Ghostlight/notes/ghostlight-implementation-plan.md
2026-05-01 22:17:15 +01:00

4.6 KiB

Ghostlight Implementation Plan

Near-Term Sequence

  1. Freeze the first canonical state model. In progress.

    • Define the latent variable families.
    • Define first-class dimensions, including volatility, attachment-seeking, and distance-seeking.
    • Define canonical state versus perceived state.
    • Keep authoring outputs explainable from the state, not just generated from stylish prompt fog.
    • Maintain schemas/agent-state.schema.json, schemas/agent-state.required-fields.json, and the example fixtures as the first executable contract.
  2. Freeze the first event and classifier schema.

    • Define event labels.
    • Define canonical expression labels.
    • Define listener perception labels.
    • Define mask or presentation labels.
    • Define confidence, uncertainty, and evidence accumulation rules.
  3. Build the prompt projection layer for authoring.

    • Use docs/architecture/prompt-projection-contract.md as the first renderer contract.
    • Use docs/architecture/schema-future-mechanisms.md to keep appraisal, transition, belief, action, conversation, resource, dyad, institution, memory, and embodiment needs visible without prematurely bloating the v0 agent-state schema.
    • Render scene and dialogue scaffolds from structured state, relevant memories, scene pressure, cultural defaults, and perceived state.
    • Keep the prompt renderer a projection surface instead of a second brain.
    • Make generated moves inspectable: why this character deflects, confesses, threatens, softens, bargains, or lies.
  4. Build the projection distillation loop. Started.

    • Use docs/architecture/projection-distillation-plan.md as the teacher to student roadmap.
    • Use docs/architecture/aetheria-lore-grounding-architecture.md to keep Aetheria grounded training separate from procedural Elysium branch generation.
    • Maintain schemas/projection-example.schema.json, examples/projections/, and tools/validate_projection_examples.py as the first reviewed artifact seam.
    • Use authored historical Aetheria flashpoints, not malleable Elysium gameplay outcomes, as the first grounded lore-training corpus.
    • Maintain schemas/lore-grounding-digest.schema.json, examples/lore-grounding/, and tools/validate_lore_grounding.py as the cultural and factional pressure seam for grounded fixtures.
    • Use examples/lore-grounding/cold-wake-panic.ganymede-corridor.json as the first draft historical grounding digest.
    • Build the first grounded projection fixture from one specific Cold Wake room rather than from the whole crisis at once.
    • Use docs/experiments/cold-wake-story-lab.md and examples/agent-state.cold-wake-story-lab.json as the first story-shaped writing lab.
    • Treat projection outputs as response turns, not dialogue-only turns: characters may speak, act, withdraw, stay silent, use violence, or combine speech and action when state and scene pressure justify it.
    • Use a frontier teacher model to produce reviewed projection artifacts from structured state and lore grounding digests.
    • Save accepted and rejected projection examples as supervised data.
    • Train or adapt a smaller student projector only after the artifact schema, input slicer, and evaluator are stable.
  5. Build the first drama-scaffolding loop.

    • memory updates
    • relationship updates
    • goal pressure
    • situational activation
    • cultural and institutional pressure
    • conflict beats derived from incompatible goals and values
  6. Build the first Aetheria authoring consumer.

    • use the Call of the Void slice described in AetheriaLore as the first concrete content-generation context
    • treat Call of the Void and other Elysium gameplay-era fixtures as procedural branch material unless specific outcomes are explicitly promoted
    • support dialogue scaffolding and procedural drama for narrated stories across the Aetheria timeline
    • do not treat Cold Wake as the implementation target unless explicitly revived later; it is currently training feedstock

Discipline

  • Do not start with an MMO and call it ambition.
  • Do not train a classifier before the label schema is stable enough to deserve the trouble.
  • Do not let prompt-writing outrun the structured state model.
  • Do not fine-tune a projector before the projection artifact schema, input slicer, and evaluator have stopped sliding around.
  • Do not let elegant theory notes multiply faster than implementation seams.
  • Do not accidentally rebuild a game slice when the requested consumer is an authoring and story-generation organ.