Kayla Young (Gilbert)
← Back to selected projects

Professional QA investigation

Making a platform-wide QA investigation usable

A platform-wide investigation needs a useful ending. I directed an AI-assisted staging pass, reviewed the evidence, and turned a large set of findings into patterns, data risks, and decisions for product and engineering.

Result: 250 documented findings grouped into 23 recurring patterns, with data risks and five unresolved decisions routed to owners.

Scope
Four-day staging pass; one QA account, Chromium, seeded data
My role
Test strategy, investigation direction, evidence review, reporting

Problem

A new platform had reached staging ahead of launch decisions. The team needed to know which observations threatened data, which appeared connected, and which depended on a product decision about intended behavior. A long list of issues would leave that sorting work to the reader.

My job was to make the investigation usable: show the relationships, prioritize the risks, and be clear about where coverage stopped.

Decisions

I organized the pass across application sections, administration settings, and cross-page behavior. I directed AI-assisted browser harnesses and checked their work. A Save click was the beginning of the check: refresh, stale forms, repeated actions, and changes elsewhere could tell a different story.

I grouped the 250 documented findings into 23 patterns and prioritized data risks for the handoff. The readout separated observed behavior from questions about intended behavior. Five unresolved decisions went to product and engineering owners.

Illustrative reconstruction · No product data

FindingA saved setting appears to revert after a refresh.
Evidence neededRecord the save, refresh, and readback separately. A retest needs the starting values and an unaffected control.
RiskA user may rely on a state that did not persist.
Open decisionWho defines the intended persistence behavior?
Untested scopeOther accounts and devices are outside this pass.
This fictional readout demonstrates the reporting structure. It is not a product screenshot or an actual finding.

Validation

I repeated checks and used a second review to try to disprove the higher-impact findings. Configuration checks led to withdrawals, and one result became flaky when another run did not establish it. A claim about stored data needed an actual save and readback. Correcting a finding was part of the investigation.

The footage audit separated full matches, partial support, and steps the recording never showed. A video of a Save click proves the click. A persistence claim still needs the reload and readback.

Evidence

Findings
250 documented findings in 23 recurring patterns
Handoff
Five decisions routed to product and engineering owners

The four-day pass produced 250 documented findings organized into 23 patterns. The readout highlighted 22 data risks covering 53 flagged issues and routed five unresolved decisions. It gave readers a prioritized summary with detailed evidence behind it.

Evidence audit · 314 rows

239 matched55 partial20 not shown on video
These 314 audit rows are evidence rows, not additional bugs. The split does not establish full video verification for every finding.

I separated observed behavior from five unresolved product decisions and routed those questions to product and engineering owners. The VP of Engineering shared the readout with the engineering team, pointing readers to the summary before ticket filing. The result documented here is the investigation and handoff; completed fixes remain outside this case study.

Limits

Coverage used one QA account, Chromium, and seeded staging data. Two windows on one account did not establish cross-user behavior. The work did not establish cross-device behavior or full platform coverage.

The next useful move would be to turn those recurring patterns into repeatable probes, include the deferred coverage, and identify decision owners earlier. The findings give that pass a starting point, with the unfinished work still visible.