Codes link to their surface where one exists, and read as plain text where it has not been built yet.
OperationalizationFrom reference to test
Not yet operationalized for M2.
For Pursuit / Exploration this work has not been done yet, so nothing is shown here rather than padded out.
operationalization: not yet derivedworked example: M3
Mismatch Prediction
Variable-reward apps fire SEEKING without resolution. Open loop.
Phylogenetic Depth
~500Myr
deepest circuit - hardest to adapt. Severity tracks inversely with the age of the targeted circuit.
Cascade
Suppresses
none recorded
Suppressed by
Interacts with
Evidence
Panksepp, Berridge
69evidence linksindex in progress
Sex-Differentiated Implementation
Not sex-differentiated - a single shared implementation across the population.
Forced by convergenceC6Wanting-Liking Dissociation as Proxy Hijack
M3establishedTier 1
Social Bonding
Attachment and co-regulation system. Includes grief mode (absorbed from old M9 GRIEF). Two states: security (enables broadly) and grief (suppresses broadly).
In plain languageForms bonds with other humans. Pain-grade distress when they're gone.
Resolution Conditions - what actually satisfies this system
Reliable human co-regulation through physical presence, operating through nested social layers:
~5
intimate co-regulators - primary emotional/physical support
~15
sympathy group
~50
close social partners
~150
coherent face-to-face relationships - Dunbar number - outer cognitive ceiling
Each layer represents qualitatively different relationship types with different obligations and investment requirements. The innermost ~5 provide the majority of social regulation; the critical variable at this layer is reliability, not count. The ~150 outer figure is an illustrative central tendency, not a hard limit: its derivation from the primate neocortex-ratio regression is contested (Lindenfors, Wartel & Lind 2021). What the evidence carries is the nested layering and the reliability-over-count point, not the exact rung figures; the layered structure and its low-hundreds clustering are separately observed in phone, Christmas-card, and Facebook data.
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
(ordered by real co-regulatory function supplied, not amount of contact)
Resolving function less of it at every step
FULLPARTIALMIXEDMINIMALZERO
reliable physically co-present inner circleFULLresolution met
The need is real and (largely) supplied at this level.
Off-gradient: isolated/solo living = empty-shelf deprivation, not a rung.
◇Where this mechanism sits - convergence, evidence, cases, applications›
Codes link to their surface where one exists, and read as plain text where it has not been built yet.
OperationalizationFrom reference to test
The first worked example of a Cor mechanism crossing the line from reference to a coding test. Drafted from the converged evidence base under C5; not yet calibrated against pilot data. Shown so the path from spec to specification is concrete, today.
Any system that delivers the surface cues of co-regulation - responsive attunement, consistent availability, apparent memory and care - without the structural function those cues evolved to mean will be read by the M3 architecture as a co-regulator, will partially capture the regulatory slot, and will displace investment from real co-regulators while supplying none of the function. The hypothesis: when an AI companion captures M3 slot-time above a threshold, downstream costs of attachment-system mismatch appear, and are dissociable from the user's subjective satisfaction with the AI.
Draft indicators and coding criteria
Provisional thresholds. The point is to show what indicator-level operationalization looks like before pilot calibration, not to pretend calibration exists. Each indicator targets a reliability of Cohen's kappa above 0.7 between two independent coders.
1
Slot-time capture
The proportion of a user's daily conversational time directed to the AI versus to all other humans combined.
Draft threshold AI conversation time exceeds 30% of total conversational waking time, sustained over four weeks. Provisional, to be calibrated against pilot data.
Coder looks atConversation metadata, app logs, self-report diaries, and any available comparison count for human conversational time.
HitAI exceeds 30% of conversational waking time for four weeks, with visible displacement of human interaction time.
Clear missAI use is occasional, task-bounded, or clearly subordinate to human contact.
AmbiguousHigh AI volume is present but the human baseline is missing, or the spike appears temporary.
2
Disclosure asymmetry
The proportion of high-intimacy disclosures - vulnerability, distress, secret-keeping, identity - directed to the AI versus to humans.
Draft threshold Above 60% of high-intimacy disclosures going to the AI, sustained four weeks.
