St. Pierre Creative Engine Automation SOW

Created July 8, 2026. Source standard: St. Pierre Media Buying Standard.

Objective

Automate creative ideation, production, Meta distribution, activity logging, and performance monitoring.

Why it matters

Ship more hook packs with less manual work, then scale based on cost per qualified booked call.

You need to

Approve build order or revise scope - 5 min.

Status

Scope ready. Build not started.

Bottom line. Build one creative operating system where the hook owns the ad set, every creative format is tracked separately, every action writes to a ledger, and the dashboard shows what is winning by hook, format, ad, spend, lead quality, and booked-call cost.

Approve the build order needs you

★ Recommended default: build the ledger and dashboard first, then automate production, then automate distribution. Reply "build v1" to start with this order.

Order	Workstream	Reason	Output
1	Ledger + IDs	If activity is not captured first, automation creates invisible mess.	Every idea, render, upload, ad, budget move, and result has a durable row.
2	Dashboard	Akash wants to monitor, not inspect raw Meta tables.	Hook board, format board, pack status, spend, leads, booked calls, winners.
3	Ideation + production	Once the ledger exists, the engine can safely create work.	Hook packs with static, carousel, motion, and AI UGC briefs/assets.
4	Distribution	Publishing is the risky external write, so it comes after QA and tracking.	Paused Meta ad sets organized by hook, then controlled activation.

Lock the organizing rule needs you

Rule. One ad set equals one hook. Static images, carousels, motion graphics, and AI UGC are creative formats inside that hook. They are not separate hook tests by default.

Layer	Owns	Rule	Example
Campaign	Testing lane	Monthly ABO hook-pack campaign unless the media-buying standard changes.	`SP TEST \| 2026-07 \| ABO Hook Packs`
Ad set	Hook	One hook, one budget, one verdict clock.	`Pack 014 \| Hook: Stop buying shared leads \| Launch 2026-07-14`
Ad	Creative execution	One asset or carousel under the parent hook.	`p014-h01-static-03-notes-app`
Dashboard slice	Format	Compare format performance without splitting the hook by default.	`static`, `carousel`, `motion`, `ai_ugc`

Exception. Split a hook by format only after the hook already has enough spend to justify a cleaner read, or if Meta delivery clearly starves a format. The default remains hook-owned ad sets.

Scope of work no action

Track	In scope	Not in v1
Ideation	Generate hook packs from persona, angle, offer, market, proof, objections, and historical winners.	Fully autonomous brand strategy changes.
Production	Create briefs, manifests, render jobs, QA checks, and output assets across four format families.	Unlimited net-new template design without a template library.
Distribution	Create paused Meta campaigns/ad sets/ads, upload assets, attach tracking, and log IDs.	Uncontrolled active spend or new account/payment setup.
Ledger	Append-only operational ledger plus Activity Ledger rows for external writes.	Chat-only notes or hidden approval queues.
Dashboard	Monitor production throughput, deployment status, and performance by hook, format, ad, and spend.	Replacing Ads Manager as the full source of billing truth.

Format-specific production tracks no action

Static Images

Fastest testing format.
Uses template families like X post, notes app, text bubble, messenger, reminders, lock screen, napkin, Safari, proof screenshot, and direct offer card.
Default output: 8-20 PNGs per hook pack.
QA checks: readable mobile text, hook match, no broken layout, no unsupported claims.

Carousels

Used when the hook needs a sequence, proof stack, teardown, myth-vs-truth, or "how it works" flow.
Default output: 5-8 cards plus primary text and headline.
Tracks each card set as one carousel creative, with card-level source files in the manifest.
QA checks: card order, first-card hook, last-card CTA, no tiny copy.

Motion Graphics

Template-driven movement over the same hook: animated proof, numbers, callout overlays, whiteboard, Loom-style explainer, or kinetic text.
Default output: 6-20 second MP4 in feed and vertical versions when practical.
Runs after static/carousel hooks show promise, unless a motion-specific idea is clearly cheap to render.
QA checks: first 2 seconds, captions, safe-area, file size, thumbnail, CTA.

