Affect Primitives

Document type: Architecture reference
Scope: V3 harm-affect register (SD-019 cluster); benefit/wanting register deferred to V4
Last updated: 2026-04-20
Related claims: SD-011, SD-019, SD-019a, SD-019b, MECH-219, Q-036, SD-032 cluster

Motivation

The clinical and experimental pain literature draws a firm three-way dissociation among dimensions of the pain experience that are commonly collapsed into a single “harm signal”:

Dimension	What it measures	Controllability effect	Neural substrate
Intensity	Sensory magnitude of noxious input	None	Lateral pathway: A-delta, S1/S2, VPL thalamus
Unpleasantness	Immediate affective valence — “how bothersome right now”	None (Loffler et al. 2018)	Medial pathway: C-fiber, ACC/insula, fast component
Suffering	Accumulated psychological burden	Selectively reduces suffering (Loffler et al. 2018)	Medial pathway: ACC/insula + salience network (slow, persistent)

The Löffler et al. (2018, Pain Reports) three-way dissociation is the empirical anchor: identical stimuli, identical unpleasantness ratings, but selectively reduced suffering under controllable conditions. Controllability is definitionally orthogonal to intensity; that it also leaves unpleasantness unchanged means suffering is a third, functionally separate construct requiring its own computational account.

Salomons et al. (2004, JNeurosci) confirm the neural substrate: perceived controllability modulates ACC/insula activation for identical stimuli — the suffering circuit, not the sensory circuit.

De Ridder et al. (2021, Neurosci BioBehav Rev) map this anatomically: the medial pathway (ACC + insula, overlapping with the salience network) is the substrate of suffering; “salience and context” — not sensory magnitude — determine the medial pathway’s output. This directly connects suffering computation to the SD-032 salience-network coordinator cluster.

The Harm Affect Register

REE V3 maintains three named harm affect primitives. Each has a computational definition, a substrate, and a distinct set of downstream effects.

Primitive 1: harm_intensity

Field	Value
REE signal	`z_harm_s` (sensory-discriminative harm stream)
Claim	SD-011 (dual nociceptive streams)
Character	Instantaneous; tracks current hazard proximity / noxious contact magnitude
Time constant	Fast — rises and falls with the hazard state within a single step
Controllability dependence	None
Biological analogue	A-delta fiber (fast nociception), spinothalamic tract, S1/S2, VPL thalamus

Downstream effects:

SD-029: comparator input for agency attribution (residual = z_harm_s_observed − E2_harm_s(z_harm_s_{t-1}, a_actual))
E3 trajectory scoring: immediate penalty weight on trajectories that approach hazard zones
SD-020: harm prediction-error weighting

Primitive 2: harm_unpleasantness

Field	Value
REE signal	`z_harm_un` (new; pre-integration affective valence channel)
Claim	SD-019a
Character	Immediate affective valence — the “I want this to stop now” signal
Time constant	Fast-to-medium; faster than suffering, slower than raw sensory intensity
Controllability dependence	None (Loffler et al. 2018 key finding)
Biological analogue	C-fiber input to ACC/insula (fast medial component); unpleasantness dimension of the McGill Pain Questionnaire

Downstream effects:

AIC urgency input (SD-032c): z_harm_un is the primary source of the urgency salience signal driving the salience-network coordinator
E3 short-horizon trajectory scoring: negative valence weight distinct from z_harm_s (a trajectory ending harm NOW scores differently from one that reduces future harm probability)
MECH-219 input: z_harm_un is what MECH-219 integrates over time to produce suffering (z_harm_a)

Design note: z_harm_un is not required to be nonredundant with z_harm_s at every instant — the two can be correlated under typical harm conditions. The nonredundancy constraint (SD-019) applies between z_harm_s and z_harm_a (suffering), not between z_harm_s and z_harm_un. However, z_harm_un must encode something beyond raw intensity — it is a medial-pathway signal that includes affective/motivational weighting — so complete collinearity with z_harm_s would be architecturally wrong.

