How to Document Creator-Vetting Evidence

What matters most

Evidence records should answer what was observed, where, when, by whom, how it relates to the campaign, what alternative explanations exist, and what decision it supported.

Preserve coverage and absence carefully. Not finding evidence in the reviewed sample is not proof it never existed; unavailable, deleted, private, unmeasured, and measured-with-no-finding are different states.

Use role-based access, redaction, retention schedules, correction history, and approved evidence types. High-risk personal data should not be copied into general campaign notes.

A practical workflow

1. 01

   Define standard evidence fields and approved source types.

2. 02

   Attach source, date, context, observation, inference status, and confidence to each finding.

3. 03

   Record coverage, contradictions, alternative explanations, and missing data.

4. 04

   Connect findings to outcome, conditions, reviewer, and recheck date.

5. 05

   Apply access, correction, retention, deletion, and incident controls.

Three readers, three different needs

A vetting record is written for three people who will never read it together. The first is the next reviewer, six months from now, who needs to know what was checked, what was found, and what is no longer current, so they do not repeat the whole exercise.

The second is a colleague defending the decision internally, who needs the reasoning: which finding drove the outcome, what alternatives were considered, and who signed it off. A record that stores a conclusion without the reasoning cannot support anyone.

The third is the reader you hope never arrives: someone assessing the decision after a dispute, a complaint from the creator, or a regulatory question. That reader needs sources, dates, and a clean separation between what was observed and what was inferred. Writing for all three at once is not much extra work, and it is nearly impossible to reconstruct afterwards.

Design for what you will not keep

The instinct in vetting is to save everything, and it is the wrong instinct. Every piece of personal information about a creator that you retain is something you must secure, may have to disclose, and could be obliged to correct or delete, and much of it does not affect any decision you will make.

Decide what the record needs before you start collecting: the finding, the source, the date, the relevance, the outcome. That is usually enough. Wholesale archives of someone's personal posts, screenshots of unrelated material, and speculation about their private life are liabilities that do no analytical work.

Be especially careful with anything touching health, beliefs, politics, sexuality or ethnicity. In several jurisdictions those are special categories with additional obligations attached, and they end up in creator files by accident rather than by decision, usually because somebody screenshotted a whole profile rather than the specific thing that mattered.

Every record has a shelf life

A vetting conclusion is a statement about a moment, and it decays. Content is published and deleted, obligations expire, audiences change, and a creator who was a poor fit for one campaign may be right for the next. A file with no date on it will be reused indefinitely and will eventually be wrong.

Attach two things to every material finding: the observation date and a recheck trigger. The trigger can be time-based, or event-based, such as before any campaign over a stated value, or before any renewal.

Distinguish stale from false when you revisit. A finding that is simply old should be re-verified rather than deleted or trusted. A finding that has been superseded by later evidence should be updated with the new evidence and the date, keeping the correction history, because the record of how a view changed is frequently the most useful thing in the file.

Common mistakes

• Recording only a red, yellow, or green score.
• Saving screenshots without URL, date, or context.
• Treating not found as proof of absence.
• Copying sensitive personal material into broadly accessible tools.

Working checklist

• Material conclusions link to dated source evidence.
• Direct observation and inference are distinguishable.
• Coverage and missing states are explicit.
• Decision, conditions, reviewer, and recheck trigger are recorded.
• Access and retention are proportionate.

Questions and answers

How long should vetting records be kept?

Long enough to support the decisions they informed and no longer, with a written retention period rather than an implicit one. Keeping personal information indefinitely because nobody decided to delete it is the default failure here. Your legal or privacy team should set the period; the job of the process is to make sure deletion actually happens when it arrives.

Should a creator be able to see their file?

Assume they might. Depending on jurisdiction and what you hold, a creator may have a right to request the personal information you keep about them. Writing every record as though the subject could read it is a good discipline regardless: it eliminates speculation, unverified assertions and casual remarks, none of which were helping the decision anyway.

What is the difference between a finding and an inference?

A finding is something observed with a source and a date, such as a specific post on a specific day. An inference is a conclusion drawn from findings, such as a judgment that a pattern indicates a risk. Both belong in the record and they must be labelled differently, because an inference that gets recorded as an observed fact is how a record becomes indefensible.

Where should vetting records live?

Somewhere with access control, an audit trail and a retention policy, which usually rules out a shared drive and a group chat. The sensitivity of the content, not its volume, sets the requirement: a handful of records containing findings about named individuals needs more care than a large file of campaign metrics.