Recall interference

Which information must survive?

Delayed cue asks for the first binary cue after a blank delay. recall_interference inserts later cues on the same receptor channels. Both have three inputs—left, right and go—and two effectors. One neural frame is one world tick.

initial cue (8) → delay, possibly with distractors → go / response (8)

The preset bank has five delayed-cue cells with delays 0, 8, 32, 128 and 256. Interference adds six cells: delays 32, 128 and 256 crossed with one or three four-tick distractors.

from brainlesslab import CompositionSpec, NodeSpec, TaskSpec, profile, resolve
from brainlesslab.tasks import capacity_probe_presets

cell = next(cell for cell in capacity_probe_presets()
            if cell.id == "recall_interference__delay_32_distractors_3")
target = resolve(profile(CompositionSpec(
    node=NodeSpec("sorn"),
    task=TaskSpec(cell.task, options=cell.task_options),
    count=32,
))).targets[0]
print(cell.id, target.horizon, target.n_inputs)

Expect the selected ID, horizon 48 and three inputs. Nonzero distractors require at least 32 delay ticks and enough blank spacing between events. The distractor schedule lies inside the delay and does not extend it.

Distinguish visibility from retention

cue_mode accepts transient, persistent and hidden. distractor_mode accepts random, matched, opposite and absent. Random draws are retained when visibility changes. During a visible distractor, its cue replaces the initial cue channels; a persistent initial cue returns after that interval.

Matched and opposite distractors reveal the initial answer even when the original cue is hidden. Use random or absent distractors for an uninformed hidden-cue control. Persistent cue permits a current-input shortcut.

The observation-based oracle remembers the first cue and ignores distractors; a last-cue reference tests overwriting. Constant and random references answer different null questions.

Report the curve

The primary keys are recall_accuracy for delayed cue and probe_accuracy for interference. Each complete trial contributes one binary response; ties select left. Chance is 0.5 and the signed adjustment may be negative. Warm-up is zero.

Report delay, distractor count and visibility alongside each outcome. One successful short cell does not establish durable retention or interference resistance. Continue with whole-trial decoding and paired comparisons.