Verb, Object, Classifier: write findings anyone can trace

Open the blue biscuit tin in almost any Malaysian kitchen and you will not find biscuits. You will find needles, thread and buttons. Next to it sits a Milo tin full of sugar, and a jam jar of something that might be ketumbar or might be jintan. You only find out by opening each one.

A good kitchen fixes this with a strip of masking tape and a marker: “Gula pasir · beli di pasar tani · Mac”. Now anyone can find the sugar, see where it came from, and know when to buy more.

Research notes are the same kitchen. After the interviews in domain research, you have pages of notes. Every finding is in there somewhere, inside an unlabelled tin. Verb, Object, Classifier (VOC) is the masking tape. It turns each finding into one short, labelled line that anyone can sort, check and trace back to the person who said it.

✦

1

The idea in 30 seconds

Verb: the action

One active verb: check, return, compile, chase.

Watch the 30-second explainer

The idea in 30 seconds0:00 / 0:26

1. Verb: the action One active verb: check, return, compile, chase.
2. Object: what it is done to A specific thing: "KRT grant claims", not "KRT matters".
3. Classifier: the label on the tin Tags for unit, frequency, risk and, always, the source.
4. Trace every row Each row points to a line in the notes. No source, no row.
5. Label the tin once. Find it in seconds, forever.

Key takeaways

If you remember nothing else, remember these.

• A VOC row is one finding in one line: Verb | Object | Classifier.
• The Verb is a single action. The Object is the specific thing acted on.
• The Classifier holds the tags that make the row sortable (unit, process, frequency, risk) and traceable (the source).
• AI can turn interview notes into VOC rows quickly. You check that every row traces to a real line in the notes.

Why findings need a label

Notes record what people said. Findings record what happens. When the two stay mixed, three problems appear. You cannot sort them, so nobody sees that four officers described the same bottleneck. You cannot compare them, because one person wrote “claims” and another wrote “tuntutan geran”. And you cannot trace them, so a finding that reaches a briefing paper has no way back to the person who said it.

Tap between the raw note and the labelled version below.

Raw noteVOC rows

Chase | monthly KRT activity reports | PPD · monthly · risk: medium · source: Int. 1, line 1
Retype | activity details from WhatsApp photos | PPD · monthly · risk: low · source: Int. 1, line 1
Check | KRT grant claims | Finance Unit · monthly · risk: medium · source: Int. 1, line 2
Return | grant claims with missing receipts | Finance Unit · as needed · risk: medium · source: Int. 1, line 2
Compile | district data for HQ requests | PPD · during Parliament sittings · risk: high · source: Int. 1, line 3

What you typically get back

Five rows you can sort by unit, filter by risk and trace to the exact line. Put a second interview through the same pattern and the rows line up.

All names, places and notes on this page are fictional.

The three parts

Here is one finding taken apart. The note is what the officer said. The VOC row is what you keep.

Verb: one action you could watch someone do. “Check”, “return”, “approve”, “compile”, “chase”. Avoid “handle”, “manage” or “deal with”, because they hide several actions inside one word. If a sentence needs two verbs, it is two rows. The Bloom and Simpson guide that comes next gives you a ladder of precise verbs.

Object: the specific thing acted on. “KRT grant claims”, not “finance”. “Monthly activity reports”, not “reports”. A good test: could a new officer pick up the object from a desk? If not, make it more specific.

Classifier: the label on the tin. The classifier is a short set of tagsShort labels attached to a row so you can sort, filter and group it. For example: which unit, how often, how risky, and which source it came from. separated by a dot. It does two jobs:

• Sorting. Tags such as unit, process, frequency and risk let you filter the table later: “show me everything the Finance Unit does monthly with high risk”.
• Tracing. The source tag points back to the exact interview and line. This is what makes a finding traceableAble to be followed back to where it came from. A traceable finding names its source precisely enough that someone else can open the notes and find the original words..

Choosing your classifiers

Pick a fixed set of tags before you start, and use the same set for every row. If one row says “monthly” and another says “every month”, sorting breaks. Agree the allowed values first.

