{ "candidate_mainlines": [ { "allowed_work": [ "Prompt packet preparation and formatting.", "Ad-hoc manual QA and consistency checks (non-provider).", "Execution log bookkeeping.", "DeepSeek advisory reviews (expert opinion only)." ], "blocked_work": [ "Any provider/manual media generation.", "Production lane acceptance.", "Canonical KG / raw KG mutation.", "Overlay writeback, feedback loops, replay.", "Outputs index mutation." ], "deepseek_consultant_role": [ "Review prompt drafts for completeness, constraint coverage, and adherence to formula.", "Flag missing asset references or ambiguous entities.", "Provide suggestions logged separately—never as acceptance or evidence." ], "gate_criteria": [ "All prompts in batch pass visual/logical QA checklist without provider invocation.", "Execution log is complete and matches batch manifest." ], "id": "Now-32", "name": "Prompt Readiness QA & Manual Execution Log Contract (Active)", "objective": "Establish an evidence-gated, provider-free QA contract for platform-neutral prompt readiness, recording manual review and execution logs. Serves as the anchor gate post-P计划.", "pipeline_advantage_to_unlock": [ "Basic prompt validation and human-in-the-loop logging.", "Establishes the pattern of evidence gating before any provider execution." ], "required_artifacts": [ "Prompt readiness QA report per batch.", "Manual execution log entries (timestamped, operator-annotated).", "DeepSeek advisory log (non-binding)." ] }, { "allowed_work": [ "Running entity extraction/linking against existing draft rows (using KG model).", "Augmenting prompt packets with `entity_refs`, `asset_refs`, and `continuity_keys` metadata.", "Manual/assisted validation of entity links (human + DeepSeek).", "Schema updates to prompt packet model to carry link data." ], "blocked_work": [ "Mutating the canonical KG based on this pass (read-only linking).", "Overwriting accepted overlays or production indices.", "Any provider execution or media generation." ], "deepseek_consultant_role": [ "Assist in entity disambiguation and suggest canonical entity matches.", "Review asset binding choices for contextual fit (e.g., does the prop reference match the scene description?).", "Produce a `linking_advisory_log`—not evidence." ], "gate_criteria": [ ">=90% of non-held rows carry at least one entity link and, where applicable, one reference asset binding.", "All 240 held rows are flagged for linkage feasibility (manual triage).", "No mutation of production-grade data stores." ], "id": "Now-33", "name": "KG/Entity-Link & Asset Binding Strengthening", "objective": "Retrofit existing draft rows with explicit KG entity links (characters, locations, props, motifs) and bind reference assets (images, audio, video references) into the prompt packets, creating a traceable production graph.", "pipeline_advantage_to_unlock": [ "Persistent entity identifiers across shots and chapters, enabling later cross-shot continuity verification.", "Asset binding allows downstream provider prompts to reference the exact visual/audio resources, a capability direct text generation cannot offer.", "Establishes a canonical asset/item lineage that can be validated before provider dispatch." ], "required_artifacts": [ "Entity-link manifest per chapter/scene (JSON).", "Asset binding registry (which reference asset applies to which shots).", "Updated prompt packet schema with link/asset slots.", "Validation report: link coverage, unresolved entities." ] }, { "allowed_work": [ "Schema design: define `VideoPromptPacket` fields for camera scale/angle/motion, transition type, asset references, audio/SFX cues, negative constraints, timing windows.", "Develop industry-gate validators (syntax, completeness, constraint coverage).", "Upgrade prompt readiness QA checklist to include industry-benchmark items.", "Retrofit approved draft rows to new schema (as a migration pass, without altering content)." ], "blocked_work": [ "Any provider execution (the new schema is a preparation step only).", "Enforcing aesthetic or stylistic judgments as gate failures—only structural/compliance checks." ], "deepseek_consultant_role": [ "Compare schema design against industry documentation and identify gaps.", "Generate adversarial prompt examples to stress-test validators.", "Provide a `schema_assessment_log`; decisions remain with operator." ], "gate_criteria": [ "All ready/non-held rows conform to new schema with zero structural violations.", "Validator passes a curated benchmark set derived from industry sample prompts.", "DeepSeek review confirms schema captures critical industry design patterns." ], "id": "Now-34", "name": "Industry-Benchmark