End-to-End Worked Example

This fictional example demonstrates how a raw podcast excerpt becomes a story sequence, a camera/lighting plan, an edit, a graphic, and several delivery versions.

1. Project brief

project_id: grammar_demo_01
format: video_podcast
master_target: 6 minutes
short_target: 45 seconds
audience: independent creators
outcome: understand why transcript-first editing needs context graphs
central_question: How can software cut speech without changing meaning?
tone: clear, skeptical, constructive
hard_constraints:
  - preserve quotation meaning
  - mark unsupported factual claims
  - captions required
  - no generated imagery presented as evidence

2. Source setup

CAM_A = MASTER two-shot
CAM_B = HOST_PRIMARY medium close
CAM_C = GUEST_PRIMARY medium close
CAM_D = GUEST_CLOSE
AUDIO_H = host isolated microphone
AUDIO_G = guest isolated microphone
SCREEN_1 = software demonstration capture

3. Raw transcript excerpt

[00:12:03.200 HOST]
So you are saying the transcript itself is not enough?

[00:12:06.480 GUEST]
Not exactly. The words are enough to search, but not enough to make every cut.
A sentence can look complete and still depend on what “it” means, which question
was asked, or whether the speaker corrects themselves ten seconds later.

[00:12:20.040 HOST]
Give me an example.

[00:12:21.210 GUEST]
Yesterday I said, “We doubled conversion.” That sounds like a result. But the next
sentence was, “Only in the returning-customer group, and the sample was small.”
If an editor takes the first sentence by itself, the words are real, but the claim
has changed.

[00:12:40.000 HOST]
So the clip is authentic, but the meaning is not.

[00:12:43.020 GUEST]
Right. A trustworthy editor has to store the dependencies around the sentence.

4. Transcript segmentation

[
  {
    "id": "u1",
    "speaker": "host",
    "range": ["00:12:03.200", "00:12:06.100"],
    "type": "QUESTION",
    "text": "So you are saying the transcript itself is not enough?",
    "topics": ["transcript_editing"],
    "links": ["answers:u2"]
  },
  {
    "id": "u2",
    "speaker": "guest",
    "range": ["00:12:06.480", "00:12:19.500"],
    "type": "QUALIFIED_CLAIM",
    "propositions": [
      "words are sufficient for search",
      "words alone are insufficient for every edit",
      "meaning can depend on referents, question scope, or later correction"
    ],
    "links": ["answers:u1", "example_requested_by:u3"]
  },
  {
    "id": "u3",
    "speaker": "host",
    "range": ["00:12:20.040", "00:12:20.950"],
    "type": "SPECIFICITY_QUESTION",
    "text": "Give me an example.",
    "links": ["answered_by:u4"]
  },
  {
    "id": "u4",
    "speaker": "guest",
    "range": ["00:12:21.210", "00:12:39.600"],
    "type": "EXAMPLE",
    "claims": ["c1", "c2"],
    "links": ["answers:u3", "summarized_by:u5"]
  },
  {
    "id": "u5",
    "speaker": "host",
    "range": ["00:12:40.000", "00:12:42.700"],
    "type": "SYNTHESIS",
    "text": "So the clip is authentic, but the meaning is not.",
    "links": ["confirmed_by:u6"]
  },
  {
    "id": "u6",
    "speaker": "guest",
    "range": ["00:12:43.020", "00:12:48.900"],
    "type": "TAKEAWAY",
    "text": "A trustworthy editor has to store the dependencies around the sentence."
  }
]

5. Claim objects

c1:
  text: "We doubled conversion."
  scope_as_spoken_initially: unspecified
  evidence_state: CONTEXT_DEPENDENT
  risk: high_without_qualification
  dependency: c2

c2:
  text: "The doubling applied only to returning customers, and the sample was small."
  function: QUALIFICATION
  evidence_state: EXAMPLE_WITHIN_SPEAKER_ACCOUNT

The graph contains:

c2 --QUALIFIES--> c1
u3 --QUESTION_ANSWER--> u4
u4 --EXEMPLIFIES--> u2
u5 --SYNTHESIZES--> u2 + u4
u6 --TAKEAWAY_FROM--> u2 + u4 + u5

6. Candidate clips

Candidate A — sensational fragment

Range: 00:12:21.210–00:12:27.100
Text: “Yesterday I said, ‘We doubled conversion.’ That sounds like a result.”
Context completeness: 0.32
Hook strength: 0.82
Risk: removes essential qualification
Decision: REJECT_STANDALONE

Candidate B — complete example

Range: 00:12:20.040–00:12:39.600
Includes question, initial claim, qualification, and meaning risk.
Context completeness: 0.93
Hook strength: 0.70
Educational value: 0.94
Decision: SELECT_LONG_FORM

Candidate C — compact short-form unit

Range: 00:12:21.210–00:12:48.900
Removes the host’s first question but includes example, synthesis, and takeaway.
Context completeness: 0.90
Self-containment: 0.88
Decision: SELECT_SHORT_WITH_OPENING_CARD

Opening card:

“Can a real quote become a false clip?”

