TRACE / configs /task_analysis /compatibility.yaml
Ewakaa's picture
Initial public release: TRACE v1.0.0
c427f62 verified
# Task Analysis clinical-consistency rules — TRACE v1
# Module -> typical learner profiles (some modules only fit older learners)
module_to_learner_profiles:
basic_living: [school_age, adolescent, adult]
home: [adolescent, adult]
community: [adolescent, adult]
vocational: [adolescent, adult]
independent_living: [adult]
# Mastery state -> acceptable reinforcement schedules (chaining-specific)
mastery_to_reinforcement:
emerging: [crf_per_step]
developing: [crf_per_step]
approaching: [crf_per_step, differential_per_step]
near: [differential_per_step, token_economy]
mastered: [terminal_reinforcement, token_economy]
generalization: [terminal_reinforcement]
# Chain type tends to follow skill preference; fallback is total_task
use_preferred_chain_type_probability: 0.75
# Sampling across modules (weighted toward typical clinical frequency)
module_sampling_weights:
basic_living: 0.35
home: 0.20
community: 0.15
vocational: 0.15
independent_living: 0.15
# Prompt strategy bias by mastery state
mastery_to_prompt_strategies:
emerging: [graduated_guidance, most_to_least]
developing: [graduated_guidance, most_to_least]
approaching: [most_to_least, least_to_most]
near: [least_to_most]
mastered: [least_to_most]
generalization: [least_to_most]