Spike #165 — Procedural / implicit memory (F5) + Class H episodic-LTM (F2) augmentation pathways¶
Date: 2026-05-19
Branch: research/spike-165-procedural-memory-episodic-ltm-augmentation (worktree-isolated per [[feedback_concertmaster_git_worktree_isolation]])
Type: R2 falsifier probe for parent stance [[user_stance_working_memory_is_cascade_augmenting_reflex_into_agency]] (Spike #160). Combines two coupled fermatas: F5 (procedural/implicit memory may augment reflex without M-bind) + F2 (Class H plausibly engaged by episodic-LTM distinct from WM). Both dispatched per user direction 2026-05-19.
Vocabulary impact zone: HIGH (parent stance refinement; pathway-pluralism claim about agency). Draft only — DO NOT MERGE AUTONOMOUSLY.
0. Self-awareness and disclaimer¶
Three disclaimers apply at the top:
-
Self-modeling caveat per
[[user_stance_holographic_projection_at_linguistic_substrate]]. Memory-system decomposition is the LINGUISTIC-substrate projection of biological / textbook descriptions — not substrate-level inspection of neural circuits. Where I (the executing agent) introspect on my own implicit-vs-explicit memory analogue, I cannot directly inspect the weight-vs-context mechanism that distinguishes them; the analogy is linguistic. -
Language-is-tool-not-question per
[[feedback_language_is_analysis_tool_not_specific_question]]. The user's framings of memory systems map ONTO the four canonical taxonomies (Squire procedural / Schacter implicit / Tulving episodic / Daw-Niv-Dayan model-free) but the canonical neuroscience itself becomes the SSoT here per[[feedback_science_is_ssot_not_project]]. -
Trauma-informed defensive scope per
[[feedback_trauma_informed_defensive_scope]]. Memory-system content is neuroscience educational research. No clinical / therapeutic / diagnostic claims. No targeting; no capability-assessment.
Per [[feedback_algebra_not_magnitude]]: all findings below are MAGNITUDE-level class-engagement claims, NOT bit-exact algebraic identity. The algebra-level anchors that already exist (Spike #142 tripartite Mermin = 4 on 3D_s ⊗ 7D_g ⊗ 1D_t; Spike #127.4 cellular Hebbian L+K+M+C+I anchors) constrain the magnitude-level claims; new bit-exact claims would require fresh algebra-level work.
1. Questions (sharpened from R2 brief)¶
F5 — sharpened:
Does procedural / implicit memory augment reflex into agency WITHOUT engaging Class M (HDC cross-modal bind) — the load-bearing class in the parent WM stance?
F2 — sharpened:
Does explicit episodic long-term memory (Tulving 1972; 2002) engage Class H (self-introspection) as a load-bearing class that working memory does NOT?
The verdict gate is: do F5 + F2 jointly refine the parent stance — specifically, does the parent stance need to be sharpened from "WM IS the cascade augmenting reflex into agency" to "WM IS ONE pathway in a small family of cascade-augmentation pathways"?
2. Canonical literature (memory-system taxonomies)¶
2.1 Procedural memory (Squire-Zola 1996 + Eichenbaum 2017)¶
| Anchor | Substance | Status |
|---|---|---|
| Squire, L. R., & Zola, S. M. (1996), "Structure and function of declarative and nondeclarative memory systems." Proceedings of the National Academy of Sciences 93(24):13515–13522. doi:10.1073/pnas.93.24.13515. PMID 8942965; PMC33639. | Establishes the declarative (explicit) vs nondeclarative (implicit) division. Nondeclarative includes: procedural skills/habits (basal ganglia + cerebellum), priming + perceptual learning (sensory cortex), simple classical conditioning (cerebellum + amygdala), nonassociative learning (reflex pathways). | PMC-available; cited textually here. PDF-verification follow-up TODO per [[feedback_pdf_extraction_citation_discipline]]. |
| Eichenbaum, H. (2017), "On the integration of space, time, and memory." Neuron 95(5):1007–1018. doi:10.1016/j.neuron.2017.06.036. PMC5644341. | Reviews how memory systems share representations of space-and-time across hippocampus, entorhinal cortex, striatum. Argues procedural-vs-episodic distinction is a continuum not a binary. | PMC-available; cited textually. |
