Scripting Skill · Reference Dossier

format-referenceThe format taxonomy and scripting mechanics library. Tells every scripting agent which formats exist, how each one is structured, and why format diversity matters for Meta's Andromeda algorithm.

Owner: Reference document — loaded as context by ad-scripter-write and Format-Hermes; no independent runner · Asel · 2026-06-11
What it does

This is a reference document, not an executable skill. It has two layers. The first is the format taxonomy — the full set of ad video formats available in the system, their production profiles, their Meta Andromeda Entity ID signals, and the visual characteristics that make each one distinct from the others. The second layer is scripting mechanics — the detailed section-by-section structure for the three non-standard format modes (Skillshow Notebook, 2-Person, and FAQ Phone Scroll). The ad-scripter-write skill loads this reference as context whenever a script is requested in a non-standard format. Format-Hermes reads the taxonomy layer when making format recommendations for a batch. The production-brief skill reads it to validate format slugs. Because it is a reference, it gets a currency check on each sweep — not golden-case evals.

Format count — PRD target vs repo

The system's product requirements document (PRD) targets approximately 22 formats across video, static, carousel, and user-generated content families. The repository currently implements 14 video formats in style-system.md (the taxonomy source) and 3 scripting modes in format-reference.md. Static, carousel, and formats such as podtalk and skit do not yet have repository specifications. This dossier documents what exists today. The delta to ~22 is flagged as a build gap for operator review.

01 Model routing — how this reference is read

This file does not run a model. It is loaded as context by the skills that do.

ConsumerWhat it readsHow it uses it
Format-Hermes (in-batch function)The 14 format profiles — EC Prior (Entity Capture Prior) values, Entity ID signal taxonomy, format × arc matching table, Entity ID separation matrixGenerates format mix recommendations for a batch. Selects formats that spread the batch across at least 4–5 distinct Andromeda entity classes. Emits recommended_formats[] on the Construct Brief that ad-ideate then weights by.
ad-scripter-write (operator command)The scripting mechanics section — section-by-section structure for SKILLSHOW_NOTEBOOK, 2-PERSON-FRIENDS, 2-PERSON-RECOGNITION, FAQ_PHONE_SCROLLWhen Format Mode on the concept card is not SOLO, the scripter loads the relevant mechanics block and produces a format-compliant script (correct section count, word count, stage direction conventions, proof placement rules).
ad-ideateFormat profiles — EC Prior, arc fit, pathway availability from production-pathways.mdSelects format for each concept card from the Format-Hermes recommended mix, filtered by non-BLOCKED pathways.
ad-ideatePathway status per formatRejects any concept whose format is BLOCKED across all production pathways. Flags IN-TEST pathways with a validation caveat.
ad-production-briefrc-ai-ugc slugs and format profile metadataValidates the format slug from the concept card before generating the filming brief or AI user-generated content handoff package.

Deterministic-first note: Format selection logic — counting formats per entity class, checking pathway availability, flagging blocked formats — is plain code. No model runs those checks. The model only reasons over the format profiles when synthesising a format recommendation narrative.

02 The spec — format taxonomy

The 14 formats in the system today, how they are built, and why format diversity matters.

Why format diversity matters — the Andromeda Entity ID mechanism

Meta's Andromeda algorithm assigns each creative an Entity ID based on visual signals it reads from the ad — environment, camera geometry, subject count, and format mechanics. Two ads with different Entity IDs compete for different audience sub-segments and scale independently without cannibalising each other. Running ten scripts in the same visual format gives the algorithm ten ads that look structurally identical, regardless of the words. Entity ID separation falls, fatigue windows compress, and the ceiling on reach shrinks. Format diversity is not a stylistic preference — it is the mechanism by which a winning message multiplies across new Entity IDs before it fatigues in the current one.

