creator_vetting v1.0.0

name: "creator_vetting" description: "Vet a creator before a brand deal: audience authenticity checks, engagement quality over follower counts, fake-engagement and pod signals, brand-fit review, past-post red flags, disclosure history, and a scored go or no-go sheet produced by a bundled scorer. Triggers when someone asks to vet, audit, check, or evaluate an influencer or creator before paying, gifting, or signing them, or asks whether a creator's followers or engagement are real."

Creator Vetting

Purpose

Decide, with evidence, whether a creator is worth paying before any money, product, or contract moves. Follower count is the weakest number on the profile. This skill checks the audience that actually shows up: whether engagement is real, whether the audience matches the buyer the brand needs, whether growth was earned, and whether the creator's past posts create compliance or brand-safety risk. The output is a scored sheet with a clear verdict: GO, CONDITIONAL, or NO-GO, plus the evidence behind it.

Vetting is not deal-making. Pricing, negotiation, and contracts live in creator_collab_deals and brand_deal_negotiation. Briefing a creator who passed lives in ugc_creator_briefs. This skill ends at the verdict.

Workflow

1. Set the bar before looking at the creator. Write down the target buyer first: countries, age bands, language, and the niche the campaign needs. A creator cannot fail a fit test that was never defined. Also write the brand's own red-line list (topics, claims, competitor ties) before opening the profile, so the scan runs against fixed criteria instead of vibes.

1. Pull the raw sample. Use the creator's public profile plus free history tools where available. Collect:

• The last 10 to 20 posts with likes, comments, shares, saves, and views per post.
• Follower growth history, long enough to see jumps and flat stretches (growth-chart tools such as Social Blade plot this).
• A comment sample: skim at least 50 comments spread across at least 5 different posts, never one post alone.
• Past sponsored and gifted posts, enough to judge disclosure habits. Scan at least 20 past posts covering roughly 90 days for brand safety.
• First-party analytics from the creator: audience geography, age bands, gender split, and recent post performance screenshots. Ask before scoring fit. A refusal to share basic analytics is recorded as a gap, not treated as a pass.

1. Compute engagement the right way for the platform. Instagram rate: (likes + comments) / followers. TikTok rate: (likes + comments + shares) / views. The denominators differ, so never compare a TikTok rate to an Instagram band. Judge the average across the sample against the tier band in references/rubric.md, and judge the creator against their own baseline too: real audiences react unevenly, so a rate that barely moves from post to post is itself a flag.

1. Run the authenticity checks. Work through the signal catalog in references/rubric.md: growth shape, comment quality, repeat commenters and pod patterns, saves and shares versus comment volume, views versus followers, geography and language match. One flag is a question. Several flags pointing the same way are a verdict.

1. Review brand fit and past posts. Score fit separately from authenticity. An honest creator with the wrong audience is a NO-GO for this campaign and no accusation of fraud. Then scan past posts and public coverage for red flags: undisclosed ads, competitor deals inside the exclusivity window the campaign needs, claims the brand cannot repeat, and controversies with evidence attached. For creators above roughly 100k followers, check what others say about them (news, podcasts, comment threads), not only what they post.

1. Check disclosure history. Under the FTC Endorsement Guides (16 CFR Part 255, revised 2023), a material connection such as payment, free product, or a discount must be disclosed clearly and conspicuously, in plain terms a consumer cannot miss. The advertiser is responsible for monitoring its endorsers, so a creator who buries or skips disclosures on past sponsored posts is a compliance risk the brand inherits. Record the share of past sponsored posts with a clear disclosure.

1. Score it. Fill the JSON input described in bin/score_creator.py and run it:

python3 bin/score_creator.py input.json python3 bin/score_creator.py --demo

The scorer returns section scores out of 100, evidence notes, hard fails, and the verdict. Hard fails force a NO-GO no matter the total: a severe red flag, 60% or more generic comments, an unexplained one-day growth jump of 50% or more of the follower count, or clear disclosure on under half of past sponsored posts.

1. Write the sheet and stop at the verdict. Deliver the output contract below. Name the evidence for every score. If a section could not be measured (no analytics shared, private account, too few posts), say so and score CONDITIONAL at best until the gap closes.

Output Contract

One vetting sheet per creator, containing:

• Header: creator handle, platform, follower count, date vetted, campaign and target buyer the fit was judged against.
• Verdict: GO, CONDITIONAL (with the exact conditions), or NO-GO. One line, up top.
• Scores: the five sections (engagement health 25, comment authenticity 25, audience fit 20, growth integrity 15, disclosure and safety 15) and the total, as printed by the scorer.
• Evidence notes: the measured facts behind the scores: average engagement rate and its tier band, generic and repeat commenter shares from the sample, audience match percentages, largest growth jump and whether a content event explains it, disclosure rate, red flags found with post links or dates.
• Gaps: anything requested but not provided, or not measurable, and what it would take to close it.
• Next step: proceed to deal terms, proceed only if the conditions are met, or pass, with the reason in one sentence.

Operating Rules

1. Numbers before narrative. Compute the engagement math and read the comment sample before reading the media kit. Media kit numbers get cross-checked against public data, never copied into the sheet as fact.
2. Averages, never cherry-picked posts. Judge views and engagement across the full sample. A single post the creator chose to share proves nothing either way.
3. Flags are not proof. One anomaly gets a question and a note. A verdict of fraud needs converging signals. Write what was measured, not what was suspected.
4. Fit failure is not fraud. Keep the two findings separate in the sheet. Wrong audience, wrong language, or wrong niche ends the deal for this campaign without impugning the creator.
5. Bands are starting points. Tier bands in the rubric reflect commonly reported 2026 ranges and shift by niche and platform changes. When a niche has its own known norm, use it and note the substitution.
6. Evidence or it did not happen. Every red flag in the sheet links to a post, a date, or a named source. Unverifiable claims about a creator do not go in the sheet.
7. The scorer does not decide alone. A human reads the hard fails and any severe flag before a NO-GO is final. The scorer's job is to make sure nothing measurable was skipped.
8. No outreach from this skill. Vetting produces a sheet. Contacting the creator, negotiating, or contracting happens in the deal skills, after a GO.