AI UGC

Scripted founder-style, operator-style, skeptic-style, or explainer-style clips generated from approved claims.
Default output: script, shot list, avatar/voice instructions, generated video, caption file, and policy check.
Never fabricates a customer, result, territory, or network endorsement beyond approved proof.
QA checks: disclosure risk, claim support, voice fit, pacing, face/hand artifacts, landing-page match.

Ideation engine no action

Input	How it is used	Ledger event
Offer	$1 start, 10+ inspections in 7 days, week-to-week, pay per qualified booked inspection.	`offer_snapshot_used`
Persona	Founder, skeptical owner, operator peer, proof narrator, systems explainer.	`persona_selected`
Angle	Rented demand, missed inspections, agency distrust, territory lockout, sales-team capacity.	`angle_selected`
Hook	The exact test identity. It becomes the ad set name and pack ID.	`hook_pack_created`
Format mix	Static, carousel, motion, AI UGC assignments for the same hook.	`format_mix_selected`
Historical results	Winners get expanded; losers inform blocked angles and new counter-hooks.	`performance_used_for_ideation`

Required output per hook pack

pack_id, hook_id, hook_text, persona, angle, offer_snapshot_id, and launch_intent.
10-20 creative concepts grouped by format.
Production manifest with filenames, template IDs, copy, CTA, landing URL, UTM tags, and QA state.
Verdict criteria before launch: target CPL, target CPQBC, kill cap, promotion trigger.

Production workflow no action

Create the hook pack. Assign pack ID and write the hook, persona, angle, offer, format mix, and target metric.
Generate concepts. Produce format-specific concepts and primary text variants, but keep the parent hook fixed.
Render assets. Run static templates, carousel card builder, motion renderer, or AI UGC generation jobs from one manifest.
Run QA. Check file dimensions, text fit, readability, policy-sensitive claims, proof support, CTA consistency, and landing route.
Freeze the pack. Once approved for distribution, the pack becomes a record. Fix errors only; do not rewrite strategy mid-test.
Store outputs. Keep assets under the project creative batch folder and write a manifest row for every file.

State	Meaning	Allowed next move
`idea`	Hook exists, no assets yet.	Generate concepts.
`briefed`	Concepts and format assignments exist.	Render assets.
`rendered`	Files exist locally.	QA.
`qa_passed`	Ready for paused upload.	Create Meta objects.
`staged_paused`	Meta ads exist but are not spending.	Activate or hold.
`active`	Spend is on.	Monitor ladder.
`graduated`	Winner duplicated into scale pool.	Monitor scaling.
`held`	Paused for poor read or QA issue.	Archive or learn from it.

Distribution automation no action

Spend guard. The system may create paused Meta objects automatically when logged and reversible. Turning on spend requires an explicit activation control, budget cap, and rollback path.

Step	Automation	Guardrail	Ledger event
1	Create or find monthly testing campaign.	Campaign starts paused unless explicitly in an active controlled test lane.	`campaign_found` or `campaign_created`
2	Create one ad set for the hook pack.	Name includes pack ID, hook, mix, launch date. Daily budget follows the standard.	`adset_created`
3	Upload each asset to Meta.	Store image hash or video ID against the asset row.	`asset_uploaded`
4	Create one ad per creative execution.	Ad name includes pack, hook, format, sequence, and idea slug.	`ad_created`
5	Verify campaign, ad set, ad, creative, link, status, and preview fields.	Block activation if any ad is not paused, missing tracking, or using wrong hook.	`distribution_verified`
6	Activate only through a controlled command.	Activation logs before/after status and budget. Budget moves follow ladder.	`status_changed` and `budget_changed`

Tracking requirements

utm_campaign = campaign lane and month.
utm_content = asset/ad slug.
utm_term = format.
sp_pack_id = hook pack ID.
sp_hook_id = durable hook ID.
sp_format = static, carousel, motion, or ai_ugc.