Format families

The 14 formats group into four entity-class families based on the primary Andromeda signal they generate:

FamilyPrimary signalFormats
Solo DTC (direct to camera) — talking headSubject, camera geometry, environmentSelfie White Wall · Outdoor Walking Selfie · Seated Interior · Mirror Selfie · Street Interview · Park Bench
2-PersonSubject count (1→2 is a primary Andromeda signal)Pub Chat (Friends) · Stage Interview
Format-mechanic formatsStructural visual mechanic — text overlays, overhead angle, split-screenFAQ Phone Scroll · Skillshow Notebook · This-or-That · Green Screen
Multi-scene / lifestyleMultiple environment cuts within one adMulti-Scene Founder · Day in the Life

Static, carousel, and formats such as podtalk, skit, and this-or-that (static variant) are PRD targets not yet specified in the repository. See the warning callout at the top of this dossier.

The 14 format profiles

1 · Selfie White Wall

rc-ai-ugc slug: selfie-white-wall · Scripting mode: SOLO
FieldValue
Production methodPhone-held selfie, face forward, clean indoor background — white wall or minimal décor.
Entity ID signalsFace-forward, single subject, minimal environment detail, static or slight movement.
EC Prior vs baselineLOW — the most common format in the avatar-match universe. Strong baseline for a first batch; low differentiation for iteration rounds.
Best arcsAny arc — this is the universal fallback. Never more than 3 scripts in this format per batch.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (VALIDATED) · EXTERNAL-ACTOR (VALIDATED)

2 · Outdoor Walking Selfie

rc-ai-ugc slug: outdoor-walking-selfie · Scripting mode: SOLO
FieldValue
Production methodPhone-held selfie while walking outdoors. Dynamic environment, ambient light, changing background.
Entity ID signalsOutdoor environment, movement (camera and subject), ambient audio, changing background.
EC Prior vs selfie-white-wallMEDIUM — outdoor + movement = two-dimension separation from white wall. Different entity class.
Best arcsDirect Callout, income/lifestyle angles, energy-forward hooks.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (VALIDATED) · EXTERNAL-ACTOR (VALIDATED)

3 · Seated Interior

rc-ai-ugc slug: seated-interior · Scripting mode: SOLO
FieldValue
Production methodSubject seated indoors — desk, home office, studio, sofa. Face to camera. Static or minimal movement.
Entity ID signalsSpecific interior environment (contextual props, décor), seated posture, controlled lighting.
EC Prior vs selfie-white-wallMEDIUM — interior context adds environment signals. Desk/office signals authority differently from a white wall.
Best arcsAuthority/credentials angles, Gatekeeping Reveal arc, longer-form explanations.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (VALIDATED) · EXTERNAL-ACTOR (VALIDATED)

4 · Mirror Selfie

rc-ai-ugc slug: mirror-selfie · Scripting mode: SOLO
FieldValue
Production methodSubject filming their reflection. Phone visible in-frame. Bathroom, dressing room, or gym context.
Entity ID signalsMirror reflection (subject appears at angle), phone in frame, specific room type.
EC Prior vs selfie-white-wallMEDIUM-HIGH — reflection mechanic creates unique visual composition. Phone-in-frame is a structural feature Andromeda likely classifies distinctly.
Best arcsPersonal transformation angles, identity callout hooks, health and body verticals.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (VALIDATED) · EXTERNAL-ACTOR (VALIDATED)

5 · Street Interview

rc-ai-ugc slug: street-interview · Scripting mode: SOLO
FieldValue
Production methodSubject on a busy street or urban environment. Blurred pedestrians, traffic, ambient noise behind them.
Entity ID signalsOutdoor urban environment, environmental depth (blurred movement behind subject), ambient sound texture.
EC Prior vs white wallHIGH — outdoor urban creates clear multi-dimension separation: environment + ambient movement + lighting conditions.
Best arcsSocial proof delivery, income-first hooks, testimonial-style narration. Creates "overheard on the street" authenticity.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (UNTESTED) · EXTERNAL-ACTOR (BLOCKED — public filming complications)

