Process Selection and Definition Criteria¶

Date: 2026-04-19 (updated) Purpose: Reference document specifying how the 172-process HED cognitive process catalog was scoped, how processes were selected, and the rules for definitions, naming, categorization, references, and task linkage. Companion to tasks_criteria.md (task-side criteria). Per-category editorial notes (scope, out-of-scope, issues, history) are now in process_details.json; The prior reference and categories documents have been archived.

1. Process Selection Criteria¶

What counts as a process in this catalog?¶

A cognitive process is a mental operation hypothesized to occur in a participant during a trial — something with an identifiable onset, an eliciting condition, and an in-principle measurable signature. Concretely, a candidate is a cognitive process if we can plausibly answer three questions:

When does it happen? — at what point in a trial, relative to stimulus and response, would we expect this process to be engaged?
How is it elicited? — what stimulus structure, instruction, or contingency would drive it?
How is it measured? — what behavioral, physiological, or neural signature would index it (even if imperfectly)?

This is a forward-looking plausibility test, not a gating filter. We keep candidates where we expect the answers exist in the literature, even if we have not yet found them. The test exists to rule out things that are categorically the wrong kind of object — not to demand finished evidence.

A process row should sit at the level where the inclusion-test triple has a single answer. If the question “when does it happen” has multiple, non-reducible answers (e.g., “working memory” can mean encoding, maintenance, manipulation, or updating, each at different trial phases), the candidate is too coarse and should be either split into sub-processes or treated as a category label only.

What is excluded?¶

The following do not count as cognitive processes in this catalog:

States and traits: Fear, arousal, anxiety, empathy-as-a-trait, boredom, mood. HED handles these as stimulus/state labels.
Specific emotions as states: Happiness, disgust, anger, sadness. These are what participants experience, not operations they perform. Emotion recognition and emotion regulation are processes; the emotions themselves are not.
Stimulus categories: Faces, words, tones, music. These are inputs to processes, not processes themselves.
Individual-difference constructs: Fluid intelligence, working-memory capacity as a score, impulsivity as a trait, handedness, dexterity. These describe between-person variation, not within-trial operations.
Dimensions and task parameters: Valence, arousal as a dimension, working-memory load, stimulus-onset asynchrony. These are properties of stimuli or experimental designs.
Umbrella terms: Attention, Memory, Language, Perception, Reasoning, Motivation, Social cognition, Consciousness, Emotion, Cognitive control, Motor control, Decision making. The category label carries the umbrella role; no process row should repeat it. Each umbrella that was in the original list has been either dropped entirely or split into concrete sub-processes that individually pass the inclusion test.
Analysis methods and modeling frameworks: Drift-diffusion modeling, Bayesian inference, representational similarity analysis. These are ways of measuring or modeling processes, not processes themselves.
Paradigm names used as process labels: A paradigm name (e.g., “Stroop effect”) is not a process. The underlying operations (interference control, response conflict) are.

Process vs. phenomenon¶

When the literature uses the same noun for both a phenomenon and the process producing it (extinction, directed forgetting, retroactive interference, masking), we treat it as a process and fold the phenomenon into the same row. We do not create separate rows for the process and the phenomenon it produces unless the distinction earns its keep.

Modality-abstract vs. modality-specific¶

Processes at the process level are modality-abstract when the same mechanism operates across modalities. Masking is a single process row (modality-abstract); the Auditory Masking Task and Visual Masking Task carry the modality distinction on the task side. The same pattern applies to priming and similar constructs: one process row, multiple task rows.

2. Naming Conventions¶

Process IDs¶

Every process has an identifier of the form hed_<slug>, where <slug> is a snake_case string. Examples: hed_response_inhibition, hed_visual_working_memory, hed_working_memory_updating.

The hed_ prefix is this catalog’s working identifier — it is a naming choice, not a claim about HED-schema membership. How (or whether) these IDs relate to the HED schema in the future is a separate, downstream decision.

