The takeaway
A useful manifest explains what is in the package, how records join, how the data changed, and where the known limitations are.
Make the data navigable before asking someone to evaluate it
A folder of exports tells a buyer that records exist. A connection manifest explains how to use them together. It is the difference between discovering a task’s evidence by trial and error and having a documented path to inspect.
The manifest can be a structured file with accompanying human-readable notes. Its format matters less than whether a reviewer can find the source, understand the keys, reproduce the joins, and identify missing context.
This is one of the evidence categories in our AI-Ready Data framework. Confirm the actual manifest deliverable and format when scoping an assessment with Clear Harness.
Start with a precise inventory
Name the dataset and version, source systems, export dates, covered periods, included tables or files, and the owner of the documentation. Distinguish when a record was created from when it was exported.
| Group | Examples |
|---|---|
| Package identity | Version, generation date, file inventory, checksums where useful |
| Schema | Field definitions, data types, allowed values, missing-value conventions |
| Relationships | Join keys, cardinality, direction, evidence, matching confidence |
| Time | Event timestamps, timezone rules, history coverage, known gaps |
| Transformations | Redaction, de-ID version, normalization, deduplication, exclusions |
| Scope | Permitted review context, limitations, unresolved questions, contact |
The manifest should describe the delivered package, not the complete source environment that remains inside the business.
Show how the joins work
For each important relationship, name both fields and the expected interpretation. Explain whether the key is unique within a system, a tenant, or the whole package. Note when a relationship is one-to-many rather than one-to-one.
Illustrative join: issues.project_id → projects.id connects an issue with its project. comments.file_id → files.id connects a comment with a file. A separate version reference may still be needed to identify the file revision being discussed.
Separate explicit foreign-key links from inferred matches. A similarity score does not turn an inferred relationship into ground truth. Cross-reference the linkage methodology and any manual review.
Record lineage and transformation choices
W3C’s PROV model provides a vocabulary for describing origins and the activities associated with producing data. A business manifest can apply that idea by naming the export, transformation, and dataset version behind a reported result.
Document what happened to identifiers, timestamps, file paths, deleted records, and duplicate exports. Link the explanation to the de-ID retention checks. Keep re-identification keys and protected reference material outside the recipient’s package.
Documentation should also state what cannot be reconstructed. If file version history was unavailable, a reader should see that directly rather than assuming the latest file represents every earlier state.
Test the manifest with someone who did not build it
Give an authorized reviewer one workflow and ask them to locate its request, actions, supporting files, decision, and outcome using the documentation. Capture where they need undocumented knowledge.
Then revise the manifest rather than simply answering the questions in a private conversation. That turns a one-time explanation into something the next reviewer can use. Keep a change log when joins or definitions change.
The exercise also exposes high-impact preparation fixes: ambiguous keys, undocumented status codes, mismatched timezones, or missing attachments.
Make the next review easier to scope
Explore quality ranking with Clear Harness, our partner, and ask which documentation and evidence can accompany the assessment. Bring your source inventory and any existing schema notes; you do not need to start by sending an unrestricted export.
For the commercial starting point, take the Troveo business assessment (referral link; we may be compensated). A manifest explains the package, while a licensing discussion establishes the proposed use and business terms.
Sources & further reading
Sources checked October 9, 2026.