sd-jieshuo-engine

sd-jieshuo-engine · Explanatory Narrative Category Specialization Engine (Track ③)

I. Role Positioning

ForTrack ③ Explanatory Narrative Category(historical events / historical figure introductions / popular science knowledge explanations), it provides a complete rule set. Itdoes not producean entirely new workflow—the authority for the upstream locking system, feeding format, and machine-check scripts still rests with the master control; this engine is responsible for the **adaptation specifications for "how the explanatory category uses this workflow"**.

The reason for this engine's existence in one sentence:

The "lines" of the explanatory category are not dialogue, but narration. Dramatic performance carries information through "dialogue + performance + plot"; explanatory narrative carries information through "narration/commentary + visual annotations + subtitles". Modeling the same rule set on dramatic dialogue willsystematically write the explanatory category incorrectly—wrong speech-rate specifications, misjudged length thresholds, wrong storyboard units, misaligned asset constraints. All rules of this engine grow from this premise.

II. Relationship with the Master Control and the Set

Track source: This engine is mounted only when the master control 〇·0 track routing lockselects ③ explanatory category. It is not activated when the track is not selected.
Entry method: The complete workflow enters from master control short-drama-director ; the master control locking system (〇·0 track → 〇·1 material source → A·model / B·aspect ratio / C·delivery / D·segment seam) locks first, then enters the script layer.
Relationship with other sub-skills：
- With sd-script-engine it is asuperpositionrelationship (does not replace sd-script-engine)—the "script" of the explanatory category is essentially a **narration script (VO Script)** rather than a dialogue screenplay; this engine defines the structure of the narration script, its writing method, and its mapping to visuals,sd-script-engine the "source registration" discipline is still executed as usual (historical / popular science subjects have especially high requirements for source grading).
- With sd-recon-engine：when the material source locks ③ benchmark / ④ hotspot, the field-research engine is still mounted as usual to first produce the "Benchmark Intelligence Card"; for historical / popular science subjects, thehistorical fact and source disciplineis strengthened by §VII of this engine (T1 historical materials may be directly quoted in visuals; T2 discussions are only orally paraphrased).
- With sd-asset-engine(assets) / sd-storyboard-engine(storyboard) / sd-feed-engine(feeding) / sd-qc-engine(quality check) it is adownstream transferrelationship—scene style consistency > character consistencythe asset specifications, the storyboard specifications from narration sentence → visual, and the C14 quota exemption must be transferred stage by stage.
Does not own: the locking system (〇·0/〇·1/A/B/C/D), the seven-segment feeding format, the H3 format, asset numbering, machine-check scripts—these authorities all rest with the master control; this engine only provides the adaptation specifications for "how the explanatory category uses these".

III. Trigger Words and Entry

Trigger words: Commentary, film breakdown, listicle, frame-by-frame analysis, talking-head film review, historical figures, historical events, popular science, trivia, knowledge niche, "understand it all in one go", documentary narration, talking-head knowledge.

When the user directly says the trigger word, if not yet in the full pipeline, first state one sentence:

For full production (assets, storyboard, video), please enter from the master control short-drama-director ; for the track, first confirm according to master control 〇·0 (1 domestic drama / 2 overseas drama / 3 commentary category / 4 other). This engine is responsible for the rule set of the commentary narrative category.

Prerequisite for entry: the track has been confirmed as ③ commentary category, and thesubtypehas been clarified (historical figure chronicle / historical event or battle / popular science knowledge commentary, choose one of three). When the subtype is undetermined, stop first and ask—the three types share one skeleton, but thehook and emotional arc differ(see §五).

