Compliance Audit Before Posting¶

The final check before a workflow goes live. A clean compliance audit catches banned words, unapproved claims, and platform-risk content before it costs you a takedown.

When you run an audit¶

flowchart LR
    A[Workflow generated] --> B[Candidates picked]
    B --> C[Compliance audit]
    C --> D{Clean?}
    D -->|yes| E[Post]
    D -->|issues| F[Edit + regen affected scenes]
    F --> C

The audit happens after generation, before posting. The Generation Runner produces the final outputs; you've picked candidates; now you check for compliance issues before publishing.

What the audit checks¶

Check	Looking for
Banned words	Words in the brand file's banned list (e.g. "cure", "guarantees")
Unapproved claims	Claims about the product that aren't in the product file's approved-claims list
Off-brand angles	Angle categories the brand has banned (e.g. "weight loss" if the brand doesn't support that angle)
Platform-risk content	Things TikTok / Meta / Facebook might flag (specific medical claims, before/after framing, etc.)
AI Label Trick presence	For TikTok Shop content — was the dark-screen flash applied?
Sales channel CTA correctness	Does the CTA match the configured sales channel? (e.g. "comment KEYWORD" makes no sense for a Meta Shop video)
On-screen text	Text overlays in post-production — do they match the script and brand voice?

How to invoke¶

You: run a compliance audit on XYZS1-V1 before I post it.

Claude:
  Loading the brand file (XYZ Wellness) for banned words and approved angles.
  Loading the product file (magnesium-glycinate) for approved claims.

  Walking the script across all account tabs...

  AUDIT RESULTS:

  XYZS1-V1 — Account A
  - Banned words: clean
  - Approved claims: all referenced claims are in the approved list
  - Off-brand angles: clean
  - Platform risk: minor — Scene 04 says 'completely fixes' which TikTok
    sometimes flags as a strong claim. Suggest softening.
  - AI Label Trick: NOT YET APPLIED (post-production pending)
  - Sales channel CTA: matches TikTok Shop (uses 'tap Shop button')

  XYZS1-V1 — Account B
  - Banned words: clean
  - Approved claims: clean
  - Off-brand angles: clean
  - Platform risk: clean
  - AI Label Trick: NOT YET APPLIED
  - Sales channel CTA: matches TikTok Shop

  ACTION ITEMS:
  1. Soften Scene 04 in Account A's tab — 'completely fixes' →
     'really moved the needle for'
  2. Apply AI Label Trick in post-production for both accounts

  Want me to do (1) as a surgical edit and bump V1 → V1-1?

The audit runs across all account tabs in a multi-account workflow. Each tab is reported separately.

What to do with the results¶

Clean audit: Proceed to posting. The audit is your green light.
Minor issues: Surgical edits to the affected scenes. Bump V0-N or decimal-bump as appropriate. Regenerate the affected scenes. Re-audit.
Significant issues: May warrant a Lvl 1-2 variant if multiple scenes need wording changes. Bump per variant rules.
Critical issues: Multiple banned words, unapproved claims, off-brand angles — likely means something went wrong upstream. Reconsider the brief or escalate to whoever owns the brand strategy.

Common audit findings¶

Strong-claim language slipping in: "Completely fixes", "100% works", "miracle". These slip through when the Script Writer's compliance filter is set too loose or the brand file is missing the word. Add to the brand's banned list, regenerate.
Adjacent-claim leakage: The product file approves "supports sleep onset" but the script says "improves sleep quality" — same general territory but not the exact approved claim. Tighten the script to use approved wording.
Off-brand angle drift: The brand bans "weight loss" angles, but a script targeting "feeling lighter and more energized" reads as a weight-loss euphemism. The audit catches the implied angle even when the explicit words aren't there.
Platform-specific risk: TikTok specifically penalizes "before/after" framing for supplements without proper disclaimers. Even if the brand allows the angle, the platform won't.
Missing AI Label Trick: For TikTok Shop workflows, the audit flags whether the post-production AI Label Trick has been confirmed applied. If not, do that before posting.

Compliance audit vs. compliance enforcement¶

Two distinct things:

Enforcement: Happens at write-time. The Script Writer reads the brand and product files and filters out banned words / unapproved claims as it writes. This catches 95% of issues at the source.
Audit: Happens at post-time. A final check to catch anything that slipped through enforcement — newly banned words, edge cases, things added by variants or surgical edits, post-production text overlays.

You can't skip either. Enforcement is upstream protection; audit is the final gate.

Documenting audit results¶

Save audit results to Assets/{workflow}/audit-{date}.md:

# Compliance Audit — XYZS1-V1

**Date:** 2026-05-12
**Auditor:** Claude (the Manager)
**Workflow version:** V1
**Status:** PASSED with 1 minor edit

## Per-account results
[as above]

## Action items completed
- [x] Softened 'completely fixes' to 'really moved the needle for' (Account A, Scene 04)
- [x] Bumped V1 → V1-1 with the surgical edit
- [x] Confirmed AI Label Trick will be applied in post (timestamp in editor: TBD)

## Sign-off
Workflow is cleared for posting to all target accounts.

This becomes the audit trail. If a workflow later gets flagged on a platform, the audit file documents what was checked at posting time.

When you're ready¶

→ Next: Client Delivery via Drive — packaging and sharing the final assets with the client.