Stability policy: During the current audit phase, IDs are explicitly provisional. They may change when a definition is refined, a near-duplicate is merged, or a better slug is chosen. Every rename must cascade atomically through process_details.json, task_details.json, and every derived file (task_names.json, process_task_index.json, process_task_crossref.md). Post-stability (after a declared freeze milestone), IDs are frozen: renames become aliases, and ID reassignment is a breaking change.

Process names¶

Process names use sentence case: the first word is capitalized, the remainder is lowercase except for proper nouns. Examples: “Associative learning”, “Working memory updating”, “Pavlovian conditioning”.

Process names do not end in “process” or “mechanism” unless those words are part of the established term (none currently do).

Aliases¶

When multiple names exist for the same process, one is designated canonical (process_name) and the others are recorded in an aliases array. Each alias carries a name and an optional note explaining the terminological distinction or the reason for the merge. An alias is not a separate process — it fails the test of having a distinct When/Elicited/Measured triple.

The aliases field parallels the task-side aliases array: it captures alternative names that a researcher might search for, preserving the terminological or theoretical distinction without creating a separate process row. Aliases do not affect process counts, task linkage, or category membership.

Schema:

{
  "aliases": [
    {
      "name": "Operant conditioning",
      "note": "Skinnerian terminology; emphasizes the operant response and reinforcement schedules."
    }
  ]
}

Current resolved aliases:

Canonical	Alias(es)	Rationale
Instrumental conditioning	Operant conditioning	Skinnerian terminology; emphasizes reinforcement schedules. Merged 2026-04-18.
Perspective taking	Mentalizing, Theory of mind	Mentalizing is the process verb (neuroimaging); Theory of mind is the capacity noun (developmental).
Word recognition	Visual word recognition	“Visual” is the default modality; auditory word recognition has its own name.
Risk processing	Risky decision making	Risk processing names the cognitive operation; risky decision making names the task class.
Emotion recognition	Emotion perception	Recognition is operationalized (identification of a named category); perception is broader.
Working memory updating	Updating, Updating (WM)	Merged 2026-04-17; plain “Updating” was memory-context-underspecified.
Active maintenance	Maintenance	Generic term for holding information over a delay; merged into active maintenance 2026-04-19.
Set shifting	Cognitive flexibility	Cognitive flexibility is a capacity-level construct that fails the single-answer inclusion test; set shifting is its primary experimental operationalization. Dropped as separate process 2026-04-19.
Deductive reasoning	Logical reasoning	Logical reasoning is an umbrella over deductive + inductive; dropped as separate process 2026-04-19.
Gustatory perception	Gustation	Renamed 2026-04-15 for consistency with other modality-perception rows.
Somatosensory perception	Somatosensation	Renamed 2026-04-15 for consistency with other modality-perception rows.
Sustained attention	Vigilance	Synonymous when operationalized as a continuous-performance measure.

3. Definition Standards¶

Requirements¶

Every process must have a definition field that is:

Non-empty. A blank definition is a data quality failure.
One sentence minimum. The definition should be a self-contained statement of what the process is, understandable to a researcher in a neighboring subfield.
Operationally grounded. The definition should connect to how the process is elicited or measured, not just what it “is” in the abstract.

What makes a good definition¶

A good process definition states what the process does (or what the participant does), in terms that connect to observable paradigms. Compare:

Weak: “The process of inhibiting responses.” (Circular.)
Better: “Suppression of a prepotent or already-initiated response when it becomes inappropriate, indexed behaviorally by stop-signal reaction time or commission errors.”

Definitions that begin with “See 5.6” or cross-reference a section number from an earlier document are stale artifacts and should be rewritten.

Cross-references in definitions¶

If a process definition says “See instrumental conditioning” or similar, this indicates a synonym relationship. The definition should still state what the process is, even if briefly, rather than being a bare pointer.

