Script Supervisor Audit Engine

The Script Supervisor Audit Engine is the director’s mental pre-flight checklist and verification gate for the Auteur Script framework. It enables the writer/director to systematically evaluate the structural integrity, spatial geometry, prop lifecycles, and temporal continuity of a scene before dispatching to generation.

1. Context-Aware Auditing: The 4 Scene Topologies

A script supervisor adapts continuity rules to the narrative topology of the scene. Applying linear continuity to a trailer or montage is an error, just as allowing jump-cuts in a real-time dialogue is an error.

Topology Type	Structural Formula	Directorial Scope & Invariant Focus
1. Continuous Real-Time	\(S_0 \mapsto S_1 \mapsto \dots \mapsto S_n\)	Single unbroken takes or real-time coverage. Strict 180° axis, prop hand-tracking, matching eyelines, and physical state accumulation.
2. Parallel Cross-Cutting	\(A_k \parallel B_k\)	Intercut scenes in separate locations (e.g., phone calls, dual perspectives). Dual independent State Ledgers with zero cross-stream bleeding.
3. Discontinuous / Montage	\(S_a \xrightarrow{\text{CUT}} S_b \xrightarrow{\text{CUT}} S_c\)	Elliptical time, trailers, flashforwards, and memory montages. Explicit cut tags (`[HARD CUT]`, `[MATCH CUT]`) to eliminate AI morphing risks.
4. Spatial Journey / Portal	\(S_{n} \xrightarrow{\text{traverse}} S_{n+1}\)	Continuous subject tracking across spatial thresholds (e.g., street \(\to\) speakeasy). Lighting and acoustic modulation at boundary thresholds.

Topology 1: Continuous Real-Time Scene (Linear \(S_0 \mapsto S_1 \dots \mapsto S_n\))

Scope: Unbroken takes, standard dialogue coverage, room-scale scenes, continuous action choreography.
Continuity Mandate: Complete state inheritance (\(S_n = f(S_{n-1})\)).
Enforced Rules:
1. Strict 180-degree axis discipline and matching eyelines across reverse angles.
2. Persistent prop hand-tracking (which hand holds an object across cuts).
3. Progressive condition accumulation (dirt, sweat, tears, blood, fabric damage persist).
4. Stable Screen-L / Screen-C / Screen-R placement relative to environment anchors.

Topology 2: Parallel Cross-Cutting / Intercut (Dual Stream \(A_k \parallel B_k\))

Scope: Intercut scenes occurring simultaneously in distinct physical spaces (e.g., telephone calls, dual perspectives, heist coordination, countdowns).
Continuity Mandate: Dual independent State Ledgers.
Enforced Rules:
1. Cross-Stream Isolation: State \(A_{k}\) inherits strictly from \(A_{k-1}\), never from \(B_{k-1}\). Environmental attributes from Location A (rain, darkness) must not bleed into Location B.
2. Temporal Synchronization: Elapsed physical action in Location A must correspond realistically with elapsed time in Location B.
3. Acoustic Bridge Discipline: Clearly separate local room acoustics from off-screen sound bridges or telephone audio filters.

Topology 3: Discontinuous / Montage / Trailer (Elliptical Time)

Scope: Movie trailers, training montages, music video rapid cuts, memory flashbacks, dream sequences.
Continuity Mandate: Decoupled spatial continuity with strict boundary demarcation.
Enforced Rules:
1. Suppressed Checks: Disable cross-cut spatial position matching, prop persistence, and axis locks across scene boundaries.
2. Hard-Cut Demarcation: Enforce explicit transition tags ([HARD CUT], [SMASH CUT], [MATCH CUT], [FLASH]) to prevent AI video generation models from morphing or blending scenes together.
3. Self-Contained Micro-Anchors: Every individual vignette must establish its own standalone subject, lighting, and environment coordinates.
4. Pacing & Audio Cues: Audit rhythmic audio impacts, braams, music drops, and sound risers aligned to visual cut points.

Topology 4: Spatial Journey / Portal Transition (Unbroken Environment Shift)

Scope: Continuous tracking shots that traverse distinct spatial thresholds (e.g., exterior street \(\to\) doorway \(\to\) interior hall \(\to\) stage).
Continuity Mandate: Subject persistence with dynamic background and lighting modulation.
Enforced Rules:
1. Threshold Modulation: Explicitly shift environmental lighting ([ENV]) and acoustic properties ([AUDIO]) at physical portal boundaries.
2. Anchor Re-establishment: Subject remains unbroken while background anchors dynamically update.

