Probabilistic Stimulus Selection

Field	Value
Name	Probabilistic Stimulus Selection
Version	v0.1.0-dev
URL / Repository	E:/Taskbeacon/T000040-probabilistic-stimulus-selection
Short Description	Chinese probabilistic reinforcement-learning task with three learning pairs, 15 transfer pairings, and kana stimuli.
Created By	TaskBeacon
Date Updated	2026-04-04
PsyFlow Version	0.1.12
PsychoPy Version	2025.2.4
Modality	Behavior
Language	Chinese
Voice Name	zh-CN-YunyangNeural

1. Task Overview

This task reproduces the classic probabilistic stimulus selection paradigm in a Chinese-localized PsyFlow implementation. Participants learn three two-option pairings, AB, CD, and EF, where the higher-probability role is rewarded at 80/20, 70/30, and 60/40, respectively. After learning reaches criterion, the task presents all 15 possible pairings without feedback to measure transfer bias and learned preference.

The six visible symbols are Japanese kana glyphs rendered from PsychoPy text stimuli. Their mapping to roles A-F is randomized per participant, so the same learning structure is preserved while the concrete symbols vary across runs.

2. Task Flow

Block-Level Flow

Step	Description
1. Parse mode/config	main.py loads human, qa, or sim mode and the matching YAML config.
2. Initialize runtime	Window, keyboard, triggers, and stimulus bank are initialized.
3. Assign roles	Six kana symbols are shuffled into roles A-F deterministically from the subject seed.
4. Run instructions	The participant reads the Chinese instruction page and presses space to start.
5. Run learning blocks	Learning blocks repeat until all pair criteria are met or max_learning_blocks is reached.
6. Run transfer block	One transfer block presents all 15 pairings with no feedback.
7. Finish	The goodbye screen appears and all trial data are saved.

Trial-Level Flow

Step	Description
Block ready	A centered ready screen appears for 3.0 s before each block.
Choice screen	Two kana symbols appear left and right; the participant responds with the left or right key within 4.0 s.
Feedback, learning only	A probabilistic 正确 or 错误 screen appears for 1.0 s based on the chosen role's reward probability.
No-feedback transfer	Transfer trials skip feedback and go straight to the inter-trial interval.
Inter-trial interval	A fixation + appears for 1.0 s before the next trial.

Controller Logic

Component	Description
Role assignment	build_role_assignment(...) shuffles the six kana stimuli into roles A-F per participant.
Learning schedule	build_learning_block_schedule(...) creates exactly 20 trials per learning pair with exact within-block left/right balance.
Transfer schedule	build_transfer_block_schedule(...) creates exactly 10 trials per transfer pair across all 15 pairings.
Probabilistic feedback	On learning trials, the chosen role is sampled against its pair-specific win probability.
Criterion check	evaluate_learning_block(...) requires AB >= 0.65, CD >= 0.60, and EF >= 0.50.
Transfer summary	summarize_transfer_block(...) reports role choice rates and timeout counts for audit logs.

Other Logic

Component	Description
Hidden score	A hidden cumulative score is tracked from learning feedback outcomes, but it is not shown to participants.
No-response policy	Learning timeouts count as incorrect; transfer omissions are logged but do not create feedback.
QA and sim modes	config/config_qa.yaml, config/config_scripted_sim.yaml, and config/config_sampler_sim.yaml reuse the same task logic with shorter smoke-test schedules.
Stimulus rendering	All participant-facing text and symbols are PsychoPy text stimuli defined in YAML; no external media is required.

3. Configuration Summary

All human settings are defined in config/config.yaml. QA and smoke-test overrides live in config/config_qa.yaml, config/config_scripted_sim.yaml, and config/config_sampler_sim.yaml.

a. Subject Info

Field	Meaning
subject_id	3-digit participant identifier used to seed the kana role assignment.

b. Window Settings

Parameter	Value
size	[1280, 720]
units	pix
screen	0
bg_color	black
fullscreen	false
monitor_width_cm	35.5
monitor_distance_cm	60

c. Stimuli

Name	Type	Description
instruction_text	text	Chinese instruction page with left/right key labels injected at runtime.
block_ready	text	Ready screen for learning or transfer blocks.
iti_fixation	text	Centered + used during the inter-trial interval.
feedback_correct	text	Green Chinese correctness cue 正确.
feedback_incorrect	text	Orange Chinese error cue 错误.
good_bye	text	Final exit screen.
kana_1 to kana_6	text	Six kana glyphs (あ, い, う, え, お, か) that are randomly assigned to roles A-F per participant.

d. Timing

Phase	Duration
Block ready	3.0 s
Response window	4.0 s
Learning feedback	1.0 s
Inter-trial interval	1.0 s

e. Triggers

Event	Code
Experiment onset	1
Experiment end	2
Block onset	10
Block end	11
Trial onset	20
Left response	31
Right response	32
Timeout	33
Feedback onset	40
No-feedback onset	41
ITI onset	50

f. Adaptive Controller

Parameter	Value
phase_order	learning -> transfer
learning_conditions	train_ab, train_cd, train_ef
learning_probabilities	AB 80/20, CD 70/30, EF 60/40
learning_trials_per_block	60
learning_trials_per_pair	20
learning_criteria	AB >= 0.65, CD >= 0.60, EF >= 0.50
max_learning_blocks	6
transfer_conditions	15 pairings: AB, AC, AD, AE, AF, BC, BD, BE, BF, CD, CE, CF, DE, DF, EF
transfer_trials_total	150
transfer_trials_per_pair	10
response_key_left / response_key_right	a / l
response_key_labels.left / response_key_labels.right	A / L
left_right_balance_policy	exact_within_block
symbol_role_shuffle_policy	random_per_subject
no_response_policy	timeout_counts_as_incorrect_on_learning
correct_score_delta / incorrect_score_delta	1 / 0

4. Methods (for academic publication)

Probabilistic stimulus selection is a classic reinforcement-learning paradigm used to measure reward sensitivity, choice bias, and transfer from learned feedback contingencies. The canonical structure contains three acquisition pairs with progressively weaker reward contingencies and a transfer test that presents all pairwise combinations without feedback.

This implementation preserves that structure while using six kana text stimuli and Chinese participant instructions. The learning phase uses probabilistic feedback, exact within-block left/right balance, and criterion-driven repetition. The transfer phase keeps the same symbols but removes feedback, allowing the analysis of learned preferences rather than immediate reinforcement.