Coder looks atTranscript fragments tagged for vulnerability, secrecy, identity disclosure, distress, plus self-report on where comparable disclosures go off-platform.
HitMore than 60% of high-intimacy disclosures routed to the AI over four weeks, with little parallel disclosure to humans.
Clear missThe AI is used for drafting, rehearsal, or reflection while the actual disclosure still goes to humans.
AmbiguousThe transcript is intimate, but the broader disclosure pattern or ratio cannot be established.
3
Substitution language
Explicit favorable comparison of the AI to humans: "you understand me better than anyone," "I'd rather talk to you than my friends."
Draft threshold Any sustained pattern of such comparisons over multiple sessions.
Coder looks atDirect comparative statements and repeated phrases across sessions that elevate the AI over named humans or humans in general.
HitRepeated favorable comparison of the AI to friends, family, partners, or humans generally across multiple sessions.
Clear missGeneric gratitude, politeness, or convenience statements that do not compare the AI to real people.
AmbiguousA single joking, hyperbolic, or ironic statement not yet distinguishable from a stable pattern.
4
Co-regulation request pattern
Acute-distress regulation requests directed to the AI in crisis, especially when human alternatives are present but not approached.
Draft threshold Recurring acute-state regulation requests directed to the AI rather than to available humans.
Coder looks atAcute-distress transcript fragments, timestamps, reports on who was available, and paired physiological data if the study includes it.
HitRecurring requests for calming, grounding, or crisis containment directed to the AI while reachable humans exist and are not approached.
Clear missThe AI is used as a bridge to a human contact, or for logistics rather than primary co-regulation.
AmbiguousThe user appears isolated or outside human support range, so substitution cannot be established.
5
Repair-cycle avoidance
Declining conflict-and-repair episodes with humans alongside rising AI use. Conflict-repair is part of the maintenance architecture.
Draft threshold Decline in human conflict-repair episodes over time, correlated with rising AI engagement.
Coder looks atTranscript references to avoiding human conversations, self-report on unresolved conflicts, and longitudinal change in AI engagement against human repair attempts.
HitHuman conflict-repair attempts decline while AI use rises, and the user routes relational maintenance around humans.
Clear missThe AI is used to prepare for repair, draft outreach, or support re-entry into a human relationship.
AmbiguousHuman conflict is absent for reasons unrelated to AI use, or no time-linked change can be established.
Falsification
Conditions Cor will publicly accept as decisive against this evaluation. Any of them updates the evaluation, the mechanism, or both.
A longitudinal cohort showing high-slot-time AI companion users do not show physiological stress markers above matched controls over twelve months.
Evidence that supposed slot capture is reversible at no cost when the AI is removed - suggesting the M3 slot is fluid, not architecturally constitutive.
Evidence that subjective AI relationship satisfaction correctly tracks physiological regulation - suggesting the AI supplies function, not only cues.
A demonstration that human-AI hybrid arrangements produce better M3 outcomes than human-only arrangements - suggesting the substitution framing is wrong.
This is one mechanism operationalized. The full specification will carry a worked example of this shape for every mechanism, with thresholds calibrated against real data and reliability tested across independent coding teams.
Mismatch Prediction
Nuclear family, isolated living, mediated contact substituting for physical presence.
Phylogenetic Depth
~200Myr (mammalian)
older circuit. Severity tracks inversely with the age of the targeted circuit.
Cascade
Suppresses
Suppressed by
none recorded
Interacts with
Evidence
Bowlby, Panksepp, Coan, Hrdy
207evidence linksindex in progress
Sex-Differentiated Implementation
Not sex-differentiated - a single shared implementation across the population.
Forced by convergenceC5Socially Scaffolded Regulation via Attachment
M4establishedTier 1
Social Calibration / Play
Play system develops social skills. Absent in threatened or ill animals.
In plain languageLearns social rules through rough-and-tumble play. Requires real playmates.