Activity ledger design needs you

★ Recommended default: use a dedicated creative_activity_ledger for every creative-system event, and continue using the existing Activity Ledger for external writes such as Meta uploads, status changes, Drive uploads, and dashboard deploys.

Field	Purpose	Example
`event_id`	Unique row ID.	`evt_20260708_001234`
`ts`	When it happened.	`2026-07-08T14:02:11Z`
`actor`	Who or what did it.	`codex`, `akash`, `renderer`, `meta_sync`
`run_id`	Groups one automation run.	`creative-engine-20260708-01`
`pack_id`	Hook-pack link.	`p014`
`hook_id`	Hook link.	`h-stop-buying-leads`
`format`	Creative family.	`static`, `carousel`, `motion`, `ai_ugc`
`asset_id`	Specific creative file or concept.	`p014-h01-static-03`
`surface`	Where the action happened.	`brain`, `filesystem`, `meta_ads`, `drive`, `dashboard`
`action`	What happened.	`asset_rendered`, `ad_created`, `winner_graduated`
`before_image`	Restorable before-state when relevant.	Previous budget/status/template path.
`after_image`	Result state.	New Meta IDs, file hashes, status, metrics snapshot.
`reverse_how`	Undo instructions.	Pause ad set, restore prior budget, archive pack.
`verified_at`	Proof timestamp.	When the API read-back matched intended state.

Required event types

hook_pack_created, concept_generated, asset_rendered, qa_failed, qa_passed.
asset_uploaded, campaign_created, adset_created, creative_created, ad_created, distribution_verified.
status_changed, budget_changed, performance_snapshot, winner_graduated, pack_held.
dashboard_refreshed, data_gap_detected, rollback_executed.

Performance dashboard design needs you

★ Recommended default: build a local project dashboard first, then deploy it as a persistent dashboard only after the data joins are correct.

Panel	Shows	Primary question
Command strip	Active spend, leads, qualified actions, booked calls, CPQBC, best hook, worst leak.	What needs attention today?
Hook board	Each hook pack by status, age, spend, CPL, CPQBC, first qualified action, verdict clock.	Which hook is winning?
Format board	Static vs carousel vs motion vs AI UGC by spend, CTR, CPL, qualified rate, booked-call rate.	Which format deserves more production?
Ad leaderboard	Top and bottom ads, filtered by hook and format.	Which exact creative should graduate or die?
Production ledger	Ideas created, assets rendered, QA fails, uploads, deploys, cycle time.	Is the machine shipping enough?
Budget ladder	Current budget tier, promotion trigger, kill cap, next allowed move.	What can be scaled without guessing?
Data health	Missing Meta IDs, missing UTMs, stale metrics, unmatched Typeform/GHL conversions.	Can Akash trust the numbers?

Minimum metric set

Meta: spend, impressions, reach, frequency, CPM, link clicks, CTR, CPC, landing-page views if available, leads/conversions.
Funnel: applications, qualified applications, $1 starts, booked inspections/calls, show rate if available.
Creative: hook, angle, persona, format, template, asset ID, ad ID, creative ID, launch date, status.
Decision metrics: CPL, cost per qualified action, cost per qualified booked call, lead-to-booked rate, spend to first qualified action.

Data model no action

Table / file	Grain	Key fields
`creative_hook_packs`	One hook pack.	`pack_id`, `hook_id`, `persona`, `angle`, `status`, `launch_date`, `target_cpqbc`.
`creative_assets`	One rendered asset or carousel.	`asset_id`, `pack_id`, `format`, `template_id`, `file_path`, `sha256`, `qa_status`.
`creative_distribution_objects`	One Meta object mapping.	`asset_id`, `campaign_id`, `adset_id`, `creative_id`, `ad_id`, `status`.
`creative_activity_ledger`	One activity event.	`event_id`, `run_id`, `action`, `before_image`, `after_image`, `reverse_how`.
`creative_performance_snapshots`	One metric snapshot per object per day or pull.	`object_id`, `level`, `date`, `spend`, `leads`, `qualified_booked_calls`, `cpqbc`.
`manifest.csv`	One production batch.	Portable backup of rendered files, copy, tracking, and Meta IDs.