四、Track definition: the essential differences between the commentary narrative category (③) and the plot performance category (①)

Dimension	① Domestic drama (plot performance)	③ Commentary narrative category (this engine)
Information carrier	Dialogue + performance + plot	Narration/commentary + on-screen annotations + subtitles
Sound structure	Dialogue-dominant (≥70%), VO/OS ≤30%, capped at 60~80 characters/minute	Narration is the main body (80~100%), dialogue can be zero; character count is calculated separately according to commentary speech rate
Characters	Continuous character performance, cross-segment consistency, FACS expressions	May have no fixed characters; people appear as "historical images/illustrated images/video footage", no continuous performance required
Storyboard logic	Plot beats → shot length	Narration sentence → image: one narration sentence paired with 1~2 images
Dialogue field	`台词：【角色（语气）："…"】` locks lip sync	Commentary usesoff-screen VO, does not lock lip sync; only when dialogue appears (interview/historical source quotation) is the dialogue field used
Visual materials	Scene asset images + character 4View + props	Period scenes + diagrams/maps/graphic-text cards + illustrated characters + documentary footage (annotated)
Asset consistency	Cross-shot character consistency is a hard constraint	Scene style consistency > character consistency; characters may appear in a single shot
Quality gate	First-shot anchoring, placement must enter frame, character ledger	Audio-visual information gap check, narration sentence-image mapping, historical/popular science accuracy, punchline ending
Viral formula	Conflict→reversal→satisfying point→emotional closure	Hook→breakdown/narrative→reveal/Wow→elevation(see §五 for details)

Judgment signals (write into track-routing §二): commentary / film breakdown / listicle / frame-by-frame analysis / talking-head film review / historical figure chronicle / historical events / popular science / trivia / knowledge niche / "understand it all in one go" / documentary-style narration.

五、The three major subtypes and corresponding viral formulas

A. Historical figure chronicle (character introduction)

Emotional arc model（0100%): Contrast persona hook (010%) → Origins and hardship (1030%) → Key turning point/highlight (3065%) → Downfall/ending (6585%) → Historical evaluation and golden-quote elevation (85100%）。
Underlying viral narrative formula：极致人设反差 + 出身逆境/微光 + 2~3 个关键转折 + 命运结局 + 后世评价金句
Hook examples: "At 48 he was still a rogue, at 54 he became emperor"; "He fought only one battle in his life, yet kept the Xiongnu from venturing south for ten years."

B. Historical events/battles/dynasties

Emotional arc model: Suspense result front-loaded (010%) → Background and causes (1035%) → Process unraveled layer by layer (3570%) → Key figures' decisions (7085%) → Impact and legacy (85~100%).
Underlying viral narrative formula：结果悬念前置 + 因果链逐层拆解 + 关键决策点 + 影响/遗产升华
Hook example: "A single treaty beneath the walls of Chanzhou kept the two nations at peace for one hundred and twenty years—but the emperor who signed it almost didn't dare cross the river." (Sentence pattern for demonstration only,historical facts subject to sources）

C. Popular science/knowledge explainer

Emotional arc model: Counterintuitive/pain-point hook (010%) → Entry via everyday scenario (1025%) → Knowledge breakdown in 35 steps (2570%）→ Wow Moment（7085%) → Value elevation/life application (85100%）。
Underlying viral narrative formula：反常识结论 + 身边事切入 + 类比拆解(3~5步) + Wow时刻 + 应用升华
Hook example: "Your phone battery not lasting long may not be the battery's fault—it's that you're charging it the wrong way."

Three-layer structure narrative (recommended skeleton for history content): Macro environment (international situation/economy) → National level (practical considerations/response) → Individual level (figures' decisions/memoirs). Suitable for grounding grand history in "people."

VI. Voiceover script (VO Script) writing standards (core)

The "lines" of explainer narrative content arevoiceover, completely different from dramatic dialogue:

