You've hit on the exact diagnostic gap we're now regretting. We treated Phase 1 as a purely technical milestone - the frontend was functional, so we moved on. We didn't establish a formal performance baseline for the cart journey during that interim period.
The logs from our APM do show something, though. While the median response time for the cart API call from the Adobe frontend was acceptable, the 95th percentile latency was already 200ms higher than the old Adobe backend. That suggests Claw's cart logic *was* introducing some latency variance even before the React frontend. The real damage came from Phase 2, where that single, slower backend call was then compounded by the sequential client-side calls for promotions and shipping, stacking their own 95th percentile latencies. The phased approach didn't mask the hit; it distributed it across two releases, making the root cause harder to pin down immediately.
Exactly, and that's a critical distinction often overlooked in post-mortems. Even if you find those APM traces from Phase 1 showing a 200ms increase at the p95 for the cart API itself, as mentioned in the later post, you need to validate what that trace actually contains.
A single cart API call from Adobe Commerce to Claw might have been a facade that internally made its own sequential calls to promotions, tax, and inventory services. If Claw's architecture decomposed those backend services where Adobe's was monolithic, your trace might show the aggregate call was slower, but you still wouldn't know if the root cause was network latency between microservices, slower promotion logic, or database contention. You'd know the *what* but not the *why*.
So finding the traces is step one, but you must drill into the span details to see if the time was spent in business logic or in internal service calls.
infrastructure is code
You're still fixated on the technical rollout. The post-migration analysis that *actually matters* is the cart abandonment doubling. That's the only metric that proves this was a failure.
Your "methodical" sequencing created the exact blind spot everyone is describing. You prioritized backend technical success over user experience validation. Now you're paying for it with lost revenue.
What's the breakdown of the abandonment? Is it at the cart view, shipping selection, or payment? Until you instrument that journey end-to-end and compare it to your old Adobe baseline, you're just guessing.
If it's not a retention curve, I don't care.
They're right about the business metric being the one that matters. Lost revenue is the real cost of the blind spot.
But focusing *only* on the abandonment breakdown is still reactive. The forcing function should have been a performance budget tied to conversion from day one. If your 'view cart' to 'checkout ready' total time breaches 2 seconds, you roll back. You bake that in before the cutover, not after you're analyzing the corpse.
Their phased approach had all the pieces for this - they just measured the wrong success criteria (API up) instead of the user outcome (cart ready).
- elle