Schema & Gate Upgrade", "objective": "Translate constraints from production video generation guidelines (e.g., Seedance enterprise guide, reference PV production bible) into formal schema requirements, prompt structure, gate checks, and QA rubrics.", "pipeline_advantage_to_unlock": [ "Guarantees that every prompt packet is structurally compatible with best-practice provider formulas, including multi-shot choreography, transitions, reference assets, and negative constraints.", "Enables automated validation that the pipeline output is not just prose, but a ready-to-consume production blueprint (time-coded shots, asset references, audio/SFX cues).", "Creates a durable competitive moat: competitors' direct generation would need to replicate these structured constraints from scratch." ], "required_artifacts": [ "`VideoPromptPacket` schema v2 (JSON Schema / type definitions).", "Industry-benchmark gate validator test suite.", "Migration report: which rows pass/fail the new schema.", "Updated QA checklist aligned to industry formula." ] }, { "allowed_work": [ "Build automated review harness: feed prompt packets to DeepSeek, collect scored reviews (completeness, consistency, constraint adherence).", "Store review artifacts in a dedicated `advisory/` path, strictly partitioned from evidence or production indices.", "Use review signals to prioritize held-row repair and entity-link gaps.", "Calibrate review rubrics based on industry benchmarks (Now-34 outputs)." ], "blocked_work": [ "Using DeepSeek scores to auto-accept prompts or mutate evidence state.", "Allowing DeepSeek to initiate any write to accepted overlays or KG.", "Any provider execution; reviews are advisory, not execution triggers." ], "deepseek_consultant_role": [ "Act as an automated reviewer only; output is advisory and labeled as such.", "Provide paragraph-level feedback on missing constraints, unlinked entities, or continuity gaps.", "Role boundaries enforced technically (write permissions, data partitioning)." ], "gate_criteria": [ "Review harness operates on all ready rows and produces consistent advisory output.", "Inter-rater comparison between DeepSeek and human QA shows alignment on structural checks (≥85% on objective criteria).", "No evidence mutation or production index writes from the review pipeline." ], "id": "Now-35", "name": "DeepSeek Consultant-as-Reviewer Systematic Integration", "objective": "Operationalize DeepSeek v4 pro thinking/max as a standing, non-voting reviewer that produces scored advisory reports on prompt readiness, entity consistency, asset binding, and industry compliance, without contaminating evidence or acceptance records.", "pipeline_advantage_to_unlock": [ "High-quality, scalable expert inspection that direct generation cannot self-perform (self-critique of generated text is unreliable).", "Creates a defendable human+AI review trail: operators decide based on DeepSeek reports and human judgment, maintaining evidence integrity.", "Enables continuous improvement feedback for schema, linkers, and QA rules without executing providers." ], "required_artifacts": [ "Review harness configuration (prompt templates, scoring dimensions).", "Advisory review log records (timestamped, operator-annotatable).", "Review summary dashboard (for human operators).", "Calibration report: DeepSeek review agreement with manual QA on a sample." ] }, { "allowed_work": [ "Compile all artifacts from Now-32 through Now-35 into a versioned evidence pack.", "Generate a readiness manifest that summarizes gate statuses, entity/asset coverage, held-row disposition, and reviewer scores.", "Crypto-sign or hash the evidence pack for tamper evidence.", "Operator sign-off (human) on the readiness manifest." ], "blocked_work": [ "Any provider execution—assembly only.", "Triggering downstream execution automatically; an explicit new contract is required." ], "deepseek_consultant_role": [ "Review the evidence pack for completeness and internal consistency.", "Provide a final advisory opinion on readiness, clearly marked as non-binding.", "Do not sign or accept the readiness manifest." ], "gate_criteria": [ "All prior gate criteria (Now-32..35) are satisfied and recorded.", "Readiness manifest is signed by a designated human operator.", "Evidence pack is stored immutably and referenced by the next entrypoint when execution is authorized." ], "id": "Now-36", "name": "Production Readiness Evidence Pack Assembly", "objective": "Assemble a comprehensive, self-contained evidence pack that demonstrates readiness for provider dispatch, including all artifacts, gate reports, advisory logs, and a signed-off