Project Explorer

Files

Story files

Browse the notes, research, decisions, and workshops used to develop the series.

Search 577 documents about the world, characters, plot, research, decisions, workshops, and earlier ideas.

Current document 04 Research/Full Reports/38 - Graph Resolution and Renderer Packet Projection.md

Document metadata
type: research-report
status: advisory-non-canon
updated: 2026-08-13
research_sequence: 38-of-40
produced_by: OpenAI Codex (Sol)

Graph Resolution and Renderer Packet Projection

Purpose

This report defines the compiler boundary between a validated visual subgraph and the clean GPT Image 2 packet.

Executive finding

Graph resolution and prompt projection should be two separate phases with separately validated outputs. Resolution produces a complete scene model with source-bearing claims. Projection converts only renderer-relevant, observable values into concise image instructions.

JSON Schema is appropriate for structural validation of the graph, resolved-scene record, source trace, and packet manifest. SHACL provides a useful conceptual model for graph constraints and severity-bearing validation results. W3C PROV supports tracing the resolved scene and generated image as entities derived through compilation and rendering activities.

Resolution pipeline

scene intent
  -> required node set
  -> claim collection
  -> qualifier filtering
  -> conflict/supersession resolution
  -> temporal intersection
  -> inheritance and override
  -> visibility filtering
  -> completeness validation
  -> resolved scene model
  -> renderer projection

No renderer wording should be generated before completeness validation passes.

Resolved scene model

The model should contain typed fields, not prose:

  • participants and identity references;
  • appearance states and chronological/apparent ages;
  • time, era, place, and environment overlays;
  • wardrobe resolution;
  • action and relationship participation;
  • technology capabilities and visible consequences;
  • manifestation and viewpoint;
  • image type, composition, camera, style, and output settings;
  • negative constraints;
  • claim IDs supporting every field.

Merge semantics

Each field needs one declared merge behavior:

  • replace: scene-specific value replaces inherited default;
  • append: compatible constraints accumulate;
  • intersect: allowed ranges narrow;
  • prohibit: inherited value may not appear;
  • select-one: exactly one eligible value must resolve;
  • derived: computed from explicit inputs;
  • unresolved: blocks or requires waiver.

Free-form dictionary merging is forbidden because it can silently combine alternatives.

Projection rules

Renderer projection should:

  1. preserve identity-reference paths and purpose labels;
  2. translate structured values into concrete observable language;
  3. retain unresolved epistemic boundaries only as prohibitions against invention;
  4. remove claim IDs, file paths other than required image attachments, citations, confidence, QA, and alternatives;
  5. keep execution parameters in a separate execution-plan section;
  6. produce a packet fingerprint linked to the resolved-scene fingerprint.

The trace should map packet section and field back to resolved fields, claim IDs, graph nodes, and vault sources. This mapping remains external.

Validation severities

  • BLOCK: no packet emitted.
  • WARN: packet emitted with external diagnostic.
  • INFO: provenance or optimization note.

Severity does not alter story status. A warning is not permission to invent.

Failure modes

  • Graph IDs or QA language appearing in the image.
  • Prompt prose generated before conflicts are checked.
  • Undeclared merge behavior.
  • Negative constraints contradicting positive instructions.
  • File citations replacing actual visual detail.
  • Trace and packet fingerprints becoming unsynchronized.

V2 requirements

  • Define JSON Schemas for claims, graph, scene intent, resolved scene, trace, and packet manifest.
  • Implement an explicit resolution DAG and merge policy registry.
  • Validate resolved scene before projection.
  • Add field-level trace links and fingerprints.
  • Keep clean packet, execution plan, retrieval receipt, and QA as distinct artifacts.
  • Add tests proving diagnostic content cannot reach renderer projection.

Sources

Boundary

This report specifies the v2 compiler architecture but does not implement it.