Decision Center
Every recommendation arrives with what we believe, why, and the smallest next step. A decision is a recorded prediction — that's the only reason we get to learn.
Make SSO setup optional during onboarding. Move it to a post-invite step gated only when an admin enables enterprise policies.
The SSO gate sits directly on the step where activation is bleeding, and three independent signals (funnel, churn interview, qualitative review) point to the same cause. Decoupling SSO from invites is the smallest change that addresses the dominant hypothesis.
- "Configure Workspace step 62% → 49%."
- "By the time IT got involved we'd already moved to Linear."
- "First tool that actually helps me come to a decision."
- Deferring SSO does not materially raise security incidents in week 1.
- The activation lift is causally tied to the gate, not seasonality.
- Workspaces enter production without governance — mitigate with a 14-day SSO prompt.
- If seasonality is the real driver, the metric won't move and we've spent a sprint.
Three independent evidence types converge. The proposed change is reversible behind a flag and instrumented end-to-end.
Scope SCIM down to just-in-time provisioning for Q3 and ship audit log export from the platform team in parallel.
JIT covers the procurement bar both design partners actually evaluated. Full SCIM lifecycle adds weeks for no immediate revenue. Audit log export unblocks three renewals from a team that isn't on the activation path.
- "Two design partners need SAML + SCIM by end of quarter."
- "Two are in renewal cycle Q3."
- Procurement at both partners accepts JIT for v1.
- Platform team has the bandwidth to ship audit log export in 3 weeks.
- If procurement rejects JIT, we'll need full SCIM and will be late.
Strong commercial signal, but procurement scope is unverified at one of the two partners.
Roll back the tab-state reducer change from 2.41 behind a flag; reintroduce with a regression test.
The regression is narrow, localized, and high-trust-impact. The fastest defensible move is to revert and reland with a test, not to forward-fix under load.
- "Started after release 2.41."
- The reducer change is the root cause.
- Revert reintroduces the original bug that motivated 2.41.
Causal link is plausible but unconfirmed. Bisect would raise confidence in under an hour.