readiness manifest—without executing a single provider call.", "pipeline_advantage_to_unlock": [ "Converts the pipeline's traceability (KG links, asset bindings, schema compliance, review trail) into a verifiable evidence artifact that can be audited by any stakeholder.", "Direct generation approaches cannot produce this auditable chain of production readiness; they produce only text and hope.", "Establishes a clear contract boundary: once evidence pack is accepted, provider execution can begin under a separate, explicit execution contract." ], "required_artifacts": [ "Evidence pack bundle (directory or archive), hash-tree manifest.", "Readiness manifest (JSON) with operator signature.", "Gap analysis: what remains unresolved (held rows, missing assets) and suggested handling." ] } ], "confidence": "high", "deepseek_operating_model": [ "DeepSeek v4 pro thinking/max operates as a third-party advisory consultant, never as an evidence source, acceptance authority, or execution trigger.", "All DeepSeek outputs are stored in a dedicated advisory log partitioned from production evidence, acceptance records, and canonical data stores.", "DeepSeek reviews focus on structural completeness, consistency, constraint adherence, and entity/asset linkage quality; it may also assist in creative brainstorming under explicit operator direction.", "DeepSeek recommendations must be reviewed and optionally acted upon by a human operator; automated adoption is prohibited.", "The model may be used for scoring/ranking, gap analysis, and inter-rater calibration, but scores are never used as gate criteria unless explicitly incorporated into operator-defined checklists that include human verification.", "A clear technical boundary (e.g., write permissions, separate storage path) enforces this separation." ], "do_not_do": [ "Do not advance to any candidate mainline execution. Keep all mainlines in definition state.", "Do not activate provider execution, manual evidence, feedback loops, or canonical overlay writeback.", "Do not treat DeepSeek advisory output as production acceptance or evidence.", "Do not mutate the next_entrypoint field or close the Now-32 active gate.", "Do not initiate any work on the 240 held rows beyond analysis and linking feasibility tagging without an explicit repair contract.", "Do not generate media, call external provider APIs, or trigger any provider-dependent workflow.", "Do not create new markdown artifacts unless they serve as a contract, retrospective, decision record, report, or handoff as per rules.", "Do not assume that schema or gate upgrades automatically improve creative quality—validate against benchmarks." ], "industry_benchmark_requirements": [ "Prompt formula must include: subject + action + camera language + reference assets + style/aesthetics + audio/SFX + restrictions.", "Reference assets (images, videos, audio) must be explicitly addressable and reusable across generation, editing, extension, and replacement tasks.", "Provider-facing prompts should support complex multi-shot camera choreography, transitions, physical realism, micro-expressions, sound synchronization, and story continuity.", "Story generation includes forward/backward extension and stitching between shots as first-class tasks.", "Negative constraints and safety/legal boundaries around human likeness/reference assets must be carried as explicit control data.", "Production documents should follow a layered structure: high-level style/tempo premise, scene/dialogue writing, and time-coded shot-by-shot execution plan with duration windows, picture/action, dialogue, subtitles, sound effects, music, action rhythm, and title cards.", "Character, prop, monster, and scene settings should include asset images; the production package must treat visual references as part of the script bundle, not afterthoughts." ], "overall_verdict": "After P计划 closeout, the active Now-32 evidence gate must persist. Candidate mainlines Now-33 through Now-36 should sequentially unlock KG/entity-link asset binding, industry benchmark schema integration, DeepSeek consultant review, and production evidence pack assembly—all without authorizing provider execution. The irreversible advantage of the pipeline lies in a traceable production graph that anchors every shot to stable KG entities, reference assets, and control constraints; direct generation cannot replicate this multi-shot, asset-aware consistency. DeepSeek must operate purely as a consultant and automated reviewer, never substituting for production acceptance or evidence. All provider/manual/feedback execution remains blocked until an explicit downstream