4. Category Rules¶

The 19 categories¶

The catalog organizes 172 processes into 19 categories. The categories, their process counts, and their scope are defined in the categories array of process_details.json. The 19 categories are:

Associative Learning and Reinforcement (13)
Auditory and Pre-Attentive Deviance Processing (4)
Awareness, Agency, and Metacognition (13)
Cognitive Flexibility and Higher-Order Executive Function (3)
Emotion Perception and Regulation (5)
Face and Object Perception (12)
Implicit and Statistical Learning (2)
Inhibitory Control and Conflict Monitoring (9)
Language Comprehension and Production (16)
Long-Term Memory (21)
Motor Preparation, Timing, and Execution (16)
Perceptual Decision-Making (Evidence Accumulation) (1)
Reasoning and Problem-Solving (11)
Reward Anticipation and Motivation (6)
Selective and Sustained Attention (11)
Short-Term and Working Memory (9)
Social Cognition and Strategic Social Choice (11)
Spatial Cognition and Navigation (2)
Value-Based Decision-Making Under Risk and Uncertainty (7)

Purpose of categories¶

Categories serve as organizational labels for human browsing. They are not ontological commitments. A process belongs to the category whose scope best describes its primary research tradition. Categories do not imply inheritance or class membership.

Category structure¶

Each category entry in process_details.json has:

scope: What belongs here — a brief description of the process family.
out_of_scope (optional): What tends to get put here but shouldn’t.
issues (optional): Open questions about boundaries, near-synonyms, or structural problems.
history (optional): How the category evolved during the reframe.

the prior categories document previously held this content as a markdown editorial companion; it was reconciled and archived to on 2026-04-19.

Dual-category processes¶

Every process is assigned to exactly one category. When a process could fit two categories, the rule is:

File by primary research tradition. Antisaccade is in Motor Preparation, Timing, and Execution because it is an oculomotor output measure, even though it centrally depends on prepotent-saccade inhibition (which would argue for Inhibitory Control).
File second-order processes by their second-order character. Metacognitive monitoring, self-monitoring, feeling of knowing, and judgment of learning are in Awareness, Agency, and Metacognition (not in their “first-order” home categories) because the reportable content is what the participant consciously experiences about their own cognition.
Perceptual processes go by modality when modality-specific, by mechanism when modality-abstract. Proprioception stays in Motor despite being perceptual, because it is tightly functionally coupled to motor control in the experimental literature.

Category process counts¶

Each category entry in process_details.json carries a process_count field. This must match the actual number of processes assigned to that category_id in the processes array. Drift between the count and the actual membership is a data quality issue.

5. Reference Standards¶

Reference types¶

Each process carries two reference arrays:

fundamental_references: Paradigm-defining or foundational references. These are the papers or books that established the process as a research construct. Typically pre-2000, often classic citations.
recent_references: Recent or standard-review references. These are post-2010 papers (preferred) that provide modern entry points into the literature: reviews, meta-analyses, or influential empirical papers.

Current state¶

As of 2026-04-18, all 407 reference entries in process_details.json have empty title fields. This is a known future-phase item. The citation_string field carries the human-readable citation (e.g., “Rescorla & Wagner (1972) in Classical Conditioning II”), and journal and year fields are populated. Title fill is deferred to a separate literature-lookup pass.

Quality expectations¶

Every process should have at least one fundamental reference and at least one recent reference.
Citation strings should be clean — no embedded markdown headers, no stitched-together text fragments.
References for the 6 Awareness/Agency/Metacognition rows added 2026-04-15 and the 5 problem-solving/valuation additions were written fresh; all inherited references come from the original Cognitive Atlas source documents.
The Levy & Glimcher (2012) citation was corrected on 2026-04-18: journal is Current Opinion in Neurobiology 22:1027–1038, not Journal of Neuroscience.

Reference provenance (proposed, not yet implemented)¶

A future enhancement would mark each reference with its source (atlas, supplement, gap_analysis, claude_generated, user_provided). Currently all references are indistinguishable in origin. Provenance is not a quality signal — atlas means “inherited from the Cognitive Atlas starting corpus” (uncurated), not “validated.”

6. Task Linkage Rules¶

How processes link to tasks¶

A process links to a task if the task’s inclusion test (Procedure / Manipulation / Measurement) engages that process. The link is stored in the tasks[] array of each process entry in process_details.json, using hedtsk_<slug> IDs. The reverse direction (task → process) is stored in the hed_process_ids array of each task entry in task_details.json.