7. Beat plan

8. Narrative logic

HOOK: a real quote can become a false clip
BECAUSE: a sentence inherits scope and qualification
EXAMPLE: “doubled conversion”
BUT: only one group and a small sample
THEREFORE: source authenticity is insufficient
WHICH MEANS: the editor needs a context graph

9. Camera plan

B1

picture: black field to extreme close of transcript text
camera_function: ABSTRACT/HOOK
movement: slow digital push on the words "doubled conversion"
exit: reveal cropped-off qualification below frame

B2

shot: CAM_A_MASTER then CAM_C_GUEST_PRIMARY
reason: establish relation; transfer information ownership to guest
cut: question begins on host, guest listening reaction, guest answer on CAM_C

B3

shot: CAM_C_GUEST_PRIMARY
size: MEDIUM_CLOSE
azimuth: THREE_QUARTER
movement: LOCKED_OFF
insert: SCREEN_1 transcript text
reaction: CAM_B only after qualification, if temporally true and meaningful

B6

shot: CAM_D_GUEST_CLOSE
reason: final principle and direct emphasis
hold_after_line: 12 frames before exit

10. Lighting plan

motivation: large window camera-right plus warm practical in rear
key:
  side: far_side
  quality: soft
  elevation: slight_high
fill:
  level: key_minus_2_stops
background:
  practical: warm
  separation: moderate
continuity_locks:
  - key_direction
  - practical_on
  - white_balance
story_change:
  B1 graphic is high contrast
  B2-B5 use stable readable interview light
  B6 subtly lowers background level, preserving source logic

The B6 change is minor: it increases focus without inventing a new environment.

11. Graphic plan

graphic_id: G1
task: EXPLAIN_MECHANISM
type: CAUSAL_DEPENDENCY_GRAPH
objects:
  - sentence: "We doubled conversion"
  - scope: "returning customers only"
  - limitation: "small sample"
links:
  - scope QUALIFIES sentence
  - limitation REDUCES_CERTAINTY sentence
motion:
  - ORIENT: show sentence alone
  - BUILD: reveal scope node
  - BUILD: reveal limitation node
  - EMPHASIZE: show meaning changes when links are removed
  - HOLD: display complete graph
accessibility:
  narration_alt: "The first sentence depends on two linked qualifications."
  color_redundancy: labels and line types
  source_note: "Fictional example"

12. Edit plan

00:00.000 BLACK + room tone
00:00.250 card: “Can a real quote become a false clip?”
00:02.000 J-cut guest: “Yesterday I said…”
00:02.500 CAM_C guest medium-close
00:06.200 SCREEN_1 insert: highlight “We doubled conversion.”
00:09.000 return CAM_C before “But the next sentence…”
00:11.000 G1 starts; guest audio continues
00:18.500 G1 completes and holds
00:20.000 CAM_B host synthesis
00:23.000 J-cut guest confirmation over host listening reaction
00:23.500 CAM_D guest close
00:28.500 hold 12 frames, then cut to next beat

Cut logic:

J-cut at opening creates anticipation.
Insert proves the exact language.
Graphic shows—not merely repeats—the dependency mechanism.
Host synthesis changes information owner.
Final close-up increases emotional/intellectual emphasis.

13. Audio plan

Dialogue: isolated microphones, matched room perspective
Room tone: continuous under transcript/graphic inserts
Music: none under the example; a low restrained motif may enter after synthesis
Captions: speaker IDs at first appearance; exact qualification retained
Loudness: rendered through selected online podcast profile

14. LAKA alternatives

Baseline

clean multicamera exchange in original chronology

Minor

alternate opening card; one more document insert; slight pacing adjustment

Major

open on the misleading fragment, then visibly restore its missing context

Structural

intercut this example with a second case where a pronoun is misresolved

Paradigm

turn the segment into an interactive editor: the viewer chooses a clip boundary and sees how meaning changes

15. Vertical derivative

B1 text reflows vertically.
CAM_C and CAM_D use stable face crop.
SCREEN_1 transcript becomes full-screen rather than side-by-side.
G1 becomes a vertical three-node stack.
Host and guest are stacked only during the synthesis/confirmation exchange.
Captions occupy a reserved lower-middle region clear of platform UI.

16. QA trace

PASS — qualification remains attached to initial claim
PASS — reaction, if used, remains in source time
PASS — fictional status appears on graphic
PASS — audio-only listener understands the example
PASS — vertical version preserves sentence and qualifications
PASS — no generated reenactment presented as evidence
PASS — captions include all words and speaker identity
WARNING — “doubled conversion” would be misleading as thumbnail text alone
RESOLUTION — thumbnail must include “only one group” or use a question form

17. Final reusable logic

IF a sentence sounds complete
BUT depends on scope, referent, correction, or qualification
THEN the sentence is not a standalone clip.

IF the context can be restored compactly
THEN expand the clip or add a truthful bridge.
ELSE reject standalone use.