1. The production decision envelope
Every production decision uses the same internal variables.
| Variable | Question |
|---|---|
| Object | What is being acted upon? |
| Conditions | Under what circumstances does the system operate? |
| Actions | What does the system do? |
| Tools | What mechanisms carry out the action? |
| Resources | What information, assets, people, compute, or capital are required? |
| Outcomes | What does the system produce? |
| Feedback | How does the result change later decisions? |
| Constraints | What limits performance, legality, continuity, or adoption? |
| Value | For whom is value created, and what kind? |
| Failure mode | How can the decision fail? |
2. Five change levels
| Level | Meaning | Media example |
|---|---|---|
| 0 — Baseline | Preserve structure and intent | Remove noise, correct transcript, switch to active speaker |
| 1 — Minor change | Change a local property | Slight punch-in, lower-third, shorten a pause |
| 2 — Major change | Change treatment of a beat | Replace talking head with B-roll sequence and diagram |
| 3 — Structural change | Change organization or coverage | Move a reveal earlier, rebuild an act, generate missing scene coverage |
| 4 — Paradigm change | Change governing model | Convert interview into narrated investigation, animation, or first-person essay |
A change level is not a quality score. A baseline solution can be the best solution.
3. Fourteen LAKA meta-variables
Each variable uses a normalized value from 0.0 to 1.0, plus a human-readable state.
| Meta-variable | First-principles question | Example in editing |
|---|---|---|
| Magnitude | How much changes? | Two-frame trim versus full scene replacement |
| Rate | How quickly does the state change? | Slow reveal versus smash cut |
| Direction | Toward what state? | Toward intimacy, clarity, threat, relief, proof |
| Scope | How broadly does it apply? | Word, clip, beat, scene, episode, series |
| Depth | How fundamental is the change? | Surface styling versus meaning or structure |
| Duration | How long does the effect persist? | Flash frame versus entire act |
| Frequency | How often does it recur? | One punch-in versus recurring visual motif |
| Acceleration | Is the rate increasing or decreasing? | Cuts become progressively faster |
| Variability | How consistent or unpredictable is it? | Uniform pacing versus deliberate irregularity |
| Detectability | How visible is the intervention? | Invisible continuity edit versus obvious glitch |
| Reversibility | Can it be undone without loss? | Metadata marker versus destructive render |
| Propagation | What downstream objects change? | Moving one beat shifts captions, music, graphics, chapters |
| Amplification | How much does the audience effect multiply? | Silence before a confession amplifies emotional impact |
| Accumulation | How do repeated effects build? | Repeated visual motif becomes a theme |
4. LAKA change vector
{
"magnitude": 0.35,
"rate": 0.20,
"direction": ["intimacy", "credibility"],
"scope": ["beat"],
"depth": 0.25,
"duration": 8.4,
"frequency": 1,
"acceleration": 0.0,
"variability": 0.15,
"detectability": 0.20,
"reversibility": 1.0,
"propagation": ["captions", "music_cue"],
"amplification": 0.70,
"accumulation": 0.10
}5. Volumetric dimension families
A candidate can carry values in all of these families.
- Source — origin, media type, rights, authenticity, quality.
- Time — source range, output range, duration, handles, synchronization.
- Language — words, syntax, speaker, intent, confidence, translation.
- Semantics — entities, topics, propositions, claims, evidence.
- Discourse — question/answer, support, contrast, cause, reference.
- Narrative — beat role, tension, stakes, reveal, payoff, arc.
- Audience — knowledge, emotion, needs, objections, accessibility.
- Performance — energy, pace, clarity, authenticity, gesture, gaze.
- Shot — scale, angle, lens, distance, focus, composition, movement.
- Lighting — role, direction, quality, ratio, color, dynamics.
- World — location, set, props, wardrobe, weather, time of day.
- Edit — boundary, cut type, transition, duration, rhythm, continuity.
- Graphics — information relation, encoding, placement, motion.
- Audio — dialogue, music, ambience, dynamics, spatial state.
- Color/VFX — transforms, look, compositing, tracking, cleanup.
- Delivery — aspect, resolution, codec, language, channel, duration.
- Governance — rights, consent, disclosure, provenance, approvals.
- Measurement — quality signals, engagement, errors, feedback.
6. Candidate-generation algorithm
1. Determine the object and intended audience-state change.
2. Load hard constraints.
3. Select allowed dimensions.
4. Choose a LAKA change level.
5. Generate candidate values inside the allowed range.
6. Reject candidates that violate truth, rights, context, continuity, or delivery constraints.
7. Score survivors.
8. Select a diverse set across meaningful dimensions.
9. Render previews.
10. Gather human and performance feedback.
11. Update weights without overwriting source truth.7. Utility function
A configurable starting point:
Utility(candidate)
= 0.16 StoryFit
+ 0.14 Clarity
+ 0.12 ContextIntegrity
+ 0.11 EmotionalEffect
+ 0.10 EvidenceStrength
+ 0.09 Continuity
+ 0.08 VisualOrAuralQuality
+ 0.07 AudienceFit
+ 0.05 Accessibility
+ 0.04 BrandFit
+ 0.04 TechnicalReliability
- DistortionRisk
- RightsRisk
- FabricationRisk
- FatigueCost
- ComputeCostWeights must change by project. A documentary gives more weight to evidence and provenance. A comedy clip gives more weight to setup, timing, and reaction. An educational video gives more weight to clarity and transfer.
8. Diversity selection
High-scoring candidates often collapse into one style. Use a distance function.
SelectedSet
= maximize(sum(utility))
+ diversity_weight × pairwise_distanceDistance can include:
shot scale, angle, lens, movement, lighting ratio, color,
graphic mode, cut rhythm, sound treatment, narrative perspective9. Propagation graph
A decision is not isolated.
Move beat
→ changes timeline positions
→ shifts music cue
→ shifts captions
→ changes graphic timing
→ changes chapter markers
→ changes short-form derivatives
→ may change hook and retention predictionEvery decision stores affected_objects and recompute_required.
10. Feedback grammar
OBSERVE metric
COMPARE against target and prior version
LOCATE responsible objects
CLASSIFY failure mode
APPLY smallest sufficient LAKA change
RENDER alternative
VALIDATE meaning and quality
PROMOTE or revert11. Failure-mode classes
- Meaning distortion
- Missing context
- Weak causality
- Redundancy
- Pacing mismatch
- Visual discontinuity
- Audio discontinuity
- Technical failure
- Accessibility failure
- Factual or evidentiary failure
- Rights or consent failure
- Generated-media artifact
- Format/platform failure
- Excessive change
- Insufficient change
- Metric overfitting
12. Example: emotional disclosure
Object: confession beat
Condition: high vulnerability, clear face, no evidence graphic needed
Action: hold close-up; slow or stop camera movement; lower music
Outcome: intimacy and credibility
Constraint: do not remove qualifying language
Failure mode: manipulative over-scoring or distracting cutsLAKA interpretation:
Magnitude: low-to-moderate
Rate: slow
Direction: intimacy
Scope: one beat
Depth: emotional, not factual
Duration: complete thought plus afterbeat
Frequency: rare
Acceleration: none
Variability: low
Detectability: low
Reversibility: high
Propagation: music, captions, adjacent cut points
Amplification: high
Accumulation: moderate if repeated confessions form an arc