Field note · 7 min read
A field guide to the meeting sentences that sound like decisions and are not: the hypothetical, the mumbled yes, the reversal nobody heard. Plus the five phrases that fix them, so the person reading the notes on Monday (human or otherwise) does not have to guess.
TL;DR
Most delivery meetings are 80 percent exploring and 20 percent deciding, delivered in the same tone of voice. That is fine for the people in the room, who can read a shrug. It is a disaster for everyone who reads the notes afterwards. Every topic should end with one of three words and one sentence. That is the whole post. The rest is a tour of what happens when you skip it.
The field guide
All five are real. All five were said in meetings we have sat in, run, or transcribed. Names changed, grunts preserved. Each one gets a pin, a Latin name it does not deserve, what the room heard, and what actually happened.
In the wild
What the room heard
Audit logs are in scope. Two people agreed.
What actually happened
An if-clause and two grunts. Nobody chose SSO first. Nobody chose audit logs. The sentence is true and the decision does not exist.
In the wild
What the room heard
Priya owns the migration.
What actually happened
Priya made a noise. Priya was reading Slack. The request is on the record, the answer is a vowel, and the two are 40 words apart in the transcript.
In the wild
What the room heard
Ship CSV. It was said clearly, with the word "done" attached.
What actually happened
Reversed 29 minutes later, in a subordinate clause, during a different agenda item, followed immediately by the word "anyway". The minutes still say CSV.
In the wild
What the room heard
It is on the list.
What actually happened
Which list? Nobody said parked, nobody said no, nobody said when. It now lives in three people’s heads at three different priorities.
In the wild
What the room heard
Tom is chasing the vendor.
What actually happened
Tom was on mute. Eye contact is not a RACI. Nothing in the transcript contains the word Tom, which is the only part of this that a transcript can prove.
The honest bit
Here is how Backlog.cloud keeps itself honest. Every story, bug, task, epic, spike, requirement, and action item it produces from a meeting has to carry the verbatim sentence it came from and a confidence rating of high, medium, or low. If the wording is not in the transcript, the item does not exist. Then a second model rereads the whole transcript and drops anything whose quote is not actually there, anything the conversation contradicts, and anything the team explicitly parked. Low-confidence items get a red chip that says Review. You see the quote before you push a single ticket.
That catches the made-up sentence. It does not catch the real sentence that meant something else. Look back at the five specimens. In every one, the quote is genuine. Dan really did say the thing about audit logs. Mia really did ask Priya. Sam really did say CSV, with the word done attached. The problem is never the sentence. It is the if-clause in front of it, the grunt after it, or the reversal 29 minutes later during a different topic.
The bot is not psychic. Neither is your new starter reading the minutes on Monday, or the engineer six weeks from now asking why the export is CSV. Every one of them is hunting the same transcript for the same missing sentence. The difference is that the humans will eventually walk over and ask you. The notes will not.
The one-line version
Grounding proves a sentence was said. Only you can make the sentence mean what you think it means.
The fixes
None of these are clever. That is the point. They are short enough to say without breaking the flow of a conversation and blunt enough that nobody in the room, or in the transcript, can mistake them for exploring.
Fixes: The Hypothetical
Before
“So, yeah, CSV then, probably.”
After
“Let’s decide that: the export ships as CSV. Rationale: legal is fine with it and the enterprise pilot only reads CSV.”
Three words, and the room knows a decision is happening rather than a vibe. Attach a rationale while you are there. Future you will want it.
Fixes: The Half-Park
Before
“Yeah, maybe, at some point.”
After
“Parking auth staging. Not this sprint. Revisit at the next planning, Dan brings a one-pager.”
Parking is a decision. It has a destination and a date. "Maybe later" is a shrug with extra syllables.
Fixes: The Quiet Reversal
Before
“Actually hold the export, legal wants a look, anyway, onboarding.”
After
“Scrap the CSV decision from earlier. Export is on hold until legal signs off. Sam chases legal by Thursday. Now, onboarding.”
Reversals are the sentence most likely to go missing, because they arrive while everyone is thinking about something else. Name the thing you are reversing. Then move on.
Fixes: The Owner by Proximity, and the Backchannel Commitment
Before
“Someone should chase the vendor.”
After
“Tom chases the vendor about the SLA by Friday. Tom, say yes with your mouth.”
A name, a deliverable, a date, and an audible yes from the named person. Silence is not consent, it is usually a dropped connection.
Fixes: Everything that got mentioned and never resolved
Before
“(silence, then the next topic)”
After
“To be clear, we are not moving auth to its own staging this quarter. Out of scope, on purpose.”
The out-of-scope sentence is worth more than the in-scope one. It stops the idea growing overnight in someone’s notes.
The habit
Before the chair moves to the next agenda item, they say one of three words and one sentence. That is it. Twenty seconds. If the room cannot agree on which word applies, congratulations, you have just found the real agenda item.
Decided
What we chose, why, and who owns the first step by when. If there is no owner, it is not decided, it is hoped.
Parked
What we are not doing now, where it goes, and when it comes back. A parking spot has a location. "Later" is not a location.
Reversed
Which earlier decision just died, and what replaces it. Say the old one by name so the notes can find and kill it.
Field note: the teams that do this consistently also finish meetings early, because a topic that has been closed cannot be reopened by the person who was on mute.
FAQ
Yes, and almost nobody does it. Most teams run meetings that are 80 percent exploration and 20 percent decision, and the two are said in exactly the same tone of voice. The fix is not a new framework. It is saying one of three words at the end of every topic: decided, parked, or reversed. The rest of this post is about why that habit matters more once something other than a human is taking the notes.
No. There is no magic keyword. The extractor reads intent from the conversation, and a plainly stated decision with a named owner is logged whether or not you use the phrase. What explicit phrasing changes is the confidence rating attached to the item and the signal the verifier pass has to work with. A parked topic that was never called parked has no anchor for the verifier to drop it on. A reversal buried in a subordinate clause looks, on paper, like a decision that still stands.
Edit the item. Every artefact in Backlog.cloud is editable before it is pushed, and the decision log carries the verbatim quote it was anchored to, so the record of what was said in the room stays intact even when the decision itself changes. If the whole meeting was misread, re-tag it under a different ceremony type and re-extract from the same transcript. That does not spend another generation.
Every meeting type in Backlog.cloud ships with a cheat sheet: a run of show, the pitfalls for that ceremony, and the questions worth asking in the room. Most of them are versions of the advice in this post. Say the sprint goal out loud. Park unknowns as spikes. Catch every "we could just" and park it. Name an owner on every action. The live coaching pass nudges the host when a topic is discussed with no owner, no date, or no scope edge, which is usually the moment one of the five specimens above is about to be born.
Related reading
Product
78 questions for product and delivery leads. INVEST, 3Cs.
ReadDelivery
59 questions for engineers. SMART, DACI, Ready vs Done.
ReadDelivery
56 questions for anyone running a spike. The time-box is the spike.
ReadStrategy
63 questions for epic owners. Hypothesis, metrics, slicing.
ReadSay it once, out loud
The bot joins your Zoom, Google Meet, or Teams call, or you paste a transcript or upload a recording afterwards. Every decision, action, and story it produces carries the verbatim sentence it came from and a confidence rating, gets checked by a second pass, and lands in Jira, Azure DevOps, Linear, GitHub, Notion, or ClickUp only when you say so.