Build plan no action

Phase	Build	Acceptance test
0. Inventory	Map current creative folders, scripts, Meta naming, landing routes, and media-buying standard.	One manifest can represent the 64 uploaded July 8 ads without hand edits.
1. Ledger	Create creative ledger writer and schemas. Wrap every automation step.	A dry run writes idea, render, QA, upload-intent, and dashboard events with run ID.
2. Dashboard	Create local dashboard from ledger + Meta read-back + funnel metrics.	Dashboard shows hook, format, ad, status, spend, CPL, CPQBC, and data-health gaps.
3. Ideation	Generate hook packs from offer, personas, angles, winners, and blocked claims.	One command creates a frozen hook-pack brief and production manifest.
4. Production	Render static, carousel, motion, and AI UGC outputs from the manifest.	At least one hook pack produces valid assets in all four format families.
5. Distribution	Create paused Meta objects by hook, attach UTMs, read back status, write mappings.	One hook pack stages paused ads in Meta with no mixed-hook ad set.
6. Monitoring	Refresh performance snapshots and update status recommendations.	Dashboard flags graduate/hold/kill candidates using the media-buying ladder.

Quality gates no action

Gate	Blocks when	Fix path
Hook integrity	An ad set contains multiple hooks or the asset hook does not match the ad set hook.	Move or rename asset before upload.
Format tagging	Format is missing or ambiguous.	Assign one of `static`, `carousel`, `motion`, `ai_ugc`.
Claims	Asset makes unsupported guarantees, fake proof, or fake customer claims.	Rewrite against approved proof bank.
Mobile readability	Text is clipped, tiny, or visually broken.	Re-render template.
Tracking	Missing pack, hook, format, or ad-level tracking tags.	Regenerate URL before creative creation.
External write ledger	Meta, Drive, deploy, or other external write lacks before/after and rollback.	Stop and write the ledger row before continuing.
Activation	Spend would turn on without a budget cap and rollback path.	Keep paused until activation command is explicit.

Automation commands reference only

Command	Effect	Default state
`create_hook_pack`	Generate hook pack brief and concept manifest.	`idea`
`render_pack`	Produce static, carousel, motion, and AI UGC assets from manifest.	`rendered`
`qa_pack`	Run file, visual, copy, policy, and tracking checks.	`qa_passed` or `qa_failed`
`stage_pack_to_meta`	Create paused Meta ad set and ads by hook.	`staged_paused`
`activate_pack`	Turn on spend with budget and rollback logged.	`active`
`refresh_performance`	Pull Meta and funnel metrics, write snapshots, refresh dashboard.	Dashboard current.
`graduate_winners`	Duplicate winning ads into the winner pool.	`graduated`

Risks and safeguards reference only

Risk	Safeguard
Automation makes a lot of assets but no learning.	Dashboard ranks by hook and format against CPQBC, not output volume.
Formats get mixed into unclear tests.	Ad set stays hook-owned; format is an ad-level dimension.
Meta delivery starves one format.	Dashboard shows spend share by format; split only when evidence justifies it.
AI UGC creates trust or policy risk.	Approved proof bank, claim QA, no fake customers, and manual review before activation.
Untracked external writes.	All external writes also call the canonical Activity Ledger with rollback instructions.
Dashboard numbers drift from real business outcomes.	Join Meta spend to Typeform/GHL/booked-call outcomes; show data-health gaps instead of hiding them.

Definition of done reference only

Akash can see every creative-system action in a ledger without reading chat history.
Akash can open one dashboard and know which hooks, formats, and ads are performing.
A new hook pack can be generated, rendered, QAed, and staged to Meta as paused ads with one run ID.
Meta ad sets are organized by hook, never by mixed creative themes.
Static images, carousels, motion graphics, and AI UGC are tracked as distinct formats.
Every external write has before/after state, verification, and rollback instructions.
Spend activation is controlled, logged, and reversible by pausing campaign/ad set/ad.