Morning: the brief and the read date
The illustrative day begins when the store's founder opens the Today tab. The morning brief lists what is waiting: the cart experiment on shipping-threshold messaging has reached its planned sample size overnight and its result is ready to read, and a new product-page experiment is proposed. Below that, yesterday in one line: tests running, their planned read dates, and no changes made to the live store. Spend against the month's budget sits in the corner.
The founder notices that the brief has never shown an interim result for the cart test in the weeks it ran, only its read date. That was the stop rule doing its job.
Mid-morning: a loss, reported as a loss
In the Experiments tab the concluded test shows its result: the variant with a progress bar toward free shipping did not move revenue per visitor by a detectable amount at the planned sample size. The confidence interval straddles zero. The Conversion FTE reports the test as a loss, shows the decomposition, a slight rise in order value offset by a slight fall in rate, and revises the hypothesis: the shipping cost may be the objection, but the progress bar was not the answer; showing the shipping cost itself in the cart is the next candidate.
Nothing on the screen tries to salvage a win from the numbers. The founder marks the result read, and the revised hypothesis moves into the ranked queue with its evidence.
Midday: a new experiment in Approvals
The proposed product-page experiment is next. It shows the hypothesis, that shoppers on mobile leave the product page without a delivery answer, with the evidence: exit data by device and the help desk's most common pre-purchase question. It shows the change, delivery time and the returns summary moved above the fold on mobile, previewed on a phone and a desktop. It shows the metric, the sample size at the store's mobile traffic, the expected duration, the stop rule and the rollback.
The founder approves with one edit: the returns summary must use the wording from the policy page verbatim. The rule is recorded in the remit, and the test launches on the store's testing tool.
Afternoon: a broken variant, caught
Later the brief updates with a pause. The Conversion FTE has detected that the new variant throws an error on one older mobile browser, visible in the store's error events, and has paused the test rather than let contaminated data accrue. The pause is reported with the evidence, the affected sessions, and a proposal: fix the variant and restart the sample from zero, because a test interrupted by a broken variant cannot be resumed honestly.
The founder approves the fix and the restart. The read date moves accordingly, and the brief will show the new one tomorrow.
Late afternoon: a diagnostic question
The founder asks a question in Missions: why did checkout completion fall after last week's theme update? The Conversion FTE returns ranked hypotheses with evidence rather than an answer: the checkout pages are measurably slower on mobile since the update; a payment wallet button no longer renders on one device class; paid traffic mix shifted toward a lower-intent campaign the same week. Each hypothesis carries what the FTE can see and what it cannot, and the speed one is ranked first because the evidence is direct.
The founder picks the speed hypothesis to act on, and because checkout is on the always-gated list, the proposal that follows will be a fix for a person to approve, not a test the FTE launches.
Evening: the log
At the end of the illustrative day the Performance tab shows what happened: the cart test read and recorded as a loss with its numbers, the revised hypothesis queued, the product-page test approved with its wording rule, the pause and restart, and the mission opened, each with the person who decided and the evidence attached. Revenue per visitor for the month so far sits against the baseline, labelled as a partial month.
Nothing in the day asked the founder to trust a system. It asked them to read one honest result, approve one previewed change with one edit, and approve one restart. The record exists already, which is what an audit trail means when it is built in.