TaskParameter

Per-trial ground-truth GENERATIVE task parameters (latent task state, not behavior), in long format: one row per parameter (and grid cell) per trial — e.g., a bandit's true reward probabilities per (state, action), a slot machine's true means. Serves simulation, parameter recovery, and engine calibration.

note

Rows are keyed by the natural key (trial_index, task_index, parameter, index_1, index_2); like StimulusComponent/OptionComponent, this table has no surrogate primary key.

note

These values are legitimate inputs for simulation and recovery tooling and ILLEGITIMATE as predictors for models of behavior: analysis pipelines should refuse (or loudly flag) any model that consumes ground-truth task parameters when predicting the very responses those parameters generated.

Context

Field	Type	Requirement	Description
trial_index	id	required	Refers to the trial_index in the Response table: the trial whose generative state this row records.
task_index	index	optional	When multitask_type is not empty, refers to the individual task this parameter belongs to — different tasks have different parameters, which is exactly the ambiguity a wide encoding cannot express.

What

Field	Type	Requirement	Description
parameter	string	required	Name of the generative parameter (e.g., p_reward, slot_mean, drift_rate). Parameter definitions (meaning, units, grid semantics) belong in the dataset's codebook.
index_1	index	optional	First grid index, when the parameter is defined over a grid (e.g., state). Leave empty for scalar parameters.
index_2	index	optional	Second grid index, when the parameter is defined over a two-dimensional grid (e.g., action within state). Leave empty otherwise.
value	any	required	The ground-truth value of the parameter for this trial (and grid cell).