# Appendix G: Vertical Shot Grammar

> **Empirical grammar rules for 9:16 vertical microdramas.**
> Consumed by the storyboard agent (`storyboard_agent.md`) and enforced by the validator (`validate_storyboard.py`).
> Derived from the Visual Grammar Bible research (1,955 shots analyzed — see `/_research/visual_grammar_bible/FINDINGS.md`), adjusted for Leviathan creative direction.

---

## A. Shot Scale Targets (Leviathan-Adjusted)

Based on FINDINGS.md empirical data, adapted for JT's creative direction — more close-ups and ECUs than the average ReelShort show.

| Scale | Target | Range (per ~35 shots) | Notes |
|-------|--------|----------------------|-------|
| ECU | 10-15% | 3-5 | Core vertical advantage — eyes, hands, objects |
| CU | 20-30% | 7-10 | Emotional core, reaction shots |
| MCU | 25-35% | 9-12 | Dialogue workhorse |
| MS | 15-25% | 5-9 | Action staging, blocking |
| WIDE/LS | 5-15% | 2-5 | World-building beats, tension release |

**Rationale:** The raw corpus shows MCU 41% / CU 10% — but Leviathan is sci-fi action with world-building needs. We shift weight toward ECU and CU for emotional punch and toward WIDE for spatial grounding, pulling from MCU's dominance.

**Key difference from corpus:** Leviathan KEEPS wide establishing shots. The "No-Geography Rule" (cold open, no establishing) applies to hooks but NOT to setup beats. Leviathan is world-building — the audience needs to see the station, the shafts, the void. Wide shots provide tension release between action.

---

## B. Cut Grammar Rules

| Rule | Description | Severity |
|------|-------------|----------|
| Axis/Scale Change | Same subject consecutive: need 2+ scale steps (per `SCALE_ORDER`) OR different angle | ERROR |
| No Triple Repeats | Max 2 consecutive same `shot_type` | ERROR (ref: `CONSTANTS.md → SAME_TYPE_CONSECUTIVE_MAX`) |
| 180° Rule | Camera stays same side of action line within a scene | WARNING |
| Spatial Matching | CU must match preceding wider shot's spatial context | WARNING |
| ECU Minimum | At least 2 ECUs per episode (ref: `CONSTANTS.md → ECU_MIN_PER_EPISODE`) | WARNING |
| Scene Coverage | Each scene (4+ shots) needs tight AND wide | WARNING (already enforced) |

### Scale Order Reference

Used for calculating scale change distance:

```
ECU (0) → CU (1) → MCU (2) → MS (3) → LS (4) → WIDE (5)
```

A "2+ scale steps" change means jumping at least 2 positions in this order. Examples:
- CU → MS = 2 steps (OK)
- MCU → MS = 1 step (needs angle change to be valid)
- ECU → MCU = 2 steps (OK)

---

## C. Beat-Type Grammar Templates

Different Kill Box beats demand different visual grammar. This is the key creative lever — not all shots are created equal.

### HOOK (Action Grammar)
- **Shot types:** ECU/CU heavy, rapid cuts
- **Establishing:** NONE — cold open, in media res (3.7.21 Rule)
- **Typical count:** 1-3 shots
- **Grammar:** Drop the audience into the moment. No geography, no orientation. The first image should be a visual punch — hands, eyes, impact, texture.
- **Template:** `ECU action → CU face → [optional MS context]`

### SETUP (Establishing Grammar)
- **Shot types:** WIDE/LS → MCU → MS progression
- **Establishing:** YES — this is where geography lives
- **Typical count:** 2-4 shots
- **Grammar:** Orient the viewer in space. Start wide (or medium-wide) and move inward. This is Leviathan's world-building moment — the station architecture, the light quality, the scale of the environment.
- **Template:** `WIDE/LS location → MS blocking → MCU character`

### ESCALATION (Action Grammar)
- **Shot types:** Building tempo, ECU punches for impact
- **Establishing:** No — the audience is already oriented from SETUP
- **Typical count:** 4-6 shots
- **Grammar:** Cuts get faster. Scale gets tighter. Use Impact Beats pattern for violence (see Section D). ECUs on hands, eyes, objects during action moments. MS for spatial orientation BETWEEN action beats.
- **Template:** `MCU action → CU reaction → ECU impact → MS spatial reset → CU reaction → ECU detail`

### TURN (Pivot Grammar)
- **Shot types:** Held shot for weight, then rapid scale shift
- **Establishing:** Possible reset if location changes
- **Typical count:** 4-7 shots
- **Grammar:** The pivot moment. May include an extended take — hold on a face for the realization to land. Then rapid scale shift (wide to ECU or ECU to wide) to mark the change. The rhythm breaks here intentionally.
- **Template:** `MCU/CU held → [beat] → rapid scale shift → new rhythm`

### CLIFFHANGER (Freeze Grammar)
- **Shot types:** Final image lingers, visual rhyme with hook
- **Establishing:** Rarely — end tight, not wide
- **Typical count:** 2-4 shots
- **Grammar:** Less is more. The final image should burn into the viewer's retina. Often mirrors the hook's framing with transformed meaning. Avoid cutting fast — let the final shot sit.
- **Template:** `MCU/CU setup → ECU/CU final image [hold]`

---

## D. Action Sequence Grammar (Vertical-Specific)

### Impact Beats Pattern (57% prevalence in corpus)

Traditional film choreographs fights with wide shots showing full body movement. Vertical can't — the 9:16 frame is too narrow. Instead, violence is conveyed through **montage**:

1. **CU aggressor face** — intent, anger, determination
2. **Sound beat** — no wide shot of contact (cut away from the hit)
3. **CU/ECU victim reaction** — shock, pain, blood
4. **MS/MCU room reaction** — witnesses, environment response

This is faster, cheaper, and more emotionally effective than choreography on a phone screen.

