Two-Armed Bandit Task (HTML Companion)

Field	Value
Name	Two-Armed Bandit Task
Version	html (v0.1.0-dev)
URL / Repository	https://github.com/TaskBeacon/H000038-two-armed-bandit
Short Description	Browser companion for the Chinese two-armed bandit learning task with block-wise reward contingencies and cumulative score feedback
Created By	TaskBeacon psyflow-web py2js port
Modality	Behavior
Language	Chinese

1. Task Overview

This HTML companion mirrors the canonical local T000038-two-armed-bandit task. On each trial, participants choose between two machines, receive a stochastic reward outcome, and accumulate score across repeated trials. The browser port keeps the same stage order, response mapping, block-wise probability schedule, and reduced-data semantics while using the shared psyflow-web runner.

The participant-facing copy follows the 1000-point starting score described in the local instructions.

2. Task Flow

Block-Level Flow

Step	Description
Load Config	Load the browser preview config and stimulus bank.
Collect Subject Info	The shared runner collects subject_id before the task starts.
Show Instructions	Present the Chinese instruction text.
Countdown	Show a three-second countdown before each block begins.
Run Trials	Present fixation, bandit choice, choice confirmation, feedback, and ITI for each scheduled condition.
Show Block Break	Display the current block summary after each block.
Finalize	Display the final score and export data through the shared runner.

Trial-Level Flow

Step	Description
Fixation	Show a central fixation cross.
Bandit Choice	Show two machines and capture keys F and J.
Timeout Fallback	If no response is made, the task imputes a choice using no_choice_policy and still scores the trial.
Choice Confirmation	Highlight the selected machine and show the confirmation text.
Outcome Feedback	Show win/loss feedback and the updated total score.
ITI	Show the fixation cross again before the next trial.

Controller Logic

Feature	Description
Condition scheduling	Uses block-wise left/right probability rows from condition_generation.block_probabilities.
Score tracking	A RewardTracker keeps the cumulative score across the session.
Response contract	Keys f and j select the left and right machines.
Adaptive control	None. This is a static probabilistic choice task, not a staircase or adaptive controller.

3. Configuration Summary

Settings are defined in config/config.yaml.

a. Subject Info

Field	Meaning
subject_id	Numeric participant identifier supplied by the shared runner.

b. Stimuli

Stimulus ID	Purpose
instruction_text	Chinese task instructions.
fixation	Central fixation cross.
machine_left/right	Left/right machine panels.
machine_left_label/right_label	Machine labels.
highlight_left/right	Selected machine highlight frame.
choice_prompt	Choice prompt with response deadline.
target_prompt	Choice confirmation text.
feedback_win / feedback_loss	Reward outcome feedback with updated score.
block_break	Block summary screen.
good_bye	Final total score screen.
outcome_text	Reserved outcome text stimulus kept for parity with the canonical repo.

c. Timing

Parameter	Meaning
timing.pre_choice_fixation_duration	Duration of the fixation screen.
timing.bandit_choice_duration	Duration of the choice window.
timing.choice_confirmation_duration	Duration of the selection confirmation screen.
timing.outcome_feedback_duration	Duration of the feedback screen.
timing.iti_duration	Inter-trial interval duration.

d. Triggers

Parameter	Meaning
exp_onset / exp_end	Experiment start/end metadata.
block_onset / block_end	Block start/end metadata.
pre_choice_fixation_onset	Fixation onset metadata.
bandit_choice_onset	Choice onset metadata.
bandit_choice_left_press / bandit_choice_right_press	Left/right choice key metadata.
bandit_choice_no_response	Choice-window timeout metadata.
bandit_choice_forced	Forced-choice marker metadata.
choice_confirmation_onset	Confirmation onset metadata.
outcome_feedback_win_onset / outcome_feedback_loss_onset	Feedback onset metadata.
iti_onset	ITI onset metadata.

e. Browser Notes

Parameter	Meaning
task.save_path	Browser output target; the shared runner downloads results instead of writing to a desktop output directory.
task.voice_enabled	Keeps the browser companion text-only by default.
task.initial_score	Starting score used by the browser task.

4. Methods

Participants completed a browser-based two-armed bandit task implemented with the shared psyflow-web runtime. On each trial, they chose one of two machines with block-wise reward contingencies. The choice screen displayed the two machines and a deadline prompt, followed by a brief confirmation screen and a reward/no-reward feedback screen.

The task operationalizes learning from probabilistic feedback and score accumulation. If the participant does not respond within the choice window, the browser companion applies the configured no-choice fallback policy and still scores the trial, preserving the one-trial-one-row analysis contract.

All participant-facing wording is defined in YAML stimulus templates so the same trial logic can be reused across browser preview and future variants without code edits.

Run

1. Open psyflow-web in the shared workspace.
2. Run npm install if dependencies are not installed yet.
3. Start the runner with npm run dev.
4. Open http://127.0.0.1:4173/?task=H000038-two-armed-bandit.

Notes

• The browser companion keeps the canonical 4-block, 160-trial structure intact.
• The local trigger codes are preserved in config for traceability, but the browser runtime does not emit serial triggers.
• Pushes to main can refresh the shared runner through .github/workflows/notify-psyflow-web.yml when TASKBEACON_ORG_DISPATCH_TOKEN is configured.