6 · Park Bench

rc-ai-ugc slug: park-bench · Scripting mode: SOLO
FieldValue
Production methodSubject seated on a park bench or natural outdoor setting. Natural light. Relaxed posture.
Entity ID signalsNatural outdoor environment, seated posture (different from walking), specific bench/nature context.
EC Prior vs street-interviewMEDIUM — shares outdoor environment but seated posture and natural vs urban context create moderate additional separation.
Best arcsConversational/confessional angles, Overwhelm-to-System arc with a calm resolution register.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (VALIDATED) · EXTERNAL-ACTOR (VALIDATED)

7 · Pub Chat (2-Person Friends)

rc-ai-ugc slug: pub-chat · Scripting mode: 2-PERSON-FRIENDS
FieldValue
Production methodTwo subjects in a social venue (pub, bar, restaurant). Conversational framing. Expert and interviewer/friend.
Entity ID signalsTwo subjects (a primary Andromeda signal), social environment, conversational camera framing (wider to show both), ambient venue context.
EC Prior vs all solo formatsVERY HIGH — subject count shift from 1 to 2 is a primary Andromeda signal. Every 2-person format generates a fundamentally different entity class from solo formats.
Best arcsTransformation narratives told as conversation. Interviewer reactions create emotional proof without explicit testimonials.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (BLOCKED) · EXTERNAL-ACTOR (BLOCKED) — AI user-generated content locked in; live filming shelved pending operator approval.
Scripting mechanicsSee format-reference.md 2-PERSON section: 7-section formula adapted to dialogue, interviewer voice rules (25–35 words total), CTA trigger mechanics (interviewer triggers the sell, never the expert).

8 · Stage Interview (2-Person)

rc-ai-ugc slug: stage-interview · Scripting mode: 2-PERSON-RECOGNITION
FieldValue
Production methodTwo subjects in a professional or event context. Microphones, event backdrop, slightly formal framing.
Entity ID signalsTwo subjects, formal/professional environment, microphone props, event backdrop texture.
EC Prior vs pub-chatMEDIUM — shares 2-person signal; formal environment adds separation from social-venue framing.
Best arcsCredibility/authority angles, expert-as-guest positioning.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (UNTESTED) · EXTERNAL-ACTOR (BLOCKED — live filming needs a real event)

9 · Skillshow Notebook

rc-ai-ugc slug: skillshow · Scripting mode: SKILLSHOW_NOTEBOOK
FieldValue
Production methodOverhead camera. Subject's hands and an iPad on a flat surface. Subject swipes through slides. Face-to-camera for hook and call-to-action bookends only.
Entity ID signalsOverhead camera angle (unique in the format set), hands-and-device visual, progressive reveal mechanic (slides one by one), no or minimal face.
EC Prior vs all talking-head formatsVERY HIGH — camera geometry is the strongest single signal. Overhead is visually unlike every other format. Guaranteed Entity ID separation from any talking-head variant.
Best arcsMechanism/credentials angles, Overwhelm-to-System arc with visual teaching structure.
PathwaysAI-UGC (BLOCKED — not viable) · INTERNAL-TEAM (IN-TEST) · EXTERNAL-ACTOR (IN-TEST)
Scripting mechanics6 sections (not 7), 90–120 seconds, 250–350 words. First swipe within 5 seconds. Proof woven into teaching steps, not held for a separate Social Proof Bridge. Step 6B: Nano Banana Pro (NBP Pro) slideshow prompts generated after operator approves the script.

10 · FAQ Phone Scroll

rc-ai-ugc slug: faq-phone-scroll · Scripting mode: FAQ_PHONE_SCROLL
FieldValue
Production methodSubject face-to-camera (selfie or seated). Text overlay questions appear on screen in post-production. Expert answers conversationally, as if responding to a comment or DM.
Entity ID signalsText overlay mechanics (structural visual element on top of talking head), question-answer pacing creates a visual rhythm distinct from continuous narration.
EC Prior vs selfie-white-wallMEDIUM — text overlays add a structural visual layer Andromeda likely classifies distinctly. Same face-forward geometry limits separation.
Best arcsObjection-handling angles, warm retargeting, high-volume rapid-fire content. Best for Solution-Aware or Product-Aware traffic.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (VALIDATED) · EXTERNAL-ACTOR (VALIDATED)
Scripting mechanics5–6 sections (1 hook question + 3–4 Q&A pairs + call to action), 45–90 seconds, ~150–220 words. Expert never reads the question aloud. Each answer 15–40 words. At least 2 answers include a proof element. Expert introduces themselves in Q2, not before.