Primitive 3: harm_suffering

Field	Value
REE signal	`z_harm_a` (affective harm stream; refined definition)
Claim	SD-019b (suffering accumulator); mechanism SD-019 / MECH-219
Character	Slowly-accumulated load state; persistence depends on exposure duration and controllability
Time constant	Slow; asymmetric onset/recovery (MECH-219: rise faster than decay)
Controllability dependence	High — escapability belief gates the accumulator (Q-036)
Biological analogue	Chronic activation of medial pathway + salience network; overlap with stress system; learned helplessness substrate (DRN serotonin, vmPFC control circuit)

Downstream effects:

MECH-259 / SD-032a: primary input to the mode-switch threshold trigger; when suffering exceeds threshold (modulated by SD-032d PCC stability), the coordinator fires a whole-system operating_mode switch
SD-032e pACC autonomic coupling: slow write-back to drive_level via long-horizon integration of z_harm_a
ARC-016 urgency modulation: z_harm_a feeds commitment-threshold adjustment in E3
Behavioral withdrawal / learned helplessness: when z_harm_a is sustained and controllability estimate is low, the agent should transition from approach/escape to mode-change or help-seeking — the suffering primitive is what grounds this transition biologically

Computational Pipeline

SENSATION
  │
  ├─ z_harm_s (intensity)  ──►  E3 trajectory scoring (immediate avoidance)
  │         │                   SD-029 comparator (agency attribution)
  │         │                   SD-020 harm PE
  │         │
  │         ▼
  │   [medial pathway encoding]
  │         │
  ├─ z_harm_un (unpleasantness) ──►  AIC urgency signal (SD-032c)
  │         │                        E3 short-horizon scoring ("stop this NOW")
  │         │
  │         ▼
  │   [MECH-219: hysteretic integrator, controllability-gated]
  │         │
  └─ z_harm_a (suffering) ─────►  MECH-259 mode-switch trigger (SD-032a)
                                   SD-032e pACC → drive_level coupling
                                   ARC-016 commitment-threshold adjustment
                                   Behavioral withdrawal / mode-change

The nonredundancy constraint (SD-019) enforces structural separation between z_harm_s and z_harm_a; it operates at the suffering level, not the unpleasantness level.

What MECH-219 Actually Computes

Updated specification given the three-primitive model:

MECH-219 is a hysteretic integrator that takes z_harm_un (harm_unpleasantness) as input and produces z_harm_a (harm_suffering). The conversion is not a pure temporal integral: the accumulation rate is modulated by an escapability estimate (Q-036: controllability/inescapability signal), such that identical exposure to z_harm_un produces higher z_harm_a when harm is perceived as inescapable than when it is perceived as controllable.

Falsifiable predictions (updated):

After harm offset, z_harm_un falls rapidly while z_harm_a remains elevated (persistence property).
Two trajectories matched on current z_harm_un but with different controllability histories produce different z_harm_a.
An agent with high z_harm_a but moderate z_harm_un (chronic load, current hazard manageable) should exhibit mode-switching behavior (SD-032a trigger) that an agent with high z_harm_un but low z_harm_a (acute spike, not yet accumulated) should not.

Open Questions (Q-036 Bridge)

Q-036 asks: beyond temporal integration, which variables are required for suffering to be a genuinely distinct load state?

The Löffler dissociation establishes that controllability is load-determining and non-temporal. The current evidence (three Q-036 entries, lit_conf=0.812) supports adding an escapability-belief signal into MECH-219’s integrator. The following variables from Q-036 remain open:

Variable	Evidence status	REE implication
Controllability / escapability	Supported (Salomons 2004, Loffler 2018)	Gate MECH-219 accumulation by escapability estimate
Persistence / recovery failure	Suggestive (De Ridder 2021 acute→chronic transition)	Asymmetric decay already in MECH-219; should be validated
Inescapability (learned helplessness)	Indirect (literature pull needed)	Maps to a separate learned state on top of the accumulator
Prediction error weighted threat	Open (Fazeli & Buchel 2018 partial)	PE may modulate unpleasantness encoding, not suffering directly

Extension Register: Beyond Harm

The harm register is V3-complete (or nearly so). The affect classes beyond harm are consolidated here. This register was reconciled on 2026-06-05 against the three proto-feelings audit lit-pulls (thought_intake_2026-06-01_protofeelings_audit_register.md §7) so that every confirmed-distinct stream is represented — previously the table listed only a single under-specified “Relief” row and had not caught up to the registered safety cluster (MECH-303/304) or the newer streams. claims.yaml status is tracked per row; rows marked gated still need a claims.yaml claim (a follow-up governance decision, not done here).