The first four are up to you and your research question. Source is never optional. A row without a source is an opinion with good formatting.

Brief AI to build the VOC table

Sorting forty pages of notes into rows is exactly the kind of grunt work AI handles well. The trick is to give it your fixed tags and force it to quote its source. Tap each coloured part of the brief below.

Dissect a prompt

A brief that turns notes into VOC rows

ExploreTest yourself

Tap a coloured part of the prompt to see what it does.

You are a careful research assistant who turns interview notes into structured findings. Below are notes from three interviews with district officers, labelled Int. 1 to Int. 3, with numbered lines. Names have been removed. Allowed tags: Unit (PPD, Finance Unit, State office, KRT committee), Frequency (daily, weekly, monthly, as needed), Risk (low, medium, high). Extract every work activity mentioned into a table with columns Verb, Object, Unit, Frequency, Risk, Source and Quote. One verb per row. Source is the interview and line number; Quote is the exact words, under 20 words. If a tag is not stated in the notes, write "unclear" instead of guessing.

Here is the full brief, ready to copy. Paste your anonymised notes where it says so.

You are a careful research assistant who turns interview notes into structured findings for a district unity office.

Below are notes from interviews, labelled Int. 1, Int. 2 and so on, with numbered lines. Names, MyKad numbers, addresses and phone numbers have been removed.

Allowed tags:
- Unit: PPD, Finance Unit, State office, KRT committee
- Frequency: daily, weekly, monthly, as needed, seasonal
- Risk: low, medium, high

<notes>
(paste your anonymised notes here)
</notes>

Extract every work activity mentioned into a table with the columns Verb, Object, Unit, Frequency, Risk, Source, Quote.
- Verb: one active verb, in the imperative (Check, Return, Compile).
- Object: the specific thing acted on.
- Source: interview and line number, for example "Int. 2, line 7".
- Quote: the exact words from that line, under 20 words.
Use only the allowed tag values. If a tag is not stated, write "unclear". Do not add activities that are not in the notes.

Check that every row traces to a source

AI is fast at this, and it is also confident when it is wrong. It may merge two findings, soften a quote, or add a sensible-sounding row that nobody said. Your job is the check. It takes ten minutes, not a day.

1. Count. Does the number of rows look right for the notes? Far fewer rows usually means merging; far more usually means invention.
2. Spot-check quotes. Pick five rows at random. Open the notes at the stated line. Are those the words?
3. Hunt for orphans. Any row with no source, a vague source (“Int. 2”) or a quote you cannot find comes out of the table.
4. Read the “unclear” tags. Each one is a question for your next interview.

You can also ask a second, separate check pass:

You are a strict reviewer. Here is the VOC table and the original numbered notes. For each row, find the quoted words at the stated line. Return only the rows where the quote is missing, different, or at a different line, and say what you found instead. Do not fix the table.

A clean result from the reviewer is not the end. It means the quotes match. It does not mean the findings are complete or that the risk ratings are right. Those are your call.

Common mistakes

• Two verbs in one row. “Check and approve claims” hides a handover. Split it into two rows; the handover is often where the delay is.
• A vague object. “Manage KRT” tells you nothing. Ask “manage what, exactly?” until you reach a thing you can count.
• Free-text tags. “Monthly”, “every month” and “bulanan” will not sort together. Fix the allowed values first.
• Trusting the table because it looks tidy. Neat formatting is not evidence. The Quote and Source columns are.

Where this fits

VOC sits between collecting and designing. Domain research gave you the notes. VOC turns them into clean, traceable findings. Next, Bloom and Simpson helps you sharpen the verbs, so you can see what each activity needs someone to know and do.

Check your understanding

Check yourself

VOC in five questions

1/5

Pick an answer, then tap Check answer. You can also use the arrow keys to move between answers.

Question 1 of 5: Which of these is a well-formed VOC row?

AHandle | KRT matters | PPDBCheck | KRT grant claims | Finance Unit · monthly · source: Int. 1, line 2CFinance is slow with claims

Check answer

Nice work!

Lesson complete

You made it to the end. Take the quiz above if you have not yet, for bonus XP.

• 0 XP
• 0 day streak