2. The 5 Script Supervisor Audit Lenses

Every script is inspected through 5 specialized audit lenses prior to finalization:

Lens	Audit Domain	Core Inspection Focus	Key Rule Verified
Lens 1	Spatial & Optics	180° Axis, eyelines, Screen-L/R anchors, optical depth (`[FOCUS]`)	No unmotivated axis flips; stable spatial geography.
Lens 2	Temporal & Wear	Physical degradation (sweat, tears, soot, blood), elapsed time, framerate (`[SPEED]`)	No unprompted resets of physical wear or injury.
Lens 3	Props & Inventory	Hand allocation, item permanence, autonomous mechanics (`[PROP]`)	No vanishing objects; consistent hand-tracking.
Lens 4	Acoustics & Voice	Foley, room sound, score (`[MUSIC]`), `(O.S.)` / `(V.O.)` status	Clear separation of diegetic sound from score and voiceover.
Lens 5	Blocks & Grammar	Block orthogonality, Single-Dimension Law, character isolation	No redundant lookbook text in `[EXECUTION]`; tag purity.

Lens 1: Spatial Geography, Axis & Optics

First-Frame Anchor: Does [OPENING] establish a stable static composition, and does line 1 of [EXECUTION] inherit those exact coordinates?
180° Line of Action: Is the imaginary axis between subjects maintained?
Axis Crossing Protocol: If the camera crosses the line, is it accompanied by an explicit neutral buffer shot (POV, direct frontal CU) or a continuous tracking arc?
Screen Direction & Eyelines: Do subjects on Frame-Left look Screen-Right, and subjects on Frame-Right look Screen-Left?
Screen Position Stability: Are characters anchored consistently to Frame-Left, Frame-Center, or Frame-Right across cuts?
Optical Depth & Focal Plane ([FOCUS]): When rack focusing, is the depth transition between foreground, midground, and background clearly separated from camera rig motion?

Lens 2: Temporal Logic, Framerate & Physical Degradation

State Accumulation: Do physical conditions (sweat-sheen, tears, soot, blood, damp fabric, shivering) persist without unprompted resets?
Physical Action Plausibility: Are described character movements humanly achievable within the shot duration without instantaneous teleportation?
Framerate & Temporal Ramping ([SPEED]): Are high-speed capture (120fps/240fps) or speed ramps anchored cleanly to physical impacts or atmospheric beats?
Elapsed Time Integrity: Do dynamic elements (candle burn, dripping water, melting ice, burning fires) progress forward logically?

Lens 3: Prop & Inventory State Machine

Hand Allocation: Is the specific hand (left hand, right hand, both hands) explicitly tracked for held items?
Prop Lifecycle Integrity: Does every prop follow a valid state transition (Stored \(\to\) Drawn/Held \(\to\) Used \(\to\) Put Down/Holstered/Dropped)?
Hardware & Mechanical State ([PROP]): Are internal object mechanisms, clockwork gears, or weapon actions clearly isolated from character anatomy?
Zero Ghost Props: Do items avoid spontaneously appearing in hands or vanishing from surfaces without explicit action?

Lens 4: Acoustic & Dialogue Architecture

Dialogue Staging: Are dialogue lines correctly flagged with (O.S.) for off-screen characters and (V.O.) for narration?
Foley Synchronization: Does every significant physical impact in [ACT] or [PROP] correspond to an explicit [AUDIO] or [SFX] foley cue?
Ambient Persistence: Are continuous environmental sound textures (rain on tin, engine drone, room tone) sustained where active?

Lens 5: Block & Grammar Orthogonality

Intent Discipline & Anti-Bloat: Does [INTENT] focus purely on directorial vision, theme, and stakes (ideally adhering to the recommended 1-sentence baseline formula: “Format + Location + Characters + Stakes”) while strictly avoiding novelistic backstories or play-by-play camera moves?
Staging-to-Execution Conditioning: Does [EXECUTION] strictly operate within the physical constraints of [LOGIC] and the static lookbook of [AESTHETIC] (\(S_n = f(S_{n-1} \mid \text{STAGING})\))?
State Machine Initialization (\(S_0 \to S_1\)): Does line 1 of [EXECUTION] inherit the exact camera angle, subject coordinates, and prop hand allocations from [OPENING] (\(S_0\)) without teleportation or angle jumps?
Entity Identity Inheritance: Are canonical character names bound in [INTENT] used consistently across all downstream blocks without raw @actor reference tag leaks?
Zero Lookbook Bleed: Are static wardrobe, lighting, and camera specs defined in [AESTHETIC] kept out of [OPENING] and [EXECUTION]?
Dynamic Light Separation ([LIGHT]): Are active real-time lighting changes (headlights, flickers, shadows) governed by [LIGHT] without duplicating static lookbook lighting?
Background Extras Isolation ([CROWD]): Are secondary bystanders, crowds, and traffic cleanly governed by [CROWD] to prevent morphing with primary characters in [ACT]?
Single-Dimension Tag Purity: Does every Sub-State tag govern strictly one dimension (e.g., [CAM] contains only optics/moves; [ACT] contains only physical performance)?
Platform Syntax Integrity: When extensions are inactive, is baseline plain-text syntax preserved with zero accidental platform braces ({}) or tokens?

