Prompt that sets watermarks and straggler policy for event windows. Paste it into an AI tool of your choice, which then steers late events with watermarks.
Results depend on the tool, the model and what you enter.
Result preview
Example: written by our editors, not real AI output.
Window: revenue per calendar day
Event vs arrival: up to +6 h
Today: close 00:05 server time
Watermark: event time − 2 h
Stragglers: correction job + late_events counter
Test: event 23:50, arrival 02:10
Fill in
What you type stays in this browser window. We store and send nothing.
What do you process?
clickstream
orders
sensors
log import
mixed
Which aggregation?
minutes
hours
days
session
mixed
How late do things arrive?
seconds
minutes
hours
unclear
mixed
Your prompt
Help me steer late events with a watermark.
Pipeline: [Pipeline]
Window: [Window]
Tolerance: [Tolerance]
Steps: 1. Read event time, processing time and how windows close today. 2. Explain the difference between event time and arrival time. 3. Propose a watermark and a lateness tolerance. 4. Decide what happens to events after close: side output, correction job or drop with a counter. 5. Describe effects on metrics and user reports. 6. Write tests with intentionally late events.
Output: watermark rule, tolerance, straggler policy and tests.
Take particular care to make the lateness boundary visible and count stragglers separately.
Before you answer, check: Are all instructions met? If key information is missing or a placeholder in square brackets is still in the text, ask instead of guessing. Mark assumptions as assumptions.
You have edited the prompt. Changes in the fields above no longer overwrite it.
This is an instruction for an AI, not the result itself.
Prompt that explains a cohort table, compares cohorts at the same age and separates chance differences from real trends.
Prompt · Intermediate · about 8 min
You may use our prompts freely, including at work. Please do not republish them as your own collection. We give no warranty for AI results; check them yourself. Terms of use