evidence contract is activated.", "parse_status": "pass", "pipeline_advantage_model": [ "The production pipeline's unique advantage is a traceable, verifiable production graph built on KG/entity links and asset bindings. This graph can persist across shots, chapters, and provider invocations, ensuring continuity and enabling asset reuse.", "Unlike direct generative models, the pipeline can explicitly reference and validate assets (images, audio, video), enforce negative constraints, and carry time-coded shot plans—all in a structured schema that is machine-validated before provider dispatch.", "This model creates an industrial-grade preparation layer that is defensible, auditable, and repeatable; direct generation lacks the governance and persistence to compete at scale.", "DeepSeek augments this model as an advisory reviewer, not a creator, further strengthening the human+tool partnership without compromising evidence integrity." ], "recommended_control_plane_updates": [ "Maintain `next_entrypoint` as `platform_neutral_prompt_readiness_manual_execution_log_contract` and do not change it; candidate mainlines are defined but not activated.", "Introduce a `candidate_mainlines` registry in the control plane configuration, listing Now-33 through Now-36 with their entry conditions (explicit evidence contracts) and gate criteria.", "Add a new artifact classification: `advisory/` for DeepSeek outputs and consultant logs, strictly separated from `evidence/` and `production/`.", "Update workflow index classification rules to include the candidate mainlines as future routes, each requiring an explicit activation contract.", "Define a `production_readiness_evidence_pack` storage location and schema, to be populated by Now-36.", "Extend the blocked-work list to enforce that no provider execution, feedback ingestion, or canonical KG mutation can occur without a signed execution contract.", "Implement a `wait` state for each candidate mainline: all work items are defined but pending activation." ], "risks_and_countermeasures": [ "Risk: DeepSeek outputs are mistaken for acceptance or evidence, leading to premature confidence. Countermeasure: enforce strict data partitioning, label all DeepSeek artifacts as 'advisory', and require human sign-off for any gate decision referencing DeepSeek input.", "Risk: Entity links and asset bindings in Now-33 are incomplete or incorrect, weakening the production graph. Countermeasure: set quantitative coverage gates, conduct manual sampling audits, and use DeepSeek to flag anomalies.", "Risk: Industry benchmark requirements (Now-34) over-constrain creative flexibility, making prompts formulaic or rigid. Countermeasure: treat benchmarks as structural boundaries, not aesthetic mandates; incorporate operator override mechanisms.", "Risk: Now-36 evidence pack assembly creates an illusion of full readiness while 240 held rows and chapter 18 remain unresolved. Countermeasure: the readiness manifest must explicitly surface all known gaps and their disposition.", "Risk: Accidental provider execution due to misconfiguration. Countermeasure: maintain a hard block in the orchestration layer that requires an explicit execution contract token to invoke any provider adapter.", "Risk: Candidate mainline descriptions create expectation of immediate activation, bypassing the Now-32 gate. Countermeasure: clearly document that all candidate mainlines are inactive until a separate authorization contract is executed." ], "strategic_diagnosis": [ "P计划 closure leaves a clean gate at Now-32, but the pipeline's true differentiator (KG/entity/asset traceability) is underexploited in current artifacts.", "Direct generative models can produce creative prose but cannot guarantee persistent entity identity, asset reuse, or cross-shot continuity—this is the pipeline's lock-in moat.", "Industry benchmarks (Doubao-Seedance guide, animated short PV production bible) make it clear that production-grade video prompts require a structured formula: subject + action + camera language + reference assets + style/aesthetics + audio/SFX + restrictions, all traceable across tasks.", "DeepSeek v4 pro thinking/max can act as a powerful on-demand expert, but without strict scoping it risks becoming pseudo-acceptance, eroding the evidence gate.", "The 240 held rows and chapter 18 held-only status signal that QA/repair must be addressable via enhanced tooling, not by bypassing the gate.", "Future mainlines should be defined and required artifacts and gate criteria should be specified now, so that when an explicit contract is authorized, the pipeline can advance without re-planning." ] }