Speech rate standard: Explainer voiceover baseline 4.5~6.5 字/秒(faster than dramatic dialogue at 3.5~6, because there is no lip-sync constraint and information density is the selling point); key golden quotes/climax lines slowed to 3~4 characters/secondto leave room for resonance;Shot length = voiceover character count ÷ speech rate + visual buffer 0.3~0.8s。
Sentence pattern iron rules: Primarily short sentences (single sentence ≤25 characters); one information point per sentence; make good use of rhetorical questions ("Why?"), parallelism, and antithesis;no written-style tone("This event had a profound impact on later generations" → "This thing changed the course of the next few hundred years").
Analogy first: Unfamiliar concepts must be analogized (Popular Science China standard: "A black hole is like a cosmic drain—even light can't escape it").
Burden-reduction principle(Yushifang standard):Names: if they can be omitted, omit them(use "that general defending the city" instead of obscure names); for place names/official titles/years, keep only the 2~3 key ones that affect understanding; numbers must be made concrete ("donated 350,000" → "equivalent to him pedaling a tricycle around the Earth X times"——number conversions must be traceable）。
Hook writing(first three sentences): conflict style / result front-loaded / suspense style / counterintuitive style, choose one,≤15 characters/sentence, landing within 0~3s.
Punchline ending: The ending must have 1 screenshot-shareable elevated line(same style as the black-background golden line), echoing the opening hook.
Sound-image information gap discipline (VO rules for commentary version): Narration is responsible for "explaining cause and effect/explaining patterns/explaining characters' inner thoughts", while the visuals are responsible for "providing evidence/providing scenes/providing process demonstrations"——Visuals are allowed to coordinate with narration (visuals are annotation, not restatement), butnarration is prohibited from reading aloud content already written in large text in the visuals(subtitles and narration do not repeat);if the narration is deleted, the information content of the visuals must not decrease。

With master control short-drama-director/references/vo-os-weaving.md relationship: Section VI of this engine is thespecialized specificationfor the commentary category; among the master control VO three disciplines, the "rigid length checkpoint" and the "five types of legitimate uses"do not apply to the commentary category(in the commentary category, narration is the main body), but the **core (do not restate the visuals)** of the "VO cold scrutiny rule" must be retained, and has been implemented in §VI.7 as the "sound-image information gap discipline".

VII. Visual and material specifications (AI production adaptation)

Era scenes take priority over characters: For historical subjects, use "era empty shots + artifacts + architecture + clothing" to establish a credible sense of period; each episode has 1 set of scene style master templates (color temperature/film grain/clothing style locked),scene style consistency is the first asset constraint for the commentary category。
Character appearance method: illustration/historical painting style characters (avoid the red line of realistic live-action facesand the controversy of "whether they look like the person"); the same character appears in ≤2 image versions(frontal portrait + full-body portrait) is sufficient,no continuous performance across shots。
Diagrams/maps/graphic-text cards: For war progression, routes, and data comparisons, use animated maps/diagrams/graphic-text cards (can be AI-generated or done in post);large-title/golden-line cards are standard for the commentary category("black-background golden line" retained).
Documentary footage/historical materials: When citing real photos/footage, on the visualsmark the source(continuing sd-recon-engine source-tier discipline,T1 historical materials can be directly cited on screen, T2 discussions are only orally paraphrased）。
Prevent AI hallucination of modern elements: Historical scene promptsmust carry era anchors(year + light source type + clothing + architecture); first produce a static reference frame to lock the art style before motion (continuing asset-first-pipeline 's P2 assets precede storyboards).
Material forms (M0/M1/M2) apply as usual: This package does not hold images——material form is determined by master control short-drama-director/references/asset-delivery-modes.md , and the commentary category is no exception.

VIII. Storyboard and feeding adaptation (commentary version)

Storyboard unit = narration sentence: each narration sentence → 1~2 visuals; shot length is back-calculated from word count (§VI.1).
Seven-segment feeding writing method: Commentary script goes in the 声音： segment (off-screen·narration (VO)), and the dialogue field is 台词：无。；no lip-sync locking, no FACS expression writing; the visual segment describes scenes/objects/process demonstrations, and posture is written only when a character appears.
⚠️ C14 Voice-over quota exemption (the most critical clause of this engine · must be explicitly declared): The commentary category isfull-film VO, and the master control's short-drama-director/references/vo-os-weaving.md of the "VO+OS ≤30%, 60~80 characters/minute" checkpoint does not apply to track ③, replaced by the commentary speech rate in §6.1 and the audio-visual information gap discipline in §6.7.validate_prompt.py must be routed by track—the output machine check adds --track 3: for track ③ scripts,wholly exempt the three C14 length WARNs(OS shot count / VO shot count / VO+OS character count),retain the three C14 FAIL red lines(audio-visual separation / written into the dialogue field / use of {} characters), and reclassify as C14n narration-picture mapping. False positives would invalidate the entire package's machine check, so it must be explicitly declared. The sole authority for the exemption criteria is this engine; the machine-check implementation resides in the master control scripts/validate_prompt.py(parameter --track 1|2|3|4), and the user manual is at sd-qc-engine/references/machine-check-manual.md。
Length and speech rate are semantic items and are not judged by machine check: The commentary script's character count/speech rate (4.5~6.5 字/秒, single sentence ≤25 characters) andthe existence of a punchline ending、audio-visual information gap(narration does not restate the picture) are subject to sd-qc-engine manual/AI semantic review and are not written as machine-check assertions—the machine check only backstops the one mechanically determinable item, C14n narration-picture mapping.
Version adaptation: The commentary category has no dialogue—2.5 uses official sound-effect characters (<>); 2.0 writes sound effects as textual sound descriptions;ambient sound / historical atmosphere sound is written per the version contract。
Segment seams (D lock) apply as usual: Multi-segment commentary tasks still connect via the D lock three-choice;do not write a continuation tail line at the end of a segment([Continuation Status] wholly abolished across the package as of 2026-10-09; machine check C11 is reversed to residual interception).

