AiBest free content-pack sample v1.2 — local export gate
Reviewed release date: 2026-10-18

WHAT CHANGED
v1.1 documented a response that stopped drafting but placed its pause message
under COPY. v1.2 adds pack_guard.py: a local, standard-library-only Python 3.10+
review/export utility. It rejects legacy COPY text instead of trying to extract it.
It is optional: keep manually reviewing drafts if you cannot use Python.
Nothing is installed, sent, scheduled or published by this utility.
No model API, paid service or third-party Python dependency is required to run it.

WHAT THIS DOES NOT FIX
This is a format/review workflow, NOT a prompt reliability fix, fact checker,
privacy filter, prompt-injection defense, authenticated approval system or proof
of useful marketing. A malicious or mistaken person can create a review receipt.
The digest only catches accidental changes after review, not tampering by someone
with local file access. Python and the file viewer are prerequisites. We exercised
this release with Python 3.14; older compatible versions were not exercised.
The original drafting-prompt.txt remains unchanged; its six model outputs were
not rerun. The added structured-drafting-prompt.txt is an untested format proposal.
A structured DRAFT can still contain false claims or status text: inspect every body.
No customer validation, time savings, marketing or revenue result is established.

TRY THE FICTIONAL FIXTURE
Open sample-pack.json and sample-source.txt in a text editor first. These are
manually authored examples, not new AI outputs or real approved business copy.
Open a terminal in this extracted weekly-content-pack directory:

python3 pack_guard.py check sample-pack.json

Then run all included automated tests (no network; temporary local files only):
python3 -m unittest -v test_pack_guard

REVIEW A REAL PACK
1. Put only approved PUBLIC business material into your own source.txt. Keep
   customer details, passwords and unpublished business records out of AI prompts.
2. Generate drafts using your chosen tool within its independently verified terms
   and costs. The structured prompt is a proposed format, not an exercised model.
3. Save valid JSON in pack.json with status DRAFT, exactly social-1/social-2/social-3/
   email-body asset IDs, and separate review_notes. Use the sample shape.
   For conflicts set status BLOCKED, assets [], and explain in review_notes.
   Never relabel BLOCKED as DRAFT merely to export. Resolve the facts and review
   fresh drafts. Do not manually migrate a paused COPY message into body text.
4. Read every body against the approved source. Check substantive reader value,
   distinct angles, confidential information, links and operational readiness.
   Different labels or no exact duplicates do not prove distinct useful content.
   If email consent/sender/unsubscribe requirements are unresolved, stop. Even
   when they are checked, this tool exports body copy, NOT a send-ready campaign.
5. Only a person who completed every check should create the local attestation:

python3 pack_guard.py review pack.json --source source.txt --receipt review.json --reviewer "YOUR REAL REVIEWER NAME" --checks facts distinct_value privacy links email_consent_setup

6. Export the exact reviewed bodies to a NEW file:

python3 pack_guard.py export pack.json --source source.txt --receipt review.json --out reviewed-copy.txt

This writes four body texts separated by ---; it excludes labels, status and
review notes. The order is three social posts followed by the email body. Inspect
the exported file again and separately approve channel settings, visuals and any
email subject, sender details, address, audience consent and unsubscribe setup.
This utility cannot send it. Changed inputs require a fresh reviewed receipt.
An existing receipt/export is never overwritten; select a new filename.
The identity supplied is a self-attestation, not authentication or human verification.
Do not use this receipt as proof that an AI output is safe or legally compliant.

EXERCISED BEHAVIOR
18 automated tests, including a real subprocess CLI review/export round trip,
BLOCKED refusal, rejected legacy COPY, missing/stale receipt, changed brief/body/
notes, duplicate JSON keys/assets and no overwrite. A negative-control test
explicitly shows unsupported claims CAN pass this format gate and be exported
after a mistaken attestation. Mechanical pass counts are not semantic safety scores.
The conflict tests repeat deterministic local rejection, not repeated AI generations.
See guard-test-results.txt for the actual saved test run.

FEEDBACK WITHOUT SIGNUP
If you voluntarily try this locally, use the site's existing contact route to tell
us: which step you reached, what failed, whether the drafts were meaningfully
useful, and what you had to rewrite. Do not send customer source material or
credentials. This feedback is not a newsletter subscription or purchase.
No payment or email address is required to download this sample.
