Method

Install, walk, mark, then talk

The instrumentation audit is a sequence we refuse to reverse. Evidence is gathered on a device with logging on. The readout exists to decide what the next build will change. Mixing those two on the same call is how findings get argued into softness.

Rows of organised notes and task cards on a wall

01

The paper journey

Before we touch the build, we print the journeys named in the brief. Each line is a human action: enter an address, fail a voucher, pay. Beside it we leave room for the event we saw, the time, and whether a required parameter arrived. The paper is the script. The analytics property is not.

02

A device that has not met the app

We install on a handset that is not a daily driver, so first_open and consent timing are honest. iOS and Android are walked separately when both stores are in scope. A finding that exists on one store is still a finding.

03

Logging, then a proxy when logging lies

We turn on the SDK’s debug view when it shows real payloads. When it does not, we use a proxy and read what left the device. Automatic screen reporting, delayed batches, and consent gates are noted as we go, not reconstructed later from a weekly chart.

04

Mark in oxide, type the same day

Silences, double fires, empty parameters, and events that arrive before the consent dialog is dismissed get a mark on the paper. We type the paper into the report the same day, while the proxy history is still on the machine. Charts stay closed until that typing is done.

05

The readout is for decisions

You receive the written findings first. The ninety minutes are for choosing what engineering will change in the next version, and what can wait. We will not discover a missing purchase_success live on the call if we already walked checkout.

Two people looking at a laptop screen together during a working session

What we ask you not to do

Do not grant production admin. Do not add a temporary SDK “so we can see more.” Do not join the walk on speakerphone and narrate what should fire; we need to see what does fire when a stranger taps.

If the candidate cannot be installed, the review waits. If debug logs cannot be produced, we will say so rather than invent a finding from the dashboard you already have.

This method belongs to the instrumentation audit. A pre-release check uses the same walk on a shorter event list. A measurement plan session does not inspect current fires at all.

Request this walk on your build