liqiang
b119135836
Build latest book artifacts / build (push) Canceled after 0s
dependency resolution / resolve (3.11) (push) Canceled after 0s
dependency resolution / resolve (3.13) (push) Canceled after 0s
deploy-pages / build (push) Canceled after 0s
deploy-pages / deploy (push) Canceled after 0s
i18n consistency check / check (push) Canceled after 0s
provider adoption tests / test (chapter2/context-compression) (push) Canceled after 0s
provider adoption tests / test (chapter2/prompt-injection) (push) Canceled after 0s
provider adoption tests / test (chapter2/system-hint) (push) Canceled after 0s
provider adoption tests / test (chapter3/log-sanitization) (push) Canceled after 0s
web-search-agent tests / test (push) Canceled after 0s
web-search-agent tests / agentbook (push) Canceled after 0s
88 lines
4.6 KiB
JSON
88 lines
4.6 KiB
JSON
{
|
|
"schema_version": 1,
|
|
"experiment": "10-5",
|
|
"title": "Stanford Generative Agents two-day society reproduction and ablations",
|
|
"preregistered_before_arm_execution": true,
|
|
"upstream": {
|
|
"repository": "https://github.com/joonspk-research/generative_agents",
|
|
"commit": "fe05a71d3e4ed7d10bf68aa4eda6dd995ec070f4",
|
|
"license": "Apache-2.0",
|
|
"base_simulation": "base_the_ville_n25",
|
|
"history_file": "environment/frontend_server/static_dirs/assets/the_ville/agent_history_init_n25.csv"
|
|
},
|
|
"provider": {
|
|
"api_style": "OpenAI-compatible",
|
|
"endpoint": "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
|
|
"credential_environment_variable": "DASHSCOPE_API_KEY",
|
|
"chat_model": "qwen3.7-flash",
|
|
"chat_thinking": false,
|
|
"embedding_model": "text-embedding-v4",
|
|
"embedding_dimensions": 1024
|
|
},
|
|
"shared_seed": {
|
|
"simulation_step": 0,
|
|
"personas": 25,
|
|
"history_rows": 25,
|
|
"history_whispers": 248,
|
|
"rule": "Generate and persist the upstream relationship-history memories once, then fork all arms from the identical saved memory state."
|
|
},
|
|
"world": {
|
|
"start": "2023-02-13T00:00:00",
|
|
"end": "2023-02-15T00:00:00",
|
|
"seconds_per_step": 10,
|
|
"steps_per_arm": 17280,
|
|
"checkpoint_steps": 360
|
|
},
|
|
"arms": {
|
|
"baseline": {
|
|
"goal": "Retain the original Valentine's party and mayoral-election seeds.",
|
|
"treatment": "none"
|
|
},
|
|
"custom_goal": {
|
|
"goal": "Replace Isabella Rodriguez's party goal with a community climate-resilience workshop at Hobbs Cafe on February 14 from 5pm to 7pm.",
|
|
"treatment": "Mutate only Isabella Rodriguez's initial currently field before the first world step; retain the shared relationship memories and every other persona field."
|
|
},
|
|
"no_reflection": {
|
|
"goal": "Measure behavior without newly generated reflection thoughts.",
|
|
"treatment": "Disable Persona.reflect for the full run and raise importance triggers defensively; retain perception, retrieval, planning, execution, and chat-memory creation."
|
|
}
|
|
},
|
|
"retained_evidence": [
|
|
"credential-free raw chat request and response receipts with provider IDs, usage, and latency",
|
|
"embedding request and response metadata with vector dimension and SHA-256",
|
|
"all checkpoint status records",
|
|
"complete per-step movement logs",
|
|
"complete final associative-memory nodes and scratch state",
|
|
"deterministic diffusion, memory, reflection, social-network, action-continuity, and attendance analysis",
|
|
"blind plausibility judgments with raw independent-model receipts",
|
|
"artifact hashes and credential scan"
|
|
],
|
|
"primary_observations": {
|
|
"baseline_event_diffusion": "Number and first timestamp of agents whose memories mention the Valentine's party.",
|
|
"election_diffusion": "Number and first timestamp of agents whose memories mention the mayoral election.",
|
|
"custom_event_diffusion": "Number and first timestamp of agents whose memories mention the climate-resilience workshop.",
|
|
"reflection": "New thought nodes, evidence-linked thought nodes, and maximum thought depth relative to the shared seed.",
|
|
"social_network": "Unique chat edges and chat-node counts.",
|
|
"behavior": "Action diversity, action transitions, and Hobbs Cafe event-window attendance.",
|
|
"plausibility": "Blind rubric scores for temporal coherence, personality consistency, memory continuity, and social responsiveness."
|
|
},
|
|
"hypotheses": [
|
|
"The seeded baseline event and election information will diffuse beyond their originating agents.",
|
|
"Changing Isabella's initial goal will change the focal event that diffuses and the event-window behavior at Hobbs Cafe.",
|
|
"Disabling reflection will eliminate new reflection thoughts and reduce at least one blinded behavioral-plausibility dimension relative to baseline."
|
|
],
|
|
"interpretation_rule": "Completion depends on executing and retaining all arms and analyses, not on confirming any hypothesis. Null or contrary results remain complete negative findings.",
|
|
"acceptance_gates": [
|
|
"exact pinned upstream commit",
|
|
"one identical 25-persona history seed",
|
|
"three arms with exactly 17280 ten-second steps each",
|
|
"final time exactly 2023-02-15T00:00:00 in every arm",
|
|
"provider receipts with real unique response IDs and positive usage",
|
|
"complete final movement and memory evidence",
|
|
"custom goal present before the custom arm's first plan",
|
|
"no new reflection thoughts in the no-reflection arm",
|
|
"deterministic and blind plausibility analyses retained",
|
|
"all retained-file hashes valid and credential scan clean"
|
|
]
|
|
}
|