Affect class	REE signal	Claim backing	Status / scope
Benefit / resource	`z_benefit` / `z_goal`	MECH-112 (wanting)	V4-deferred (full benefit register)
Relief	offset/derivative of harm stream	MECH-302	V3 (registered claim)
Safety	learned threat-absence predictor	MECH-303 (contextual) + MECH-304 (cue-specific)	V3 (registered claims)
Soothing / comfort	autonomic state-gain modulator	MECH-355 (candidate, V4-social)	V4-social (+ optional V3-minimal hook)
Autonomic rebound (endogenous parasympathetic recovery)	offset-triggered recovery-rate boost (MECH-219 recovery + SD-032e leak)	MECH-356 (candidate, V3)	V3-tractable sibling of MECH-355 (endogenous trigger; gated build)
Blocked agency (control-failure)	`z_block` (new)	MECH-353 (candidate, V3)	V3-candidate (lit-grounded; pending experiment)
Coercion / domination / injustice	(not yet specified)	—	V4-deferred (social; needs other-agent model)
Effort / fatigue (stop-recover)	`drive_level` (SD-012) + SD-048 interoceptive	MECH-354 (candidate, V3)	V3-minimal, gated
Social harm	(not yet specified)	—	Research phase

Each stream is differentiated from its neighbours below. The differentiations honour feedback_biology_before_formal_definitions: distinct biological systems get distinct rows, never one merged primitive (the SD-010→SD-011 philosophy-right/mechanism-wrong failure). The three 2026-06-05 pulls are: targeted_review_affect_stream_relief_safety_soothing (relief/safety/soothing), targeted_review_fatigue_vs_helplessness_dissociation (fatigue/stop-recover vs suffering), and targeted_review_blocked_agency_anger_stream (blocked-agency/anger). Based on articles retrieved from PubMed.

Extension Primitive (V3, registered): relief / safety

Relief (MECH-302) is an event-locked, phasic, value-coding reinforcement signal at the offset of an aversive event — it reuses the reward/goal machinery (NAc/VTA; D1+NMDA coincidence, Bergado Acosta et al. 2017) reading the SD-011 suffering derivative. It tags “this reduced suffering.”

Safety (MECH-303 contextual passive + MECH-304 cue-specific conditioned inhibition) is a learned prospective predictor that threat is absent — an active prefrontal-hippocampal-thalamic representation (mPFC IL/PL; nucleus reuniens→BLA, Silva et al. 2021; Corches et al. 2019), not mere low harm. It licenses commitment-release and approach.

Differentiation: relief is past-offset / phasic / appetitive (reinforcer); safety is future / tonic / inhibitory (predictor/gate). Neither is the other, and neither is z_goal/wanting (MECH-112) — safety is inhibitory prediction, not appetite. Enrichment APPLIED 2026-06-05 (not contradiction): Silva 2021 (NRe→BLA, conf 0.80) implicated a midline-thalamic relay (nucleus reuniens) and a remote-vs-recent (time-since-encoding) dependence absent from MECH-303/304’s stated anatomy. Both are now folded into the claims: MECH-304 carries the reuniens transmission node + the remote/recent axis (Silva bears most directly on the active cue-specific pole); MECH-303 carries the lighter cluster-level note that the safety-expression machinery is prefrontal-hippocampal-thalamic. The consolidation-age axis is recorded on MECH-304 as a candidate third safety sub-mechanism (consolidation-state-conditional safety expression interfacing the MECH-092/sleep replay machinery), deliberately not minted yet (single rodent anchor, off the V3 critical path, no consolidation-interface design). Thalamic relay + remote/recent axis are documentation-only (UNIMPLEMENTED in SD-051/SD-052). Evidence: targeted_review_affect_stream_relief_safety_soothing/ (verdict, conf 0.82).