### Leviathan Action Additions

- **ECU on impact points** — hands gripping, eyes widening, objects breaking. During action sequences, ECU is your most powerful tool. Fill the vertical frame with texture and force.
- **Wider shots BETWEEN action beats** — after an Impact Beats sequence, reset with an MS or wider shot for spatial orientation before the next hit. The audience needs a breath to locate themselves.
- **Camera static 77%** — movement is the exception. Use shake ONLY for impact moments (6% in corpus). Don't motivate movement without a storytelling reason.
- **DOF as weapon** — shallow DOF (54.5% corpus default) isolates subjects during action. Rack focus between aggressor and victim for tension.

### When to Use Impact Beats

Use this pattern when:
- Physical violence occurs (hits, falls, explosions)
- Emotional violence lands (revelations, betrayals) — same pattern works metaphorically
- The action is too complex for a single wide shot

Do NOT use for:
- Slow confrontations (use held shots instead)
- Chase/movement sequences (use tracking/pan)
- Environmental reveals (use wide → tight progression)

---

## E. Two-Shot Vocabulary

Three valid types for vertical composition:

### Stacked Depth
- **Composition:** One character foreground (soft focus), one background (sharp focus) — or vice versa
- **Blocking:** Characters at different depths on the z-axis
- **Reads as:** Power dynamics, surveillance, secrets, emotional distance
- **When to use:** One character watches/follows another, eavesdropping, power imbalance
- **Vertical advantage:** The tall frame accommodates depth composition naturally

### True Two-Shot
- **Composition:** Both characters visible in frame, sharing the space equally
- **Blocking:** Characters may be side-by-side, facing each other, or at slight angles. Cropping is a function of shot size:
  - MS two-shot = waist up for both
  - CU two-shot = faces with backs of heads cropped
- **Reads as:** Intimacy, confrontation, resolving separation, alliance
- **When to use:** Face-to-face conversation, physical closeness, moments of connection or conflict where BOTH characters' presence matters equally
- **Vertical note:** Tighter than 16:9 two-shots — one character will always be slightly more cropped. This is a feature, not a bug — it creates dynamic asymmetry.

### Over-Shoulder / Dirty Single
- **Composition:** One character's back/shoulder in foreground, other character's face in focus
- **Blocking:** Shot past one character toward the other
- **Reads as:** NOT a two-shot — this is a single with context. Dialogue exchanges, reaction emphasis.
- **When to use:** Standard dialogue coverage, when you need to show who a character is talking to without committing to a two-shot
- **Note:** The foreground character is compositional context, not a subject. The camera favors the background face.

---

## F. Edit-Pair Relationships

From FINDINGS.md (55% continuation, 23% reaction, 15% contrast):

| Relationship | Prevalence | Description | Example |
|-------------|-----------|-------------|---------|
| Continuation | 55% | Same moment, different angle/scale | CU face → MCU hands (same action beat) |
| Reaction | 23% | Character response to previous shot | CU aggressor → CU victim (Impact Beats) |
| Contrast | 15% | Tonal or spatial shift | Warm interior → cold void exterior |
| Parallel | 3.5% | Simultaneous action in different locations | Rare in microdramas — stay spatially tight |
| Answer | 1.8% | Resolution of visual question | ECU locked door → MS character finding key |
| Flashback | 1.1% | Memory or backstory insert | Use sparingly — disrupts pacing |

### How to Use in Storyboarding

For every shot after #1, the storyboard agent should specify:
1. **What cut it follows** — previous shot's scale + subject + angle
2. **What the edit achieves** — which relationship type (continuation, reaction, contrast)
3. **Whether the axis/scale change rule is satisfied** — 2+ scale steps or different angle

This creates **edit-pair thinking** — the agent considers how each shot relates to its neighbors, not just what each shot contains in isolation.

---

## G. Insert/Detail Shot Rules

> Full rules defined in `storyboard_agent.md` → Step 3 → "Insert/Detail Shot Rule (MANDATORY)".
> This section cross-references, not duplicates.

**Core principle:** Detail shots are **captured moments**, not product photography. The camera is a documentary camera that catches things in passing.

**Every detail shot must include:**
1. Action in progress (the object is being USED)
2. Oblique angle (never straight-on)
3. Context bleeds in (body/environment at frame edges)
4. Imperfect framing (documentary feel)
5. Depth of field (the object lives in a SPACE)

**In the context of vertical grammar:** Insert shots serve as **punctuation** in the edit rhythm. They break up face-to-face dialogue patterns, provide evidence (the "Proof Shot" — contracts, texts, displays), and create breathing room between intense close-ups.

**Frequency target:** 1-2 insert/detail shots per episode (2.2% of corpus). Don't overuse — every insert should earn its place by advancing plot or revealing character.

---

## Cross-References

| Topic | Document |
|-------|----------|
| Numeric constants | `/CONSTANTS.md` → Shot Grammar section |
| Storyboard agent protocol | `/agents/storyboard_agent.md` |
| Storyboard validator | `/tools/validate_storyboard.py` |
| Empirical research | `/_research/visual_grammar_bible/FINDINGS.md` |
| Lens package defaults | `/CONSTANTS.md` → Default Lens Package |
| Insert shot rules (full) | `/agents/storyboard_agent.md` → Step 3 |
| AI video strategy | `/appendix_d_ai_video.md` |
| Flux 2 prompting | `/appendix_e_flux2_protocols.md` |

---

*Derived from Visual Grammar Bible research (1,955 shots, 33 scenes, 3 mediums)*
*Adjusted for Leviathan creative direction (Feb 2026)*
*Run `python3 /tools/validate_storyboard.py` to enforce these rules mechanically*