The Catalog vs the Atlas¶

The Cognitive Atlas was the starting corpus for the Catalog, and the two overlap heavily. They are not, however, the same kind of thing, and reading one as a subset of the other gets the relationship wrong.

What is the Cognitive Atlas? describes the Atlas on its own terms. The task mapping and process mapping pages give the correspondence row by row, and the Atlas mapping method page explains how each row was decided. This page says what it all adds up to.

Two resources doing different jobs¶

The Atlas is an open ontology covering the breadth of cognitive neuroscience, and it admits anything a contributor thought worth recording: experimental paradigms alongside rating scales, neuropsychological batteries, imaging protocol labels and physiological procedures, all filed under the single heading “task”.

The Catalog admits only paradigms that produce event-structured data, because its purpose is HED annotation of experimental events. That is a narrower and more specific test, set out in the task criteria.

The difference in scale follows from the difference in purpose.

	The Catalog	Cognitive Atlas
Experimental paradigms	103	857
Cognitive processes or concepts	172	918
Links between the two	486	1422
Mean processes or concepts per paradigm	4.72	1.66
Named variations	772	not modelled
Mean named variations per paradigm	7.5	not modelled

Size is the least interesting difference. What matters is the density and the shape. The Catalog links each paradigm to 4.7 processes on average against the Atlas’s 1.66, and it models named variations at 7.5 per paradigm, which the Atlas does not model at all: it registers each implementation as a separate top-level entry instead.

How much overlaps¶

The first table in each pair says what the Atlas can tell you about the Catalog; the second says what the Catalog covers of the Atlas. The task mapping and process mapping pages list every row.

Tasks¶

Tasks in the Catalog, by the match type of their Atlas counterpart:

Match type	Tasks	Share
exact	70	68%
close	11	11%
related	4	4%
none	18	17%

Atlas task entries, by the match type of their counterpart in the Catalog:

Match type	Atlas entries	Share
exact	86	10%
close	73	9%
related	24	3%
none	674	79%

The large none share in the second table is not a gap. Only 481 of the unmatched Atlas entries are experimental paradigms at all; the rest are instruments the Catalog deliberately excludes.

Processes¶

Processes in the Catalog, by the match type of their Atlas concept:

Match type	Processes	Share
exact	102	59%
close	14	8%
related	23	13%
none	33	19%

Atlas concepts, by the match type of their counterpart in the Catalog:

Match type	Atlas concepts	Share
exact	102	11%
close	14	2%
related	16	2%
none	786	86%

Likewise, about half the Atlas concept layer (455 of 918 concepts) is asserted by no Atlas task either, so a concept with no counterpart here is often one the Atlas itself makes no use of.

What the Catalog adds¶

For the 183 Atlas entries that do correspond to something here, the Atlas records on average 2.13 concepts each, and 51 of them record none at all. Against that, every task in the Catalog carries:

an inclusion test stating procedure, manipulation and measurement, so that membership is decidable rather than a matter of name
a curated list of named variations, each with a justification for why it is a variation and not a separate task
verified references
links to 172 defined processes in 19 categories, at 4.7 per task

The Atlas carries citations on 470 of its 857 task entries (54.8%), so references are not absent, but they are contributed rather than curated: the remaining 387 entries have none, and nothing records how any of them was checked. Inclusion tests and variations have no Atlas equivalent at all.

What the Atlas has that the Catalog does not¶

The traffic runs both ways, and the Atlas holds several things worth keeping in view.

Breadth. 857 task entries and 918 concepts against 103 and 172 here. For a construct the Catalog does not cover, the Atlas usually has something.
A concept relation graph. 1398 KINDOF and PARTOF edges between concepts. The Catalog groups processes into categories but asserts no relations between them.
Stable, citable identifiers for every entry, which the Catalog now references directly in its mapping tables.
Contrasts. 1585 contrast definitions across 529 tasks, the most reusable structured content the Atlas holds after the concept links.

How the Atlas shaped the Catalog¶

The influence is concrete. A gap analysis against the Atlas in April 2026 added 15 paradigms that the Catalog had been missing, among them the multi-armed bandit, the two-stage decision task, the random dot kinematogram, remember/know, and Raven’s Progressive Matrices. All 15 now map to an Atlas entry.

The reverse also holds. 18 tasks here have no Atlas counterpart at all, and several are heavily used paradigms the Atlas simply never registered: the Mismatch Negativity paradigm, the Dictator Game, Reading the Mind in the Eyes, Multiple Object Tracking, and the Weapons Identification Task among them. The task mapping page lists all 18.

Mismatch Negativity is the starkest: thousands of published studies, a candidate clinical biomarker, and no Atlas entry at all.

What the Atlas cannot be used for unchecked¶

The mapping could not be built by matching names, and the reason matters for anyone planning to consume the Atlas programmatically. Names agree while meanings do not, and names disagree while meanings agree; the Atlas mapping method page gives the cases.

Every row in these mappings was therefore checked against the archived Atlas record. That is the practical conclusion of the whole exercise: the Atlas is an excellent source of candidate terms and stable identifiers, and a poor source of automatic answers.