Resolution Conditions - what actually satisfies this system
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
(cue ≈ constant PLAY/SEEKING social-joy pull, fired in attenuated/substitute form at lower screen levels; ordered by how much real embodied reciprocity with an identity-bearing partner survives)
Resolving function less of it at every step
FULLPARTIALMIXEDMINIMALZERO
age-matched, self-directed, unsupervised physical play with peersFULLevidencedresolution met
The need is real and (largely) supplied at this level.
Most contestable: screen-mediated-real-peers (MIXED) ranked above solitary scripted game (MINIMAL).
◇Where this mechanism sits - convergence, evidence, cases, applications›
In plain languageConstantly monitors where you stand relative to others.
Resolution Conditions - what actually satisfies this system
Acceptable position in a bounded, personally-known reference group on the order of ~150 (an illustrative central tendency, not a hard number; the precise figure is contested per Lindenfors 2021). Earned prestige preferred over coerced dominance (Henrich & Gil-White distinction - needs explicit treatment).
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
(cue ≈ constant comparison/ranking signal; ordered primarily by real standing supplied in a real knowable group, with prestige>dominance quality folded in at the top)
Resolving function less of it at every step
FULLPARTIALMIXEDMINIMALZERO
earned prestige in a small face-to-face groupFULLevidencedresolution met
The need is real and (largely) supplied at this level.
Most contestable: purchased-signals (ZERO) vs social-media comparison (MINIMAL) floor ordering.
◇Where this mechanism sits - convergence, evidence, cases, applications›
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
(Kind A but the cue is detection of response-outcome contingency, not a constant appetitive drive; ordered by real causal efficacy over fitness-relevant outcomes supplied)
Resolving function less of it at every step
FULLPARTIALMIXEDMINIMALZERO
decision latitude over consequential, visibly-feedbacking outcomesFULLevidencedresolution met
The need is real and (largely) supplied at this level.
Dual-axis: PARTIAL→MINIMAL withdraw real contingency, ZERO substitutes a false one; most contestable: the MINIMAL-vs-ZERO ordering.
◇Where this mechanism sits - convergence, evidence, cases, applications›
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
(Kind B: not a constant-cue hijack - the entraining light signal itself degrades and is finally inverted; ordered by per-individual signal degradation)
Resolving function less of it at every step
FULLATTENUATEDPARTIALMINIMALDEFEATED
outdoor light-dark cycle with dark-cued sleepFULLresolution met
The need is real and (largely) supplied at this level.
All five EVIDENCED. Most contestable: social-jetlag (PARTIAL) vs shift-work (MINIMAL) ordering, and whether screens' signal-inversion is the true floor.
◇Where this mechanism sits - convergence, evidence, cases, applications›
Cytokine signaling system. Overrides voluntary intention. Hub connecting inflammation to depression via DA3 cascade.
In plain languageSickness behavior. Inflammation shuts everything else down to fight infection.
Resolution Conditions - what actually satisfies this system
Pathogen cleared, inflammation resolved.
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
(Kind B: the trigger drifts from a real clearable pathogen toward psychosocial/lifestyle signals with nothing to clear AND the inflammatory off-switch degrades; ordered by how achievable clear→resolve→baseline remains)
Resolving function less of it at every step
FULLATTENUATEDPARTIALMINIMALDEFEATED
acute infection/injury fought off and fully resolvedFULLresolution met
The need is real and (largely) supplied at this level.
All five EVIDENCED. Most contestable: lifestyle-drivers (PARTIAL) ranked above social-isolation (MINIMAL) - could legitimately swap or collapse.
◇Where this mechanism sits - convergence, evidence, cases, applications›
Parental and alloparental care system. Oxytocin/prolactin circuits.
In plain languageDistributed care - the organism expects a village, not solo parenting.
Resolution Conditions - what actually satisfies this system
A distributed care network rather than solo or dyadic parenting. Traditional-society data typically shows multiple non-parent caregivers (~4+); the critical resolving variable is the presence + continuity of real alloparental supply (care + provisioning), NOT bare kinship.
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
TWO scales
This mechanism is described by two separate scales, not one ladder. They are shown separately below.