3. Pre-Flight State Ledger (Internal Mental Simulation)

Before generating output lines, the architect mentally constructs an Internal State Ledger to anchor all coordinates:

[INTERNAL SPATIAL LEDGER]
• Topology: [Continuous | Cross-Cut | Montage/Trailer | Spatial Journey]
• Staging Setup: [INTENT: Objective J | LOGIC: Constraint Matrix C | AESTHETIC: World State Ω | OPENING: Initial Vector S_0]
• Primary Axis: [Subject A (Frame-L) <─── Line of Action ───> Subject B (Frame-R)]
• Active Props: [Prop 1: Held in Subject A Right Hand | Prop 2: On Center Table]
• Physical Conditions: [Subject A: Rain-soaked collar | Subject B: Dry]
• Background Anchors: [Frame-L: Open window | Frame-R: Oak door | BG: Stone hearth]

4. Automated Directorial Repair Recipes

When the audit detects a continuity or grammar flaw, execute the corresponding deterministic repair:

Detected Flaw	Script Supervisor Directorial Repair
Plot Dump / Play-by-Play Intent	Strip play-by-play camera moves or novelistic backstories from `[INTENT]`, condensing it toward the recommended 1-sentence baseline formula: `"Create a [Format / Tone] scene set in [Location], featuring [Characters] [Core Stakes]."`
Staging Disconnect (\(S_0 \not\to S_1\))	Re-anchor Line 1 of `[EXECUTION]` to mutate directly from the exact framing, body posture, and prop grips locked in `[OPENING]`.
Accidental 180° Axis Break	Insert a neutral buffer shot (direct frontal CU, over-the-shoulder shift, or POV) or specify an explicit arcing camera movement across the line.
Ghost Prop / Vanishing Object	Inject an explicit hand-tracking Sub-State (e.g., `[ACT] clutches bronze key in right fist`) or item physics (`[PROP]`) to re-anchor the item before the cut.
Lookbook Bleed / Static Redundancy	Strip static color, wardrobe, and lens descriptions out of `[OPENING]` and `[EXECUTION]`, leaving only active spatial staging.
AI Morphing Risk in Hard Cuts	Insert explicit hard-cut tags (`[HARD CUT]`, `[SMASH CUT]`, `[MATCH CUT]`) and supply a self-contained micro-anchor (subject + lighting) on the new line.
Primary Actor & Crowd Blending	Isolate background extras into a dedicated `[CROWD]` tag, keeping `[ACT]` strictly restricted to primary named subjects.
Blended / Multi-Dimensional Tag	Split the blended tag into orthogonal single-dimension Sub-States (e.g., convert `[CAM] Zooms in as John falls` into `[CAM] PUSH-IN CU -> [ACT] John stumbles forward`).
Multimodal Reference Leakage	Move `@actor` / `@image` declarations into `[INTENT]` and replace downstream occurrences with canonical character names.

5. Diagnostic Audit Report Mode

When the creator explicitly requests an audit using commands such as “Audit”, “Check continuity”, “Script supervisor report”, or “Diagnostic”, output the script followed by an explicit Script Supervisor Audit Report:

=== SCRIPT SUPERVISOR AUDIT REPORT ===
• Scene Topology: [Continuous Real-Time / Cross-Cut / Discontinuous Montage / Spatial Journey]
• Staging & Block Pipeline: [Verified: INTENT (J) -> LOGIC (C) -> AESTHETIC (Ω) -> OPENING (S_0)]
• 180° Axis Health: [Verified / Repaired with neutral buffer shot]
• Prop & Hand Ledger: [Tracked: Item A (Right Hand), Item B (Table Surface)]
• State Inheritance: [Verified: S_0 -> S_1 ... -> S_n unbroken condition carryover]
• Block Orthogonality: [Verified: Zero lookbook redundancy in EXECUTION]
• Grammar & Tag Purity: [Verified: Single-dimension tags, clean extension isolation]
=======================================