Method
Session Quality Analytics, as we actually practise it
This page is the studio’s working definition. It is not a vendor comparison. If you only remember one sentence: a quality score is a claim, and claims need evidence, owners, and expiry dates.
Four layers we keep separate
Capture. What events exist, when they fire, and what they must never contain. Without a written event contract, replay and warehouse numbers will disagree and both teams will be right.
Reliability. Did the system allow the session to proceed? Crashes, timeouts, blocked assets, and permission failures live here. They are not “experience.”
Completion. Did the person finish, abandon with intent, or get stuck? Duration is almost never sufficient. We teach task models and explicit abandon signals.
Respect. Consent, PII leakage in replay, dark-pattern friction, and inaccessible flows. This layer is qualitative more often than not, and we refuse to average it into a crash rate.
How a Cloudpacketbase score is born
Students write a one-page charter before any formula. The charter names the population, the unit (usually a session, sometimes a task attempt), inclusions, exclusions, and what would falsify the score. Only then do we map events to components. Composite indices are allowed if each component can be plotted alone.
We prefer scores that expire. A quarterly review is part of the method: if capture changed, the number is retired or versioned. Silent drift is treated as a quality failure of the analytics programme itself.
Sampling without theatre
Replay is evidence only inside a sample design. The studio uses stratified draws by device, corridor, and time band, plus a small purposive set for known incidents. “The tool showed me this clip” is not a method. Cohort Replay Analysis is where this is practised in depth.
The Bangkok Field Lab
When a partner cannot export traces, we run a Field Lab day at the Bangphlat studio or on their floor. Laptops stay in the room. Students practise instrumentation notes on real constraints: vendor export gaps, Thai-language UI copy, payment-switch timeouts at month end. The lab is not a demo day. It produces a written score charter the squad can take home.
What this method will not do
It will not make a weak product look healthy. It will not replace accessibility work. It will not tell you which replay vendor to buy. It will give you language to argue, in English or Thai, about whether a session was actually any good.