IX. Rule Library Index (references/)

File	Content	When to read
references/narration-vo-script.md	Narration Script Writing Standards (core): commentary speech rate criteria, sentence pattern iron rules, analogy priority, burden-reduction principle, five hook formulas, punchline ending, audio-visual information gap discipline, VO Script structure template, seven-segment feeding writing method	Read through every time you write a narration script
references/history-visual-style.md	Period Scene and Historical Visual Standards: scene style master template, four elements of era anchoring, illustration character entrance methods, diagrams/maps/graphic-text cards, historical source citation labeling, prevention of AI hallucination of modern elements	When defining scenes and materials
references/explainer-formulas.md	Three Major Subtype Viral Formulas and Emotional Arcs: historical figure biography / historical event and battle / popular science knowledge commentary + four-module decomposition structure reuse + hook sentence pattern library	When defining subtype and structure

Read on demand: Read only the files needed for this task. The narration script standards are read for almost every case; the historical visual standards are read only for historical subject matter (popular science may read the generalized period scene portion); viral formulas are read in the corresponding subsection by subtype.

X. Cross-Stage Handoff (the main output of this engine)

When locking a script or passing across stages,mustcarry:

赛道: ③ 解说类
子类型: 历史人物志 | 历史事件战役 | 科普知识解说
旁白稿: VO Script（每句旁白的镜长预算已倒推）
场景风格母版: 色温 / 胶片颗粒 / 服饰款式（跨集复用）
信源: T1 史料清单 / T2 讨论清单（沿用 sd-recon-engine 分级）
C14 豁免: 已声明赛道③ → validate_prompt 不触发 VO 篇幅告警
金句: 结尾升华句（与开篇钩子呼应的那一句）

Downstream discipline：

sd-asset-engine：Scene style consistency > character consistency; characters serve as "illustrative figures" rather than "cross-shot continuous performance assets".
sd-storyboard-engine: the storyboard unit = narration sentence; each narration sentence → 1–2 frames; do not write FACS facial expressions.
sd-feed-engine: the commentary goes in the 声音： segment, with the dialogue field 台词：无。; the C14 exemption is passed along with the draft.
sd-qc-engine：Reclassify under track ③ C14(do not judge the VO length as exceeding quota); verify the narration sentence-to-frame mapping and the existence of the punchline ending.

XI. Prohibited Boundary Crossings

Do notreplace the six-stage chain of sd-script-engine ,do notestablish a separate script process (this engine only governs "how the commentary genre uses this chain");
Do notestablish a separate feeding format, asset numbering, or machine-check script (authority lies with master control);
Do notwrite the commentary genre's "full-film VO" standard back to contaminate the narrative track (the narrative track's VO/OS ≤30% checkpoint remains unchanged);
Do notfabricate historical facts in historical/science topics—the source-tiering discipline (T1/T2) is enforced as usual, and T2 discussions must not be elevated to fact;
Do notuse photorealistic real-person faces for historical figures (the red-line boundary is the same as master control);
Do notcross track boundaries—drama series / overseas drama tasks are not subject to this engine.