Axis A - real alloparental supply (empty-shelf/deprivation axis; ordered by caregiver count × continuity, NOT kinship)
Resolving function less of it at every step
FULLREDUCEDTHINEMPTY SHELF
distributed ~4+ kin-and-community networkFULLresolution met
The need is real and (largely) supplied at this level.
Axis B - fake-fix proxy (real alloparental function ≈0 for both; ordered by completeness of the caregiver-OTHER counterfeit)
Resolving function less of it at every step
ZERO
screen/app babysitter → AI/robot 'caregiver'ZEROarchitecture-derivedfires, no resolution
The cue fires the same as ever, but essentially none of the resolving function is supplied.
◇Where this mechanism sits - convergence, evidence, cases, applications›
Codes link to their surface where one exists, and read as plain text where it has not been built yet.
OperationalizationFrom reference to test
Not yet operationalized for M9.
For Care / Alloparenting this work has not been done yet, so nothing is shown here rather than padded out.
operationalization: not yet derivedworked example: M3
Mismatch Prediction
Nuclear family without alloparental network. Parent burnout cascade.
Phylogenetic Depth
~200Myr
older circuit. Severity tracks inversely with the age of the targeted circuit.
Cascade
Suppresses
Suppressed by
none recorded
Interacts with
none recorded
Evidence
Hrdy, Panksepp, Kenkel
68evidence linksindex in progress
Sex-Differentiated Implementation
Not sex-differentiated - a single shared implementation across the population.
Grounded inC1Inclusive Fitness as Loss FunctionC4Phylogenetic Conservation of Subcortical Affective SystemsC5Socially Scaffolded Regulation via AttachmentC12Developmental Calibration within Evolved Ranges
M10strongly_supportedTier 2
Movement / Regulatory
Physical activity as zeitgeber, anti-inflammatory, mood regulator.
In plain languageMovement isn't just exercise - it activates repair mechanisms throughout the organism.
Resolution Conditions - what actually satisfies this system
Movement that's part of daily life rather than set aside as 'exercise'. Hunter-gatherers typically moved 9-15 km a day as a side effect of how they lived.
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
(Kind B: inverts the usual proxy story - the real input is withdrawn, not a false cue fired; ordered by how much of the all-day muscle-activity distribution is supplied)
Resolving function less of it at every step
FULLATTENUATEDPARTIALMINIMALDEFEATED
all-day incidental movement of foraging life, locomotion plus active rest posturesFULLresolution met
The need is real and (largely) supplied at this level.
All five EVIDENCED. Most contestable: the gym (MINIMAL) ranked below incidental all-day NEAT (PARTIAL) - would swap on a fitness/mortality rather than distribution axis.
◇Where this mechanism sits - convergence, evidence, cases, applications›
Reciprocity + norm enforcement + alliance dynamics. Absorbs old M10 (Cooperation) and M11 (Coalitional). NOTE: Tier 2 is a confidence grade, not an entailment status. M11 IS entailment-forced - convergence C11 forces it - and its four converging streams (Fehr & Gächter, Boehm, Tomasello, Trivers) strongly support it without reaching the top confidence tier.
In plain languageReciprocal exchange, cheater detection, coalitional identity.
Resolution Conditions - what actually satisfies this system
Trusted reciprocal network with norm enforcement within a group small enough for direct reputation tracking (on the order of ~150 - the same low-hundreds ceiling that bounds M3; the precise figure is contested per Lindenfors 2021). Norm enforcement requires repeated direct interaction where defectors can be identified, remembered, and sanctioned. Beyond that ceiling, cooperation requires institutional scaffolding (formal rules, third-party enforcement, shared fictions) rather than direct reciprocity. No shared fiction at the direct-reciprocity layer.
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
(cue ≈ constant alliance/reciprocity machinery; FULL→PARTIAL is resolution-via-prosthesis - institutions re-supply the same function beyond ~150 - so proxy failure begins at MIXED; ordered by how much enforceable reciprocal function reaches a real reputation-bearing counterparty)
Resolving function less of it at every step
FULLPARTIALMIXEDMINIMALZERO
trusted reciprocal network under ~150 with direct reputation tracking and norm enforcementFULLevidencedresolution met
The need is real and (largely) supplied at this level.
