An event email is a small system
The first request sounds manageable: send students a list of upcoming events. Then the details arrive. Some events need a direct registration link. Others belong under an umbrella page. Full events should disappear, but an event without a capacity limit should remain. The time needs to say noon, or drop the extra zeroes, or follow a particular house style.
OADE, the Ongoing Adaptive Deliver Engine, gives those decisions a maintained structure. In this Gilded Build, David works through the setup in a Clean Slate test environment. The useful idea is to separate the event collection, the rules that shape it, and the layout that displays it. Each part can then be inspected when a preview does something unexpected.
The troubleshooting is useful too: empty results and awkward time ranges show where to inspect the query and its output.
In the recording · 00:55 ↗Give the event collection its own job
The recipient query begins with people. An independent Configurable Joins subquery supplies the event collection. That separation matters: the message needs events from a different dataset, then needs to decide which of those events belong in this recipient's invitation.
Dictionary exports carry the pieces needed for each row, such as the title, date, time and registration route. A for loop repeats the layout for the returned entries. This makes the collection itself a useful first inspection point. Before polishing the email, check whether the expected events are actually in the dictionary.
Write down one inclusion rule and one exclusion rule before building. A confirmed event within the selected date window might qualify. A full event or an excluded folder might not. The recording also accounts for events with no configured capacity limit. Those cases deserve distinct tests rather than one reassuring preview.
In the recording · 03:04 ↗Separate readable presentation from selection rules
Time formatting occupies a substantial part of the build because seemingly small display requests interact. Noon, a.m. and p.m., omitted :00 minutes and condensed ranges each introduce conditions. The recording builds these through layered exports that can be inspected individually.
The practical lesson is to collect the display requirements before adding another formula. Make a short list of examples with their expected output. Include a same-period range and a range that crosses noon. The final live preview does not demonstrate every possible time case; David explicitly notes that more test events would have helped.
Event visibility is a separate decision. The recording combines an international-event folder with recipient information while retaining general events. Geography and academic-interest matching are discussed as possible extensions. A new matching rule should have its own test record and a reason that someone maintaining the system can understand.
In the recording · 11:19 ↗Make the next click worth testing
A good-looking event row can still send someone to the wrong place. The build selects an umbrella registration route or a direct-event route and adds a person parameter. A fallback handles a missing umbrella value, but the underlying event configuration still needs care.
Check the full journey with approved test data: which event appears, what the button says, where it leads, and what happens when a value is absent. A correct-looking URL in a formula is only one part of that check. Keep personal information and recipient-linked URLs out of public screenshots and shared worksheets.
The demonstrated pattern assembles event content for a send. It does not rewrite an email that has already arrived in someone's inbox. The system also uses explicit rules; the word adaptive should not suggest that this demonstration contains a machine-learning model.
In the recording · 28:58 ↗Use the unfinished edges as a test plan
Near the preview, an empty event window explains missing rows. A missing separator explains a confusing time range. Those are useful moments because each suggests a diagnostic question: did the query return the expected data, and did the presentation combine it correctly?
The later seven-to-fourteen-day window example adds another boundary. David shows how the size of an event dictionary can choose a content block, but says those content blocks are not live. Treat this as an illustrated pattern requiring completion and testing, not evidence that a finished automatic switch was sent.
Start smaller than the full build. Pick one rule, create one fictional case that should pass and one that should not, then compare expected and actual behavior. Record what remains untested. That gives the next person a useful place to resume, and gives you a better reason to trust the next send.
In the recording · 55:21 ↗