Context integration

Which stream determines the answer?

context_integration presents a binary context cue, then two simultaneous evidence streams. Context 1 selects stream 1; context 2 selects stream 2. The seven receptors are two context channels, two left/right stream pairs and go. Two effectors respond during go; there is one neural frame per world tick.

context (8) → two evidence streams (32 pulses) → go / response (8)

from brainlesslab import TaskSpec
from brainlesslab.tasks import definition, resolve

task = TaskSpec("context_integration",
                options={"congruency": "conflicting", "context_mode": "transient"})
print(resolve(task))
print(definition(task))

Expect seven inputs, two effectors and horizon 48. The defaults use 32 one-tick pulses and evidence fraction 0.25.

Make selection necessary

congruency="congruent" gives both streams the same majority. "conflicting" makes their majorities opposite. "mixed" draws the irrelevant stream’s majority independently. Context and relevant answer are balanced choices. Each stream controls its final pulse as in evidence accumulation.

context_mode="transient" removes context after the initial cue; "persistent" retains it through response; "hidden" removes context only. Evidence remains visible in all three modes. Success on congruent streams alone cannot establish context-dependent selection.

The observation-based oracle counts the selected stream. context_blind always selects stream 1; irrelevant selects the other. The latter is systematically wrong on conflicting trials. These references answer different questions from the random-action null.

Report the declared difficulty

The 36 presets cross three pulse counts, three evidence fractions, congruent or conflicting streams and transient or persistent context. probe_accuracy is one binary result per complete trial; ties choose left, chance is 0.5 and warm-up is zero.

Compare matched worlds and full resets within fixed-wiring blocks. Continue with decoding to inspect context and answer information at separate observation points.