Field	Value
REE signal	autonomic state-gain modulator (`MECH-355`)
Status	MECH-355 (candidate; V4-social, `substrate_conditional`) — registered 2026-06-05; V4 substrate + update rule are later design decisions
Character	Down-regulation of the ongoing stress response (present-tense), canonically via a conspecific (social buffering)
Biological analogue	oxytocinergic + parasympathetic/HPA + prefrontal regulation (Hostinar et al. 2014; Heinrichs et al. 2003)

Computational role: a state-gain / recovery modulator that lowers the gain or speeds the recovery of the active aversive trajectory — proposed home is MECH-219 (suffering accumulator) decay + SD-012 (drive) + SD-032e (pACC autonomic coupling).

Differentiation: soothing acts on the present stress trajectory, where relief is past-offset and safety is future-prediction; its substrate is oxytocin/parasympathetic/HPA (vs NAc/VTA for relief, vs cortico-amygdalar for safety); its role is state-gain modulation (vs reinforcer, vs predictor). It is socially gated — the canonical trigger is a conspecific, which V3 cannot represent. Do NOT fold into Relief (MECH-302) or Safety (MECH-303/304), and NOT into MECH-112. Scope: V4-social primary. The non-social V3-minimal autonomic-recovery hook that the verdict left optional is now split out as its own sibling claim, MECH-356 (endogenous parasympathetic recovery / autonomic rebound) — same effector/update rule, endogenous offset trigger instead of a conspecific. MECH-355 stays the social, V4 arm; do NOT fold the two together. When the MECH-355 substrate is built it must depend on MECH-219 + SD-032e (the affective/autonomic drive_bias component, NOT base SD-012 drive_level) and be explicitly NOT-302/303/304/112. Evidence: targeted_review_affect_stream_relief_safety_soothing/ (verdict, conf 0.82). Update-rule

scope design pass: evidence/planning/mech_355_soothing_update_rule_and_scope_design_2026-06-05.md.

Extension Primitive (V3-candidate; sibling of soothing): autonomic rebound / endogenous parasympathetic recovery

Field	Value
REE signal	offset-triggered transient boost to the recovery side of the slow affective accumulators (`MECH-356`)
Status	MECH-356 (candidate; V3, `epistemic_category: standard`, `v3_pending`) — registered 2026-06-06; build gated behind in-flight cue-authority / z_goal work (off the V3 critical path)
Character	Active, present-tense acceleration of return-to-baseline after the agent’s OWN stressor ends — not the passive fade of the aversive signal
Biological analogue	vagal rebound at stressor offset despite residual sympathetic activation; central-autonomic-network (prefrontal-vagal) recovery driver (Mezzacappa et al. 2001; Thayer & Brosschot 2005; Cunha et al. 2015)

Computational role: the same decay-acceleration effector as MECH-355 soothing — a transient, state-scaled multiplier on the MECH-219 z_harm_a recovery_rate (recovery side only; onset untouched) and the SD-032e drive_bias leak — but driven by an endogenous trigger: offset-detection (d||z_harm_a||/dt < 0, the MECH-302 SufferingDerivativeComparator signal), with no conspecific and no prediction. Load-proportional (multiplicative on existing state → zero effect on a calm agent → recovery ≠ sedation). Never writes base SD-012 drive_level, z_harm_s, or z_harm_un. MECH-094 waking-only.

Differentiation: sibling of MECH-355 (same effector, endogenous trigger vs conspecific — register as siblings, not a fold-in: the literature shows social support modulates the same vagal-rebound variable, Tung et al. 2021). Distinct from passive MECH-219 recovery (active overshoot, not decay), from MECH-302 relief (shared offset trigger, different output — relief writes a reinforcer, rebound boosts recovery rate), and from MECH-303/304 safety (no prediction). Closes REE’s sympathetic-rich / parasympathetic-poor asymmetry. Evidence: targeted_review_autonomic_rebound_parasympathetic_recovery/ (verdict, conf 0.74).