11 · Green Screen

rc-ai-ugc slug: green-screen · Scripting mode: SOLO
FieldValue
Production methodSubject in front of a digital background replaced via chroma key. Background can be any image, graphic, or video.
Entity ID signalsBackground type (digital/graphic vs real environment), potential for on-screen data visualisations or supporting graphics behind subject.
EC PriorMEDIUM — depends heavily on background content. Plain colour = low separation; chart or dramatic visual = higher separation.
Best arcsData-heavy angles, contrarian/myth-bust hooks with on-screen supporting visuals.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (BLOCKED — requires physical green screen rig) · EXTERNAL-ACTOR (UNTESTED)

12 · Multi-Scene Founder

rc-ai-ugc slug: multi-scene-founder · Scripting mode: SOLO
FieldValue
Production methodMultiple cuts across different locations or contexts. Documentary-style. Voiceover drives the narrative while B-roll or multiple scenes play.
Entity ID signalsMultiple visual contexts (cuts between environments), voiceover-driven structure, documentary pacing.
EC PriorHIGH — multi-scene creates multiple environment signals within one ad. Andromeda reads scene transitions as structural complexity distinct from any single-scene format.
Best arcsOrigin story arcs, Gatekeeping Reveal, longer transformation narratives.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (BLOCKED) · EXTERNAL-ACTOR (BLOCKED) — AI user-generated content via multi-clip stitch with consistent identity anchor.

13 · Day in the Life

rc-ai-ugc slug: day-in-the-life · Scripting mode: SOLO
FieldValue
Production methodMultiple brief clips across different times and contexts within a single day. Aspirational lifestyle framing.
Entity ID signalsTime-of-day variation (morning/afternoon/evening), lifestyle location variety (home, coffee shop, workout, work), aspirational aesthetic.
EC PriorHIGH — lifestyle variety across scenes creates multi-environment signals. Lighter narration and stronger lifestyle/aspiration register than Multi-Scene Founder.
Best arcsIncome-First Curiosity arc, lifestyle-as-proof angles, aspirational identity hooks. Most effective for Unaware to Problem-Aware traffic.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (BLOCKED) · EXTERNAL-ACTOR (BLOCKED) — previously shelved; AI user-generated content makes viable via multi-clip.

14 · This-or-That

rc-ai-ugc slug: this-or-that · Scripting mode: SOLO
FieldValue
Production methodSplit-screen or rapid alternation between two contrasting options, scenes, or states. Visual binary comparison.
Entity ID signalsSplit-screen or rapid-cut mechanics, visual binary structure, comparison framing.
EC PriorMEDIUM-HIGH — split-screen or rapid alternation is a distinct structural mechanic.
Best arcsNegative Elimination arc ("not this, not that — but this"), myth-bust hooks, before/after framing.
PathwaysAI-UGC (IN-TEST) · INTERNAL-TEAM (IN-TEST) · EXTERNAL-ACTOR (UNTESTED)

Format × Pathway — quick-reference matrix

The rule for ideation: a concept may emit a format if at least one pathway column is not BLOCKED. Dual-pathway formats defer routing to scripting-approval time.