When is a link justified?¶

A link from task T to process P should satisfy at least one of:

Canonical link: The paradigm-defining paper(s) for T theorize that T engages P. Example: Stroop → interference control.
Contrast-based link: Some condition contrast in T is standardly interpreted as isolating P. Example: congruent vs. incongruent in Stroop → conflict monitoring.
Measurement-based link: T has a measure standardly interpreted as an index of P. Example: stop-signal reaction time in the Stop-Signal Task → response inhibition.

What a link is not¶

A link is not an ontology assertion (no is-a, no class hierarchy).
A link is not an exclusion — other processes may be engaged; we list only those the task is designed to probe.
A link is not symmetric in the sense that every process that lists T also makes T its primary paradigm. Some processes are engaged by a task but primarily studied with a different task.

Unlinked processes¶

As of 2026-04-19, 19 of 172 processes have task_count = 0 (no linked tasks). Some of these are legitimate: the process is real but none of the 103 tasks in the current catalog are designed to study it (e.g., gustatory perception — no taste-perception task in the catalog). Others may signal problems: the process is too abstract, or a link was missed.

7. Known Issues and Policy Questions¶

7.1 Instrumental ≡ Operant conditioning — RESOLVED 2026-04-18¶

Issue: Two separate entries existed: hed_instrumental_conditioning and hed_operant_conditioning. These are synonyms for the same process.

Resolution: Collapsed to single hed_instrumental_conditioning entry. “Operant conditioning” recorded as an alias with note about the Skinnerian terminological emphasis. References from the operant entry (Ferster & Skinner 1957, Staddon & Cerutti 2003) absorbed into the instrumental entry. Definition expanded to cover both traditions.

7.2 Working memory umbrella — OPEN (deferred)¶

Issue: hed_working_memory exists as a process row alongside its component sub-processes (verbal WM, visual WM, spatial WM, active maintenance, manipulation, updating, rehearsal, chunking). This was flagged as item 3 in the umbrella decisions pass and explicitly deferred.

Current state: hed_working_memory is retained. It has a concrete definition and task links. The question is whether tasks should link to the umbrella or only to the specific components.

Options: (a) Drop hed_working_memory and redirect all its task links to specific components. (b) Keep it as a legitimate process row (it does have a concrete definition: “system for short-term maintenance, updating, and manipulation…”). © Keep it but never link tasks to it directly — tasks always link to the specific sub-process.

Recommended action: Policy decision needed. Not blocking.

7.3 Cognitive flexibility ≈ Set shifting — RESOLVED 2026-04-19¶

Issue: hed_cognitive_flexibility and hed_set_shifting were near-synonyms. Cognitive flexibility is a capacity-level construct that fails the single-answer inclusion test; set shifting is its primary experimental operationalization.

Resolution: hed_cognitive_flexibility dropped. References absorbed into hed_set_shifting. “Cognitive flexibility” added as alias with note explaining the broader construct. Category count 4 → 3.

7.4 Attentional flexibility / mental flexibility sub-split — OPEN¶

Issue: The user flagged that attentional flexibility and mental flexibility should perhaps be separate sub-processes under cognitive flexibility.

Current state: Deferred. No action taken.

Recommended action: Needs research. Low priority.

7.5 Emotion Perception and Regulation thin on appraisal — OPEN¶

Issue: The category has only 5 processes. Coverage of appraisal-stage processes is thin.

Current state: Expected to grow if paradigms targeting appraisal are added.

Recommended action: Future-phase. No action needed now.

7.6 Antisaccade category membership — OPEN¶

Issue: hed_antisaccade is in Motor Preparation, Timing, and Execution. It could be argued into Inhibitory Control since it centrally depends on prepotent-saccade inhibition.

Current state: Retained in Motor because it is an oculomotor output measure.

Recommended action: Low-priority policy question. Current placement is defensible.

