: dual-write events to ClickHouse, point dashboard reads at ClickHouse, keep Postgres as the source of truth for OLTP. Rollback plan: cut dashboard back to Postgres reads. Acceptance: p95 rollup query < 1s, no dashboard regressions.

Headcount approval pending Sarah's sign-off. Drafted JD lives at notion://hiring/sre-jd. Mike argues SRE is the bottleneck before more product work; Sarah argues two more backend engineers unblock more deadlines. No decision yet — track on the Apr 22 leads sync.

Move authentication off Auth0 onto Clerk. Drivers: Auth0 contract renewal at +40% in May, Clerk's per-org pricing fits our enterprise tier better, native Clerk EU tenant unblocks the Frankfurt cluster (OPS-88). Migration plan: dual-issue tokens for two weeks, then switch the JWT verifier. Blocked on: webhook signature verification — Clerk's HMAC docs don't match the SDK behavior we're seeing in staging. Owner replied to Clerk support Apr 13, no resolution yet.

Adds @clickhouse/client and a thin AnalyticsWriter wrapper that writes every event to both Postgres and ClickHouse behind a feature flag (analytics.dual_write). Dashboard still reads from Postgres in this PR; the cut-over happens in #1301 once we've verified row counts match. Linked: ENG-471. Reviewed by tom.

Adds Clerk JWT verifier alongside the existing Auth0 verifier; both are queried during a two-week dual-issue window before Auth0 is removed. Marked draft because the webhook handler in src/auth/clerk-webhook.ts can't verify Clerk-signed payloads — see the failing test in __tests__/clerk-webhook.test.ts. Holding the PR until the upstream support ticket lands. Linked: ENG-503.

v2 of the public API will be removed on July 15, 2026 per the deprecation policy. v3 has been GA since Jul 2025 and is feature-complete. Migration guide: docs.acme.com/api/v2-to-v3. CSMs are reaching out to all v2 customers with a one-on-one migration session. Final security patch: Jun 1. After Jul 15, v2 endpoints return 410 Gone.

Constraint: under GDPR plus the agreed German DPA, customer event data and any derived analytics must be processed and stored inside the EU. Recommendation: stand up a fully isolated Frankfurt cluster (eu-central-1) — Postgres OLTP, ClickHouse analytics, Clerk EU tenant. No cross-region replication of event data. Identity (org/user metadata) replicates to the US for support. Owner: tom for infra, priya for the legal sign-off, marcus for the schema split. Target: Frankfurt cluster live by Jun 1.

Decision: add ClickHouse alongside Postgres. Postgres remains the system of record for OLTP. ClickHouse serves the analytics dashboard and the daily rollup jobs. Rationale: a single Postgres has worked for three years, but at 14B events the analytics workload now dominates and forces vacuum/lock contention on OLTP. Spike showed 100x latency improvement on the canonical rollup query at equivalent hardware cost. Rejected alternatives: scale Postgres up (linear cost, doesn't fix lock contention); managed Snowflake (5x the monthly cost in our usage band); Apache Pinot (less ops-mature, smaller community). Owner: marcus. Target: dual-write live by Apr 30, dashboard cut over by May 15.

All major API versions are supported for at least 12 months after a successor ships. v2 was released Jul 2024. v3 shipped Jul 2025. v2 sunset is therefore Jul 15, 2026, with one final security-only patch on Jun 1. Customers on v2 receive a monthly migration nudge and a named CSM-led migration session. Internal SLO during sunset: zero p1 outages, < 0.5% v2 traffic by Jul 1.

Priya: 'I got the redline back from German counsel. The DPA is signable as long as we can show event data never leaves the EU. The two open issues are (1) our analytics pipeline — confirmed resident now that ClickHouse is also in Frankfurt — and (2) support tooling, where US engineers occasionally pull customer data to debug. We need a tenant-aware support proxy.' Tom: 'I'll own the support proxy ticket. Sarah, can we get the EU launch slipped to Jun 15 to give us a real margin?' Sarah: 'I'd rather hold Jun 1 and cut Auth migration scope than slip the launch — we have signed LOIs.'

Tom: 'I want to lock the ClickHouse decision today. Marcus, summarize?' Marcus: 'Spike numbers are clean — 100x on the rollup, 9x smaller on disk. The risk is operational, not technical. We're adding a system to maintain.' Sarah: 'Are we sure we want this BEFORE the EU launch?' Tom: 'The dashboard is already broken under load. We can't ship EU on a query that times out. Let's accept ADR-007 and start dual-writing this week.' Priya: 'For EU we'll need the analytics data resident in Frankfurt too — flag that in the ADR.'