Extension Primitive (V3-minimal, gated): effort / fatigue (stop-recover)

Field	Value
REE signal	`drive_level` (SD-012) + SD-048 interoceptive channel — a second homeostatic deficit alongside hunger (`MECH-354`)
Status	MECH-354 (candidate, V3-minimal) — registered 2026-06-05; `v3_pending` until a discriminative experiment; build gated behind in-flight cue-authority/z_goal work
Character	A stop-AND-recover signal: accumulates with effort/time-on-task, recovers with rest/sleep
Antecedent	Cumulative effort / time-on-task, controllability-INDEPENDENT — accumulates from your own successful effort (Boksem & Tops 2008; Meyniel et al. 2013; Borbély et al. 2016)
Biological analogue	interoceptive cost-accumulator (posterior insula) + ACC effort-valuation; adenosine/Process-S for the slow sleep-pressure variant

Smallest computational form (Meyniel 2013): a two-bound (hysteretic) leaky cost-evidence accumulator — F += Se*effort during exertion, F -= Sr during rest; emit STOP at the upper bound, recover to the lower bound before re-engaging. Two time-constants: a fast within-task accumulator and a slow sleep-pressure accumulator (offline reset via SD-017 sleep). Se/Sr/bound-gap are incentive-modulated (incentive → work closer to exhaustion).

Differentiation (vs suffering / learned helplessness, SD-011 side): they converge behaviourally on stop/disengage but separate on four axes — antecedent (effort-depletion vs uncontrollable aversion), substrate (interoceptive cost-accumulator vs DRN 5-HT/CRF), computation (value recalibration, incentive-reversible vs escape-failure), and decisively recovery: fatigue remits with rest, helplessness does NOT (needs controllability/pharmacological reversal). Do NOT wire fatigue as a z_harm / SD-011 nociceptive stream — that imports the wrong antecedent, gate, and dynamics. This is the upstream fix for the DEV-NEED-002 conflation (“harm, hunger, fatigue blur together”).

Consumers: the in-task STOP feeds MECH-342 (commit-maintenance release) as a new deficit input (alongside the R-c execution-readiness deficit) and the ARC-078 goal-disengagement consumer via a cost/benefit (not aversive) channel — the release actuator is shared with the suffering pole, reached from the opposite antecedent. The recover / re-accumulation half is NOT in MECH-342; it belongs on the SD-012 / SD-017 side. Evidence: targeted_review_fatigue_vs_helplessness_dissociation/ (VERDICT).

Extension Primitive (V3-candidate): blocked_agency

Field	Value
REE signal	`z_block` (blocked-agency / control-failure stream; `MECH-353`)
Status	MECH-353 (candidate, V3) — registered 2026-06-05; lit-grounded, `v3_pending` until the discriminative experiment lands
Character	Rises when an intended/expected action-outcome is repeatedly blocked while the goal and the agent’s capacity-belief are retained
Antecedent	A negative disparity between expected and realised action-outcome — an expectation/agency violation, not noxious input (frustrative non-reward; Papini et al. 2024)
Time constant	Medium — integrates blocked attempts over a window; distinct from instantaneous harm
Controllability dependence	High, but opposite polarity to suffering: `z_block` is the capacity-RETAINED pole (assert), where `z_harm_a` is the capacity-COLLAPSED pole (withdraw)
Biological analogue	RAGE as a distinct primary-process system (Davis & Montag 2019); frustrative non-reward circuit (Papini et al. 2024); reactive-aggression assert channel under prefrontal gating (Bertsch et al. 2020)

Computational form (smallest): z_block is layered on substrate REE already has —

Detector (exists): the agency comparator (SD-029) applied to the action-outcome / goal channel — intended effect predicted by the forward model (E2), realised effect diverges (the comparator-model account of agency and its breakdown; Carruthers 2012).
Expectation (exists): z_goal / wanting (MECH-112) supplies the “expected outcome.”
New work — readout + two gates: integrate comparator mismatch over a window → z_block ↑; an attribution gate (mismatch caused by external constraint, not own motor error); a capacity gate (fires as assert only while goal/capacity is retained; hands off to z_harm_a as capacity-belief collapses).

Differentiation (each neighbour, explicitly):

vs harm (SD-011, z_harm_*): harm’s antecedent is noxious contact; z_block fires with zero noxious input (a merely-blocked/omitted outcome). Distinct antecedent, distinct biology (RAGE circuit vs nociceptive medial pathway). Collapsing them would repeat the SD-010→SD-011 philosophy-right/mechanism-wrong error.
vs suffering (SD-019b, z_harm_a + Q-036): same controllability axis, opposite pole. Capacity-belief collapsed → withdraw (suffering, already in REE); capacity-belief retained → assert (z_block). REE currently encodes only the withdraw pole — the energised assert pole is the piece this primitive adds.
vs residue (MECH-056): residue is the trace of an action taken at a value-cost; z_block is an action prevented. Opposite causal structure, different consumer.
vs commitment-hold (MECH-090 beta-gate): MECH-090 is the licit, self-imposed gating of E3→action during a committed sequence; z_block is an externally-imposed block against a live goal. Self-hold (licit) vs external-block (the new signal).