7.7 Reading: process vs. behavior — OPEN¶

Issue: hed_reading conflates reading as a cognitive process (visual processing of written text integrating orthography, phonology, and meaning) with reading as a behavior. The definition is process-oriented, which is correct, but the boundary is fuzzy.

Current state: Retained as is.

Recommended action: Low priority. The current definition is adequate.

7.8 Stale cross-reference definitions — RESOLVED 2026-04-19¶

Issue: Several process definitions contained stale cross-references from a prior numbered-section document. Three had bare “See 5.6” / “See 5.4” markers (resolved 2026-04-18). Additionally: hed_speech_production had “see also 5.9 (motor component)”; hed_emotion_recognition had “overlap with 5.10 Emotion perception”; hed_insight had a spurious \! escape in “Aha!”.

Resolution: All definitions rewritten or corrected with substantive content. Final pass completed 2026-04-19. Verified against the prior reference document before archival.

7.9 Sustained attention ≡ Vigilance — RESOLVED¶

Issue: Sustained attention and vigilance are treated as synonymous when operationalized as a continuous-performance measure.

Current state: Single row hed_sustained_attention with the synonymy noted in the definition.

Status: Resolved. No action needed.

7.10 Category scope coverage — OPEN¶

Issue: Several categories are very small (Perceptual Decision-Making has 1 process; Implicit and Statistical Learning has 2; Spatial Cognition and Navigation has 2). These may grow as tasks are added, or may warrant merging.

Current state: Retained as separate categories on the expectation of future growth.

Recommended action: Monitor. Future-phase.

7.11 Processes with no task links (20 processes) — REVIEWED¶

Issue: Originally 22 processes had task_count = 0. After review: hed_pattern_completion was linked to MST (genuine missed link); hed_maintenance was merged into hed_active_maintenance (which has links). Remaining 20 unlinked processes assessed individually in audit report — 14 are legitimate gaps (no task in the 103-task catalog), 4 were confirmed correctly unlinked, 2 are structural.

Current state: Reviewed. No further action needed until catalog expands.

Recommended action: Low priority. Re-evaluate when new tasks are added.

7.12 Empty reference titles — RESOLVED (2026-04-20)¶

Issue: All reference objects in process_details.json had empty title fields.

Current state: Resolved. Citation enrichment pass (Phase B) run 2026-04-20. 390/404 references now have structured metadata (title, authors, venue, venue_type, doi, etc.). 14 unresolved — all classic books or obscure manuals with no CrossRef/OpenAlex record; documented in .status/citation_gaps_2026-04-20.md §2.2. 4 false-positive matches identified in spot-check and documented in §2.3 of the same file; these need manual correction. Original journal, year, and citation_string fields preserved on all refs.

Recommended action: Correct the 4 false positives (§2.3 of gaps doc). Attempt manual resolution of O’Keefe & Nadel (1978) and Braver et al. (2007) (§2.2 of gaps doc). Proceed to Phase C (task reference enrichment) once process pass is approved.

7.13 Mentalizing → Perspective taking synonym status — RESOLVED¶

Issue: During the reframe, Theory of mind / Mentalizing was merged with canonical form “Perspective taking” (hed_perspective_taking). The Social Cognition category scope field still references both “Theory of mind” and the canonical ID.

Current state: Resolved. The scope field’s mention is for clarity, not a separate entry.

7.14 Source of truth for process definitions — RESOLVED 2026-04-19¶

Issue: Both the prior reference document and process_details.json contained process definitions, with the risk of drift.

Resolution: Reconciliation pass completed 2026-04-19. All 172 process definitions and all citation strings were compared programmatically. Five differences found (category field staleness in 3 categories; 2 process definition issues); all resolved in favor of process_details.json except hed_speech_production (JSON had a stale cross-reference; updated from markdown). the prior reference document and the prior categories document archived to . `process_details.json, and . They are summarized here for reference.

Umbrella drops¶

The following umbrella terms were removed from the process list entirely: Attention, Memory, Language, Perception, Reasoning, Motivation, Social cognition, Cognitive control, Motor control, Consciousness, Emotion, Decision making. The category labels carry the umbrella role. (Decisions log, 2026-04-15.)