FormatAI-UGCInternal TeamExternal Actor
Selfie White WallIN-TESTVALIDATEDVALIDATED
Outdoor Walking SelfieIN-TESTVALIDATEDVALIDATED
Seated InteriorIN-TESTVALIDATEDVALIDATED
Mirror SelfieIN-TESTVALIDATEDVALIDATED
Street InterviewIN-TESTUNTESTEDBLOCKED
Park BenchIN-TESTVALIDATEDVALIDATED
Pub Chat (2-Person Friends)IN-TESTBLOCKEDBLOCKED
Stage InterviewIN-TESTUNTESTEDBLOCKED
Skillshow NotebookBLOCKEDIN-TESTIN-TEST
FAQ Phone ScrollIN-TESTVALIDATEDVALIDATED
Green ScreenIN-TESTBLOCKEDUNTESTED
Multi-Scene FounderIN-TESTBLOCKEDBLOCKED
Day in the LifeIN-TESTBLOCKEDBLOCKED
This-or-ThatIN-TESTIN-TESTUNTESTED

Entity ID separation — pairing guide

Use format combinations with HIGH or VERY HIGH separation to maximise Andromeda entity diversity across a batch. A pioneer batch of 10 scripts should aim for at least 4–5 distinct entity classes.

Format AFormat BSeparationReason
Any solo DTCAny 2-person formatVERY HIGHSubject count shift (1→2) is a primary Andromeda signal.
Any solo DTCSkillshow NotebookVERY HIGHCamera geometry (overhead vs face-forward) — the strongest single dimension.
Selfie White WallStreet InterviewHIGHIndoor minimal vs outdoor urban with ambient movement.
Outdoor Walking SelfieSeated InteriorMEDIUMMovement + outdoor vs static + interior — two dimensions but both solo.
Selfie White WallFAQ Phone ScrollMEDIUMText overlay mechanic adds a structural layer, but face-forward geometry is shared.
Selfie White WallOutdoor Walking SelfieLOW–MEDIUMOnly one dimension of difference (indoor vs outdoor). Too similar for entity separation at scale.

03 Live format profiles representative

Three representative profiles showing how a format card looks when Format-Hermes emits it to the Construct Brief.

REPRESENTATIVE EXAMPLE — Format profile output as emitted by Format-Hermes. Not a real eval artifact — constructed from style-system.md and format-reference.md source data. Badge: b-rep.
--- FORMAT PROFILE: Selfie White Wall ---

slug: selfie-white-wall
scripting_mode: SOLO
family: Solo DTC (direct to camera)
ec_prior: LOW
entity_class: 1 (face-forward, single subject, minimal environment)

entity_id_signals:
  - Face-forward, single subject
  - Minimal environment detail (white or near-white background)
  - Static or slight movement

pathway_availability:
  ai_ugc: IN-TEST
  internal_team: VALIDATED
  external_actor: VALIDATED

best_arcs:
  - Any arc — this is the universal fallback
  - Direct Callout (strongest match)

batch_cap: 3 (never more than 3 in a batch of 10)

scripting_notes: >
  Standard SOLO 7-section formula applies. No format-specific
  mechanics. Load ad-scripter-reference for word counts and
  section targets.

--- FORMAT PROFILE: Pub Chat (2-Person Friends) ---

slug: pub-chat
scripting_mode: 2-PERSON-FRIENDS
family: 2-Person
ec_prior: VERY HIGH
entity_class: 4 (2-person, social environment)

entity_id_signals:
  - Two subjects in frame (primary Andromeda signal)
  - Social venue environment (pub, bar, restaurant)
  - Conversational camera framing (wider than solo)
  - Ambient venue context

pathway_availability:
  ai_ugc: IN-TEST        # CURRENT primary path — AI-UGC locked in
  internal_team: BLOCKED # Live filming shelved pending operator review
  external_actor: BLOCKED

best_arcs:
  - Transformation narratives told as conversation
  - Social proof via interviewer reactions (implicit, not stated)

scripting_notes: >
  Load format-reference.md — 2-PERSON-FRIENDS section.
  7-section formula adapted to dialogue. Interviewer word count:
  25–35 total. CTA trigger: interviewer references their person
  (girlfriend, mum, friend) — expert never initiates the sell.
  Location options: cafe, pub, park bench, restaurant — same
  script serves 2–3 locations, note in brief.

