112 lines
7.1 KiB
YAML
112 lines
7.1 KiB
YAML
experiment:
|
|
name: Ghostlight
|
|
status: incubation
|
|
phase: phase_0_scaffold
|
|
premise: >
|
|
Build a durable research and implementation home for socially persistent
|
|
generative agents whose behavior emerges from personality, culture,
|
|
memory, perception, incentives, and situation rather than flat persona
|
|
prompts.
|
|
|
|
mission:
|
|
primary: >
|
|
Generate psychologically grounded branching scenes, dialogue, actions,
|
|
social consequences, and eventually storylines, consumer behavior,
|
|
faction decisions, and technology/item manifests for Aetheria.
|
|
near_term: >
|
|
Produce sandboxed, source-grounded responder training data through
|
|
projected local context, responder packets, reviewed outputs, and audited
|
|
mutation receipts.
|
|
active_fixture: >
|
|
Cold Wake is a pre-Elysium historical training fixture, not the product
|
|
target. Call of the Void and Elysium-era gameplay remain procedural
|
|
consumer targets.
|
|
|
|
hard_boundaries:
|
|
- Prompt text is projection, not truth storage.
|
|
- Canonical agent state and per-agent perceived state must remain distinct.
|
|
- Characters act from local knowledge, local perception, local pressure, and bounded affordances.
|
|
- Listener misread is first-class; intent and interpretation are not the same object.
|
|
- Generated action is an event proposal; deterministic gates and reviewed mutation decide what becomes state.
|
|
- Fuzzy judgments become reviewed training artifacts before they become trained models.
|
|
- Aetheria generation is source-first; check lore for factions, bodies, locations, institutions, and constraints before projection.
|
|
- Elysium futures are branch-local canon indexed by lineage, not single-history Sol facts.
|
|
- Technology exploration must emit game-usable item, component, assembly, supply-chain, compatibility, and provenance data mapped toward Aetheria-Economy.
|
|
- Archived local-model prototype runners are receipts and plumbing tests, not the gold-data path.
|
|
|
|
live_architecture:
|
|
state_model:
|
|
- Layer personality through organization, disposition, behavior, presentation, situational activation, relationships, goals, and values.
|
|
- Canonical authored state has no certainty; perceived overlays carry observed activation, attributed disposition, and confidence.
|
|
- Voice style is a broad communication vector; defensive verbal behavior should emerge from state, pressure, relationship, and presentation.
|
|
projection:
|
|
- Project canonical state into character-local operating context before any responder sees it.
|
|
- Include scene facts, relevant memory, relationship read, stakes, embodiment/interface affordances, action affordances, projection controls, retrieved lore, and latent pressure handling.
|
|
- Hide raw state internals such as means, plasticity, current_activation, and numeric dimensions from prompt text.
|
|
responder_packets:
|
|
- `packet_only` tests runtime parity from curated source excerpts.
|
|
- `retrieval_augmented` bakes setting texture into reviewed outputs through scoped retrieval.
|
|
- `responder_scoped_repository_search` allows the responder to inspect only declared lore scope and requires consulted refs plus research summary.
|
|
- Gold responder data must preserve exact visible input, raw output, parsed/reviewed output, hidden-context refs, leakage audit, isolation method, and coordinator interventions.
|
|
coordinator:
|
|
- The coordinator is a trainable soft organ: it stages scenes, chooses next beats and acting agents, invokes specialist machinery, carries unresolved hooks, emits glue prose, and proposes game-engine-shaped state changes.
|
|
- Glue prose is training signal, not canonical truth.
|
|
training_targets:
|
|
- Coordinator/story runtime
|
|
- Memory and lore retrieval
|
|
- Projector/state interpreter
|
|
- Character agent/responder
|
|
- Participant appraiser
|
|
- State mutator
|
|
- Relationship/perception updater
|
|
- Evaluator/review classifiers
|
|
- Aetheria-tuned responder
|
|
- Institution, faction, and consumer decision models
|
|
|
|
active_subgoals:
|
|
- Keep Ghostlight focused on the architecture and data pipeline for socially persistent agents.
|
|
- Build schema-stable, sandboxed responder data before chasing larger runtime loops.
|
|
- Preserve Aetheria grounding through source-backed lore, branch labels, and reviewed assumptions.
|
|
- Shape every soft judgment as future training data, not permanent prompt folklore.
|
|
- Keep item/technology exploration database-shaped for Aetheria-Economy instead of prose-only worldbuilding.
|
|
|
|
current_artifacts:
|
|
canonical_docs:
|
|
- docs/architecture/prompt-projection-contract.md
|
|
- docs/architecture/projected-local-context.md
|
|
- docs/architecture/sequential-agent-runtime.md
|
|
- docs/architecture/ink-branching-scenes.md
|
|
- docs/architecture/aetheria-lore-grounding-architecture.md
|
|
- docs/architecture/sandboxed-responder-packets.md
|
|
- docs/architecture/soft-model-training-artifacts.md
|
|
- docs/architecture/training-plan.md
|
|
- docs/architecture/technology-item-manifest-plan.md
|
|
schemas:
|
|
- schemas/agent-state.schema.json
|
|
- schemas/projection-example.schema.json
|
|
- schemas/lore-grounding-digest.schema.json
|
|
- schemas/projected-local-context.schema.json
|
|
- schemas/coordinator-artifact.schema.json
|
|
- schemas/responder-packet.schema.json
|
|
- schemas/responder-output.schema.json
|
|
live_examples:
|
|
- examples/lore-grounding/cold-wake-panic.ganymede-corridor.json
|
|
- examples/projected-contexts/scene-02-sanctuary-intake.sella_ren.projected-context.json
|
|
- examples/coordinator/cold-wake-sanctuary-intake.v0.json
|
|
- examples/responder-packets/scene-02-sanctuary-intake.sella_ren.packet.research.v0.json
|
|
- experiments/responder-packets/cold-wake-sanctuary-intake-sella-v0.capture.json
|
|
- experiments/responder-packets/cold-wake-sanctuary-intake-sella-v0.mutation.json
|
|
archived_receipts:
|
|
- Archived local-model prototype captures remain under experiments/ and examples/ink/ for validation and lessons, not active steering.
|
|
|
|
data_flow:
|
|
- Rehydrate from map, handoff, system map, implementation plan, and `npm run state:status`.
|
|
- Work one bounded organ.
|
|
- Verify the seam that matters.
|
|
- Add distilled evidence only when future belief changes.
|
|
- Commit and push completed work.
|
|
- If state grows faster than understanding, cut it back.
|
|
|
|
current_status:
|
|
summary: Ghostlight is focused on sandboxed, source-grounded responder and coordinator training artifacts for socially persistent Aetheria agents, with archived prototype runs demoted to receipts and the live path centered on projected local context, responder packets, reviewed outputs, and audited state mutation.
|
|
next_action: Run a sandboxed no-fork research-enabled responder against `examples/responder-packets/scene-02-sanctuary-intake.sella_ren.packet.research.v0.json`, allowing only the declared AetheriaLore scope, then capture exact visible input, consulted refs, research summary, raw output, review labels, leakage audit, and any coordinator intervention. Review for institutional lore use, latent pressure, prompt parroting, omniscience, and mission-critical trauma-dumping. Do not use archived local-model prototype runners for gold responder data.
|