What goes in a data manifest, and is there a template I can copy?
Short answer. A data manifest is an inventory of what a company holds, one row per system or kind of record, with counts, sizes, date ranges and language, and no file contents. It is the first document a buyer reads, and the delivery is later checked against it. The template below has the columns we ask for, and a CSV copy you can fill in.
General information, not legal advice.
Why the manifest comes before any sample
A lab deciding whether to make an offer wants to know three things before it looks at a single record: how much there is, how far back it goes, and whether it is the kind of work its models are short of. A manifest answers all three on one page. It also lets the owner describe the archive without opening it to anyone, which is why it is the safest thing to share early.
The manifest becomes part of the deal. If a license is signed, the files delivered are compared with the rows in it. A count that was a guess, or a date range that turns out to have a two year gap, can hold up acceptance and the payment tied to it. So the manifest is worth an afternoon of checking with whoever runs each system.
The columns, one row per system
Write one row for each place records live: the chat workspace, the mailbox archive, the help desk, the project tool, the shared drive, the accounting system. If one system holds very different things, such as a drive with contracts and with photos, split it into two rows. Use the numbers the admin screen or the export report shows, and write down the day you counted.
| Column | What to write |
|---|---|
| System or source | The tool or archive, such as the help desk or the shared drive. |
| Kind of record | Tickets, threads, tasks, drawings, call transcripts, invoices. |
| File type | The format an export gives you: JSON, CSV, PDF, DWG, MP4. |
| Count | Messages, tickets, files or hours, from the admin screen or export report. |
| Size | Gigabytes on disk, as the export reports it. |
| First and last date | Month and year of the oldest and newest record. |
| Language | Mostly English, or the share in each language. |
| Who owns it | The company, or a client whose files you hold under contract. |
| Personal information inside | Which kinds: staff names, customer emails, phone numbers, health details. |
| Notes | Gaps, tool changes, retention settings, anything an export leaves out. |
How to count without opening the data
Most business tools report their own totals. A chat admin page shows message counts by channel, a help desk reports ticket volume by year, and a cloud drive shows file counts and storage per folder. Those figures are enough for a manifest. Nobody needs to read a message to fill it in, and nobody outside the company needs access to produce it.
Where a tool gives no total, run its export to a local disk and count the files there. Keep the export; it is the copy a sample is later drawn from. Note any switch of tools, because a help desk that replaced an older one often starts its history on the day of the move, and the older years sit in a separate archive.
Mistakes that cost time later
The common ones are counting attachments twice, mixing client-owned files into company rows, and rounding a date range to the founding year when the tool only came in later. Another is leaving out a retention rule. If a chat plan deletes messages after a year, the manifest should say so, because the history a lab is paying for may not exist.
Research on dataset documentation makes the same point from the buyer side. The paper that proposed datasheets for datasets argued that every set should travel with a record of how it was made and what it holds, so a user can judge whether it fits. A manifest is the short, early version of that record.
Follow-up questions
Does the manifest need exact numbers?
Use the figure the system reports and the date you read it. A rounded number is fine if it says so. What causes trouble is a guess written as a fact, because the delivery is measured against it.
Should file names go in the manifest?
No. File and folder names often carry client names, people or project codes. Describe the kind of record instead, and leave names for later, after an agreement and de-identification.
Who in the company should fill it in?
Usually an operations lead with the admins of each tool. The owner reviews it, since the owner is the one who signs a license that refers to it.
Didn't find your question? Ask us directly.