--- FORMAT PROFILE: FAQ Phone Scroll ---

slug: faq-phone-scroll
scripting_mode: FAQ_PHONE_SCROLL
family: Format-mechanic
ec_prior: MEDIUM
entity_class: 3 (text overlay, Q&A structure)

entity_id_signals:
  - Text overlay questions (structural visual element)
  - Q&A pacing rhythm (distinct from continuous narration)
  - Face-forward geometry (shared with solo DTC — limits separation)

pathway_availability:
  ai_ugc: IN-TEST
  internal_team: VALIDATED
  external_actor: VALIDATED

best_arcs:
  - Objection-handling (Solution-Aware or Product-Aware traffic)
  - Warm retargeting

scripting_notes: >
  Load format-reference.md — FAQ_PHONE_SCROLL section.
  Structure: 1 hook question + 3–4 Q&A pairs + CTA.
  Duration: 45–90 seconds, ~150–220 words.
  Expert NEVER reads the question aloud — text overlay only.
  Each answer: 15–40 words. At least 2 answers include
  a proof element (pioneer name + concrete outcome).
  Expert introduces themselves in Q2, not before Q1.
  Q1 IS the hook.
How to read it: Each block above is one format profile as Format-Hermes would emit it to the Construct Brief. The ec_prior field drives the mix — a pioneer batch needs at least two formats rated HIGH or VERY HIGH to spread across distinct Andromeda entity classes. The pathway_availability block tells ad-ideate which pathways are live — any concept tagged to a format where all pathways are BLOCKED is rejected before scripting. The scripting_notes tell ad-scripter-write which mechanics section to load and any non-SOLO constraints that apply. SOLO scripts load standard 7-section structure from ad-scripter-reference; non-SOLO scripts load the specific section in format-reference.md.

