Nine fields, every time
The receiver should be able to pick up the task cold, see how sure the sender is about each claim, and know what to send back. If a field has nothing in it, the message says so instead of leaving it out.
- SOURCE → TARGET
- Who is sending, and who should act on it.
- TASK
- The piece of work, in a sentence or two.
- STATE
- Where the work stands: new, in progress, blocked, ready for review or done.
- OBSERVATIONS
- What the sender saw directly while working, with times where they matter.
- EVIDENCE
- The claims the work rests on. Each one carries a label and says where it came from.
- HYPOTHESES
- Ideas the sender thinks are worth testing but hasn't confirmed.
- UNCERTAINTIES
- What the sender doesn't know or isn't sure about.
- REQUEST
- What the receiver is asked to do.
- RETURN
- What the receiver should send back, and in what form.
Worked example: a handoff and its reply
The person asked for a laptop under $900 for writing and video calls. Athena has the prices and hands the battery question to Sentinel.
SOURCE → TARGET Athena → Sentinel
TASK Check the battery-life claims for the two shortlisted laptops.
STATE In progress. Prices are done; battery and returns are open.
OBSERVATIONS Both seller pages were read at 9:40 this morning.
EVIDENCE [REPORTED] The maker says model A runs 14 hours. Source: maker's spec page.
[REPORTED] A review says model A ran 10 hours in its test. Source: review site.
[OBSERVED] Model A was listed at $849 on the seller's page at 9:40.
HYPOTHESES The maker measured its figure with the screen dimmed.
UNCERTAINTIES Which test is closer to video-call use. Whether the price holds past today.
REQUEST Find a battery test that matches video calls, or say that none exists.
RETURN One line per claim, with its label after checking and the source.
SOURCE → TARGET Sentinel → Athena
TASK Battery claims for model A, checked.
STATE Ready for review.
OBSERVATIONS The review's test method is on its site: video playback with wifi on.
EVIDENCE [REPORTED] 14 hours (maker). Test conditions not stated.
[REPORTED] 10 hours (review). Video playback, wifi on.
[INFERRED] For video calls, expect something closer to the review's figure.
HYPOTHESES None new.
UNCERTAINTIES No test found that uses the webcam. [UNKNOWN] Battery wear after a year or two.
REQUEST Use 10 hours in the comparison and show the maker's figure beside it.
RETURN Nothing further. Mark the battery check done.
What to notice
- The guess about a dimmed screen doesn't come back as a fact. Sentinel couldn't confirm it, so the reply says the maker's test conditions aren't stated.
- Both battery figures stay REPORTED. Neither agent ran a test, so neither figure gets promoted, even though one is now preferred.
- The price carries the time it was seen. A price from 9:40 is only known for 9:40.
- The reply goes back to Athena, who asked for the check and puts the comparison together. Nothing in either message buys anything.
Go deeper: why uncertainties get their own field (layer 3 to 4)
A receiver who can't see the gaps fills them with guesses. Keeping UNCERTAINTIES apart from HYPOTHESES tells the receiver the difference between "we don't know" and "we think so, but nobody has checked".
Those two fields line up with the UNKNOWN and HYPOTHESIS labels on the next page.
Build a message
Choose a sample task and fill in the fields. The message on the right updates as you go, and the review notes underneath point out what Sentinel would send back.
The builder needs JavaScript. The two worked examples above show complete messages.