Strategy snapshot: hip batch 1

What this is. One file for this batch of nine briefs. The first half is frozen now, before any ad copy and before any launch. The second half is empty and gets filled after the test with real Facebook and store numbers, which skill wrote each native, and what we saw. Its job is to make the brief-making skill better from real results.

Status: proposal, not approved. Full file on machine 1: snapshots/HIP-BATCH-1/STRATEGY-SNAPSHOT.md (Parts A to F).

1. The nine briefs and what was predicted

My ranking is locked now so it can be scored against real results later. One clash the data will settle: the pitch judges put "The same chair" at the bottom, Argon's later read of the full brief put it near the top, and you liked it.

2. How the batch was made

• 18 steps, each with its proof file: slots on the ClickUp matrix, one native donor per slot, one Argon consult, pitches, four blind judges, full 12-part briefs, your 80/20 virality check with Argon fed winners, a fresh checker per brief, fixes.
• Three tests: Argon High against Carbon High as the reader; three ways of planning a batch (matrix first plus consult won, 4 of 4 judges); Opus 5.5 HIGH against MAX for writing briefs (HIGH won both, small margin).
• Both chat transcripts are named by path, with the work folders.
• Models: coordinator Fable 5.2; writers Opus 5.5 HIGH; checkers and judges Opus 5.5 MAX; reader Gemini 4 Argon High; this snapshot checked by Gemini 4 Carbon High.

3. The link: brief to ad to order

Today nothing links a brief to its ads or its orders. The brief registry has no field for ad ids. In the 8 November data pack, 828 of 828 ads carry no link tag. Your data rules say: store money only for the whole funnel until a join exists.

The chain proposed:

1. Each brief has a code: HIP-B01 to HIP-B09.
2. The code is on its ClickUp card.
3. The ad set and ad names start with the code, for example HIP-B03 | The trip | 14.11 and HIP-B03-H1-P2 (hook 1, picture 2).
4. Every ad carries this tag in its link settings: utm_source=facebook&utm_medium=paid&utm_campaign=HIP-B03&utm_term={{adset.id}}&utm_content={{ad.id}}
5. The store order keeps the tag. The store's revenue report has filters for these tags (seen 11.11, read only), but they list no values yet.

Launch gate: before the batch goes live, one real click through the article to the checkout and one small real order must show the tag saved on the order. If the article's button drops the tag, money per brief cannot be read.

4. What gets filled after the test

• At launch: date, ad account, campaign, ad set and ad ids, budget and budget type, Facebook page, pictures, headline, targeting and placements, which skill and version wrote each native and its health verdict, your edits.
• At each read: spend, clicks, cost per click, landing page views, checkouts, buys (labelled pixel or incremental), verdict by your rules (still testing under $200 spend, 100 clicks or 48 hours; winner at 1.5x with 5 or more buys; signal; kill), and store orders and revenue by tag.
• Roll-ups: by teller, story type, length, jar hidden or not, donor type.
• Predicted against real: for Argon, the judges, the fresh checker and my ranking: did its top 3 beat its bottom 3?
• What changed between the brief and the ad, so a brief is not blamed for what the writer changed.

The skill changes only after 3 or more batches say the same thing and you say go. One batch gives leads, not rules.

5. Gemini Carbon High's check

Verdict: "Partly — brief lineage is clean, but unmatched photos, an unproven UTM-to-order link, and $200 sample noise can hide which brief really won."

Taken in:

• The launch gate above (a real test order must show the tag).
• One ad set per brief with its own budget, separate ads, no dynamic creative, all nine started the same day.
• More fields: landing page views, article-to-checkout clicks, the top ad's share of spend, buys on 7-day click only, share of spend on women 58 to 70.
• Do not drop briefs that are still "testing" when scoring the predictions; compare top 3 against bottom 3 across 3 or more batches.
• If all nine fail on a page with no sales history, the page or offer is the first suspect, not the briefs.

Not taken in, left to you:

• Carbon wants the same photo and headline on all nine ads, so only the text differs. But a native's picture belongs to its story. My proposal: every brief gets its pictures made the same way and in the same number, and the picture is recorded and read as its own variable.
• Carbon wants the $200 and 1.3x / 1.5x gates changed for this product. Your rule is that only you change thresholds.
• Carbon wants at most 2 hooks per brief. The writer skill makes 3.
• Carbon's platform facts are a model's advice. It marked several as unsure. The media buyer should confirm them on a live ad.

6. What I need from you

1. Is this the snapshot you had in mind? Anything to add before it is frozen?
2. Who launches, and will they use the code in the names and the tag in the link?
3. Which ad account? The hip account is disabled.
4. Pictures: each brief its own (my proposal), or one shared picture (Carbon's)?
5. Hooks: 3 per brief as separate ads, or 2?