04 Glossary

TermIn full / what it means
DTCDirect to Camera — A video style where the presenter speaks directly into the camera lens, creating a one-on-one feel. The dominant style for most formats in this system. Abbreviated DTC throughout the creative system; always spelled out here and on first use in any human-facing document.
UGCUser-Generated Content — An ad style that mimics organic, creator-style content — informal, handheld, less polished. In this system, AI user-generated content refers to creatives produced via the ugc-factory pipeline (Nano Banana Pro → Enhancor → Seedance 2.0) rather than filmed by a real person.
AI-UGCAI User-Generated Content (production pathway) — The production route that uses the ugc-factory extension (Nano Banana Pro identity lock → Enhancor → Seedance 2.0) to generate video without a real actor. One of three production pathways alongside INTERNAL-TEAM (Bonnie or internal operator films) and EXTERNAL-ACTOR (contracted actor films).
Entity IDAndromeda Entity ID — A unique identifier Meta's Andromeda algorithm assigns to each creative based on the visual signals it reads from the ad. Two creatives with the same Entity ID compete for the same audience sub-segment. Two creatives with different Entity IDs compete independently and can scale without cannibalising each other. Format diversity is the primary mechanism for generating distinct Entity IDs.
AndromedaAndromeda (Meta's ad delivery algorithm) — Meta's system for deciding which creative to show to which person. It reads visual signals from the ad — environment, camera geometry, subject count, format mechanics — and uses them to assign Entity IDs and route ads to audience sub-segments. Understanding Andromeda's signal taxonomy is the reason this format taxonomy exists.
EC PriorEntity Capture Prior — A first-principles estimate of how much Andromeda Entity ID separation a given format produces relative to what is already in the footage library. Scored per format in style-system.md. Values are currently theoretical, not measured. Used by ad-creative-strategy to score iteration paths. Format combinations rated HIGH or VERY HIGH are prioritised in batch planning.
Format familiesFormat families (entity class groupings) — The four groupings in this system: Solo DTC (direct to camera) talking head (Selfie White Wall, Outdoor Walking, Seated Interior, Mirror Selfie, Street Interview, Park Bench); 2-Person (Pub Chat, Stage Interview); Format-mechanic formats (FAQ Phone Scroll, Skillshow Notebook, This-or-That, Green Screen); Multi-scene/lifestyle (Multi-Scene Founder, Day in the Life).
rc-ai-ugc slugrc-ai-ugc slug (format identifier string) — The lowercase-hyphenated identifier used to refer to a format in production pipeline files, motion profile JSON files, and the production-brief skill. Example: faq-phone-scroll. Distinct from Format Mode names (all-caps, underscore-separated) used in ad-scripter-write — the relationship is many-to-one: one scripting Format Mode (e.g. 2-PERSON-FRIENDS) can be filmed across multiple rc-ai-ugc slugs (e.g. pub-chat, park-bench).
Format ModeFormat Mode (scripting mode identifier) — The all-caps, underscore-separated label used in ad-scripter-write to select which scripting structure to apply. Five modes: SOLO (standard 7-section formula), SKILLSHOW_NOTEBOOK, 2-PERSON-FRIENDS, 2-PERSON-RECOGNITION, FAQ_PHONE_SCROLL. One mode can map to multiple rc-ai-ugc slugs.
SOLOSOLO (scripting mode) — The default scripting mode. One expert, speaking directly to camera, 7-section formula, 190–250 words. Applies to Selfie White Wall, Outdoor Walking Selfie, Seated Interior, Mirror Selfie, Street Interview, Park Bench, Green Screen, Multi-Scene Founder, Day in the Life, and This-or-That.
PodtalkPodtalk (PRD target format — not yet in repo) — A podcast-style format where the expert sits at a desk or table with a visible microphone, as if being recorded for a podcast or interview show. The closest repo analog is Stage Interview. Podtalk is named in the system PRD (product requirements document) as part of the ~22-format target set but has no rc-ai-ugc slug, scripting mechanics, or pathway specification in the current repository.
Voice BookVoice Book (Layer 2 expert data) — The structured expert voice artifact built by ad-voice-profile. Contains signature language patterns, tone fingerprints, forbidden words, and proof delivery style for a specific expert. Ad-scripter-write loads the Voice Book at every invocation and halts if it is empty for the named expert. Previously called "Layer 3 calibration" — that name is deprecated; do not use it.
NBP ProNano Banana Pro — The image generation tool used to produce individual slide images for Skillshow Notebook format. Prompts are generated in Step 6B of the Skillshow scripting process, after the operator approves the script. Slides are loaded onto an iPad as a swipeable slideshow.
CPBCCost Per Booked Call — How much money was spent on Meta ads to get one person to book a sales call with the expert. The primary success metric for top-of-funnel (TOF) cold traffic. Target threshold per expert lives in expert_config.target_cpbc in Supabase.
TOFTop of Funnel — Cold traffic — people who have never seen the expert's content before. Most of this system's formats are designed for top-of-funnel audiences. Middle of funnel (MOF) warm retargeting uses a different format and hook mix.
Social Proof BridgeSocial Proof Bridge (non-negotiable script element) — The section in the standard 7-section ad formula where proof entries (pioneer name + concrete outcome) are delivered. Non-negotiable in all SOLO format scripts. In Skillshow Notebook, proof is woven into teaching steps rather than held for a separate bridge section.
Pioneer modePioneer mode (operating mode) — The operating mode for first-batch creative work against a new expert or new market position, where no own-performance data exists. Format selection in Pioneer mode relies on competitor intelligence and theoretical EC Prior values.
Iteration modeIteration mode (operating mode) — The operating mode for subsequent batches, where live performance data from Fibery/Supabase is available. Format selection weights toward formats that have demonstrated EC Prior separation in practice, validated by the creative-strategy skill's path scoring.
FiberyFibery (project management and system of record) — The system of record for all creative entities, production status, and Meta performance data. Downstream of scripting — ad-log writes to Fibery; scripting skills read from Supabase views, never Fibery directly at runtime.
SupabaseSupabase (database platform) — The PostgreSQL-based database hosting the SKL ad creative system data layer — all seven layers from raw ad captures through to launch learnings. Scripting skills read from Supabase views (expert config, proof bank, Voice Book).
OpenRouterOpenRouter (model routing proxy) — The API gateway through which all language model calls in this system are routed. Provides unified billing, model switching, and per-call logs (model identifier, provider, cost) that feed the agent_run audit trail.