Test different formats first.
Start with a useful carousel, an actual product demonstration and a short sensation POV. Compare the formats before spending a production cycle on three similar opening lines.
Run a small practitioner-referral pilot alongside the creative work. NSM’s existing practitioner relationships make this a particularly relevant route to test.
Watch the actual creatives.
Five complete videos with sound and playback controls, plus two carousel-cover screenshots. Each sits beside its performance evidence and a specific Stateshift test.
Three repeatable growth channels.
Ranked for Stateshift’s assets and ability to test. These are proposed pilots, with historical results attributed to their sources.
Five competitors, distinct evidence.
Paid creator content, product positioning and historical examples are labelled separately. An available post is not proof of paid amplification.
| Competitor | Message already occupied | Observed execution / CTA | What to test for Stateshift |
|---|
Rootd has two active Meta ads verified on 1 October. The other non-Condri profiles mainly describe product positioning. Full competitor inventories and paid acquisition economics remain unverified.
Earlier contextual hypotheses.
Retained from the first pass. The format tests above now take priority; after-work framing is one possible execution, not a proven winning angle.
Inspect the evidence.
Open each observation for provenance, dates and limits. Views are public metadata, sometimes rounded, and never a revenue proxy.
Excluded evidence
Three hooks. One shared body.
Shared sequence
Production
Claims and permissions
Measurement
Before launch
Reuse the existing archive.
The 6 September AppKittie collection: 102 ad records, 71 distinct creative files. It is a dated comparison set, not a current-ad feed for the five competitors above.
Refresh, review, repeat.
Run on demand. There is no scheduled job or unattended publishing process.
What is automated
- Refresh the known public pages and exact TikTok URLs. Retrieve metadata only; no new media download.
- Save a dated snapshot. Preserve each source’s last successful comparison hash if it becomes unavailable.
- Flag changed text or captions and failed reads. Keep curated observations separate from fresh retrieval timestamps.
- Rebuild this page and the Markdown brief from the canonical research file.
What still needs review
Discover new posts; inspect video/audio and paid-partnership labels; reconsider the hypotheses; update the brief; verify rights, attribution and budget before a launch. Reddit sources are reviewed manually and retained only as aggregate themes.
Refresh status
Commands
From the Stateshift repository, using Python 3 and the existing curl / yt-dlp installation:
python3 "Notes/Creative Intelligence/creative_loop.py" validate python3 "Notes/Creative Intelligence/creative_loop.py" refresh python3 "Notes/Creative Intelligence/creative_loop.py" build
To recheck just one source:
python3 "Notes/Creative Intelligence/creative_loop.py" refresh --source condri_home
After inspecting a flagged source and updating research.json where needed:
python3 "Notes/Creative Intelligence/creative_loop.py" acknowledge --source SOURCE_ID --note "What was checked and changed"
A source staying unchanged does not clear an earlier review task. An explicit acknowledgement records the review without pretending the original creative observation was made again.