Consumers it drives:

Assert / escalate-effort / try-different (new behavioural pole) — raise drive/vigor (MECH-320), search an alternative action that restores the intended outcome (reactance direct-restoration; the adaptive analogue of reactive aggression).
Decommit (MECH-342) — if assertion fails across the window, release the blocked commitment rather than escalate unboundedly; gated by the commitment-threshold (ARC-016), the prefrontal-analogue of reactive-aggression inhibition (Bertsch et al. 2020).
Withdraw — not native to this stream; reached only via hand-off to z_harm_a when capacity-belief collapses.

Scope: V3-tractable as a single-agent proxy — “an intended, predicted-to-succeed action is repeatedly blocked by the environment/constraint.” The coercion / domination / injustice extension (Stream B below) requires modelling an other agent as the source of the restriction and is V4-deferred (psychological reactance / autonomy-threat, Steindl et al. 2015; injustice/norm-violation appraisal). Anti-domination and anti-exploitation are inherently other-agent — defer; cross-link the V4 ethics cluster (thought_intake_2026-05-31_musings_on_v4.md).

Smallest V3 test (gated; not yet queued): an environment that repeatedly blocks an intended, predicted-to-succeed action while harm and goal-value are held constant; measure whether a comparator-mismatch readout rises, whether it drives assert/persist (effort escalation / alternative-action search) distinct from the withdraw signature, and whether it dissociates from z_harm_a under matched controllability.

Evidence: evidence/literature/targeted_review_blocked_agency_anger_stream/ (5 entries + VERDICT). Based on articles retrieved from PubMed.

Lit-pull status

Primitive	Coverage	Next pull
harm_intensity	Strong (SD-011 targeted reviews)	—
harm_unpleasantness	Partial (Rainville 1997 in SD-019 pulls; SD-019a not yet targeted)	Needed — medial pathway fast component, ACC/insula unpleasantness encoding
harm_suffering	Moderate (Q-036 pull 2026-04-20, 3 entries)	Needed — inescapability / learned helplessness; prediction error in suffering
relief	Strong (`relief_completion_mechanism` 2026-05-03 + `affect_stream_relief_safety_soothing` 2026-06-05)	— (MECH-302 grounded)
safety	Strong (`affect_stream_relief_safety_soothing`, 2026-06-05)	— thalamic-relay (reuniens→BLA) + remote/recent enrichment APPLIED 2026-06-05 to MECH-303/304; candidate third (consolidation-age) sub-mechanism flagged, not minted
soothing / comfort	Strong (`affect_stream_relief_safety_soothing`, 2026-06-05, 2 entries)	— (V4-social; claim registration is the gated next step)
effort / fatigue (stop-recover)	Strong (`fatigue_vs_helplessness_dissociation`, 2026-06-05, 5 entries)	— (SD-012-side MECH registration is the gated next step)
blocked_agency	Strong (`targeted_review_blocked_agency_anger_stream`, 2026-06-05, 5 entries)	— for the V3 row; Stream B (coercion/injustice) is a V4 pull when the social substrate exists

Affect Primitives

Motivation

The Harm Affect Register

Primitive 1: harm_intensity

Primitive 2: harm_unpleasantness

Primitive 3: harm_suffering

Computational Pipeline

What MECH-219 Actually Computes

Open Questions (Q-036 Bridge)

Extension Register: Beyond Harm

Extension Primitive (V3, registered): relief / safety

Extension Primitive (V4-social; V3-minimal hook optional): soothing / comfort

Extension Primitive (V3-candidate; sibling of soothing): autonomic rebound / endogenous parasympathetic recovery

Extension Primitive (V3-minimal, gated): effort / fatigue (stop-recover)

Extension Primitive (V3-candidate): blocked_agency

Lit-pull status