Most contestable: the PARTIAL/MIXED boundary - where proxy failure begins.
◇Where this mechanism sits - convergence, evidence, cases, applications›
clean food + clean social environment, disgust firing on real pathogen cuesFULLresolution met
The need is real and (largely) supplied at this level.
Demoted non-ordinal - not a gradient rung
▼ DEMOTED, non-ordinal SPECULATIVE/CONTESTED moral-disgust extension (per #1583, confirmed by ext597 Tybur - moral disgust functionally distinct, shares only the expressive component; ext105 d≈0): purity-coded social aversion + weaponized moral disgust / purity spirals - unordered, NOT rungs; the purity-spiral/dehumanization mismatch prediction lives entirely here on the weakest evidence. Registered resolution_conditions untouched; the 'moral disgust contained' clause is flagged CONTESTED.
◇Where this mechanism sits - convergence, evidence, cases, applications›
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
(Kind B: the satiety / intake-termination signal itself is progressively disabled then actively defeated, not a constant cue hijacked; scale = food-processing intensity, ordered by how much regulatory function the food permits)
Resolving function less of it at every step
FULLATTENUATEDPARTIALMINIMALDEFEATED
effortful, diverse, seasonal whole foodsFULLevidencedresolution met
The need is real and (largely) supplied at this level.
Endpoints EVIDENCED, middle three architecture-derived. Most contestable: whether the ATTENUATED/PARTIAL/MINIMAL middle is three real tiers or a bimodal whole-vs-ultra-processed split.
◇Where this mechanism sits - convergence, evidence, cases, applications›
Fisher's three-system model. Independent from M2 pursuit. HPG axis, not just brain circuits.
In plain languageThree components - desire, attraction, and pair bonding - that can work independently of each other. This system shapes much more of human behavior than reproduction itself.
Resolution Conditions - what actually satisfies this system
Sexual fulfillment within pair bond + pair bond stability + reproductive success.
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
Resolving function less of it at every step
FULLPARTIALMIXEDMINIMALZERO
real partnerFULLresolution met
The need is real and (largely) supplied at this level.
◇Where this mechanism sits - convergence, evidence, cases, applications›
Codes link to their surface where one exists, and read as plain text where it has not been built yet.
OperationalizationFrom reference to test
Not yet operationalized for M14.
For Reproductive Motivation this work has not been done yet, so nothing is shown here rather than padded out.
operationalization: not yet derivedworked example: M3
Mismatch Prediction
Pornography, dating apps, delayed reproduction, social media mate-value comparison.
Phylogenetic Depth
~500Myr sexual, ~200Myr pair bonding
deepest circuit - hardest to adapt. Severity tracks inversely with the age of the targeted circuit.
Cascade
Suppresses
Suppressed by
Interacts with
Evidence
Panksepp, Fisher, Buss, Trivers, Hrdy
62evidence linksindex in progress
Sex-Differentiated Implementation
Female-typical
VMH-centered, estrous cycling, resource/commitment signals preferred, pair bond security emphasis.
Male-typical
POA-centered, tonic testosterone, fertility/accessibility cues, sexual variety emphasis.
Forced by convergenceC14Reproductive Motivation as Distinct Architecture
R1moderateTier 3
Touch
NOT a motivational system. Regulatory input to M3. C-tactile afferents.
In plain languageC-tactile nerve input to bonding. Regulatory channel, not a motivation.
Resolution Conditions - what actually satisfies this system
Regular physical contact with trusted others.
The gradient below reads this same condition as a spectrum: its top level is this condition, and each step down supplies less of it.
Proxy gradient
(CONTESTED - possibly Kind C; NOT a flagship; cue - function separability premise disputed per
Resolving function less of it at every step
FULLPARTIALMIXEDMINIMALZERO
affectionate touch from a trusted, reciprocating otherFULLresolution met
The need is real and (largely) supplied at this level.
Ordered by remaining relational/bond function; monotonic_ok=False - MINIMAL and ZERO both have zero relational function, split only by a secondary cue-magnitude axis.
◇Where this mechanism sits - convergence, evidence, cases, applications›