Working memory SPLIT deferred¶

Item 3 from the umbrella pass. hed_working_memory retained for now alongside component rows. (Umbrella decisions, 2026-04-15.)

Category 10 dissolved¶

Memory Control and Metamemory dissolved. Directed forgetting → Long-Term Memory. Feeling of knowing, Judgment of learning → Awareness, Agency, and Metacognition. Umbrella/non-process entries dropped: Confidence, Metacognition, Metamemory, Introspection, Self-reflection. Category count 20 → 19. (Decisions log, 2026-04-15.)

Category renames¶

“Perceptual Awareness and Consciousness” → “Awareness, Agency, and Metacognition”
“Working Memory Maintenance and Manipulation” → “Short-Term and Working Memory”
“Reasoning, Problem-Solving, and Fluid Intelligence” → “Reasoning and Problem-Solving”

Sensory modality naming consistency¶

Gustation → Gustatory perception. Somatosensation → Somatosensory perception. Proprioception stays in Motor (Finding 1b, Option B accepted). (Side findings resolutions, 2026-04-15.)

Self-monitoring moved¶

Self-monitoring moved from Inhibitory Control to Awareness, Agency, and Metacognition. (Finding 2, accepted.)

Problem-solving rows added¶

Insight, Means-ends analysis, Planning, Subgoaling added to Reasoning and Problem-Solving. (Finding 3, accepted.)

Valuation rows added¶

Valuation, Choice commitment added to Value-Based Decision-Making. (Finding 5, accepted.)

Motivation axes collapsed¶

Appetitive motivation and Aversive motivation dropped (stimulus properties). Approach motivation and Avoidance motivation retained. (Finding 6, Option B accepted.)

Synonym merges¶

Word recognition (← Visual word recognition), Perspective taking (← Mentalizing, Theory of mind), Risk processing (← Risky decision making), Emotion recognition (← Emotion perception). (Finding 7, accepted.)

Updating / Updating (WM) merged¶

Consolidated into hed_working_memory_updating on 2026-04-17. Plain “Updating” was memory-context-underspecified.

Six Awareness/Agency/Metacognition rows added¶

Metacognitive monitoring, Metacognitive control, Sense of agency, Body ownership, Interoceptive awareness, Self-referential processing. Process count 172 → 178. (Subsequently reduced to 172 after merges, umbrella drops, and dedup.)

Aliases schema added (2026-04-18)¶

Added aliases array to process schema for recording alternative names with explanatory notes. Populated for 12 processes with resolved synonym/rename history. Rule added to process_criteria.md §2.

Instrumental ≡ Operant conditioning merged (2026-04-18)¶

hed_operant_conditioning removed. References absorbed into hed_instrumental_conditioning, which gains “Operant conditioning” as alias. Count 176 → 175.

Maintenance merged into Active maintenance (2026-04-19)¶

hed_maintenance removed. Goldman-Rakic (1995) reference absorbed into hed_active_maintenance, which gains “Maintenance” as alias. Short-Term and Working Memory count 10 → 9. Total 175 → 174.

Logical reasoning dropped (2026-04-19)¶

hed_logical_reasoning removed — umbrella over deductive + inductive reasoning, fails the single-answer inclusion test. Wason (1968) absorbed into hed_deductive_reasoning, which gains “Logical reasoning” as alias. Reasoning and Problem-Solving count 12 → 11. Total 174 → 173.

Cognitive flexibility dropped (2026-04-19)¶

hed_cognitive_flexibility removed — capacity-level construct that fails the single-answer inclusion test; set shifting is the primary experimental operationalization. Diamond (2013) and Dajani & Uddin (2015) absorbed into hed_set_shifting, which gains “Cognitive flexibility” as alias. Verbal Fluency Task link dropped (verbal fluency engages flexible search/strategy, not set shifting). Cognitive Flexibility and Higher-Order Executive Function count 4 → 3. Total 173 → 172.