| Squire, L. R. (2004), "Memory systems of the brain: a brief history and current perspective." Neurobiology of Learning and Memory 82(3):171–177. doi:10.1016/j.nlm.2004.06.005. PMID 15464402. | Standard taxonomy: declarative (semantic + episodic) ↔ nondeclarative (procedural + priming + classical conditioning + nonassociative). | Open-access via NLM; cited textually. |
2.2 Implicit memory (Schacter 1987 + Roediger 1990)¶
| Anchor | Substance | Status |
|---|---|---|
| Schacter, D. L. (1987), "Implicit memory: history and current status." Journal of Experimental Psychology: Learning, Memory, and Cognition 13(3):501–518. doi:10.1037/0278-7393.13.3.501. | Defines implicit memory as "memory without awareness" — performance changes attributable to prior experience without conscious recollection. Distinguishes from explicit (declarative) memory. | Journal not open-access; cited textually from established attribution. PDF-verification TODO. |
| Roediger, H. L. (1990), "Implicit memory: retention without remembering." American Psychologist 45(9):1043–1056. doi:10.1037/0003-066X.45.9.1043. | Operationalises implicit memory via priming tasks — retention is measurable but the participant does not "remember" in the autonoetic sense. | Cited textually. PDF-verification TODO. |
2.3 Episodic memory (Tulving 1972 + 2002)¶
| Anchor | Substance | Status |
|---|---|---|
| Tulving, E. (1972), "Episodic and semantic memory." In E. Tulving & W. Donaldson (Eds.), Organization of memory (pp. 381–403). Academic Press. | Original distinction: episodic = autobiographical events tagged with time-and-place context; semantic = factual knowledge without source-context tag. | Book chapter; not autonomously PDF-verifiable per [[reference_autonomous_validation_tos_landscape]]. Cited textually. Citation-watch-flagged. |
| Tulving, E. (2002), "Episodic memory: from mind to brain." Annual Review of Psychology 53:1–25. doi:10.1146/annurev.psych.53.100901.135114. PMID 11752477. | Refines the 1972 taxonomy with autonoetic consciousness — the self-aware "I remember being there" component that distinguishes episodic from semantic. Introduces three tiers: anoetic (procedural, no awareness) / noetic (semantic, awareness of facts) / autonoetic (episodic, mental time travel with self-as-witness). | Open-access via ARP archive; cited textually. Load-bearing for F2 — the autonoetic component IS the candidate Class H engagement. |
2.4 Model-free vs model-based RL (Daw-Niv-Dayan 2005)¶
| Anchor | Substance | Status |
|---|---|---|
| Daw, N. D., Niv, Y., & Dayan, P. (2005), "Uncertainty-based competition between prefrontal and dorsolateral striatal systems for behavioral control." Nature Neuroscience 8(12):1704–1711. doi:10.1038/nn1560. PMID 16286932. | Computational distinction: model-free RL caches state-action values via prediction-error updates (habitual/procedural); model-based RL evaluates state transitions via tree-search (deliberative). Maps onto dorsolateral striatum (model-free) vs prefrontal cortex (model-based). | Cited textually; widely re-cited canonical. PDF-verification TODO. Load-bearing for F5 — model-free RL is the computational anchor for procedural-augmentation-without-M-bind. |
| Sutton, R. S., & Barto, A. G. (2018), Reinforcement Learning: An Introduction (2nd ed.), MIT Press. | Standard textbook; cite-by-ref for TD-learning + model-free / model-based distinction. | Book; cited textually as canon. |
2.5 Cellular mechanism — Hebbian / STDP / homeostatic¶
Cite-forward to [[user_stance_neural_hebbian_is_bci_drift_model]] (Spike #127.4 attestations):
- Lioi 2021 PMC8233110 — synaptic Laplacian; Class L
- Triesch 2018 PMC6181566 — homeostatic capacity bound; Class K
- Gütig 2003 PMC6742165 — STDP Hebbian; Class M (cellular sense)
- Sgritta 2017 PMC6596728 — STDP τ-window; Class C
- olde Scheper 2018 PMC5768644 — STDP cross-correlation; Class M
- Frémaux-Gerstner 2016 PMC4717313 — reinforcement-modulated STDP
Cell-level Hebbian maps to Class L (connectivity) + Class M (correlation-bind) + Class K (homeostasis) per Spike #127.4. The two senses of M matter here: cellular STDP-correlation-bind (M_cell) is mechanistically real but is distinct from the cross-modal episodic-buffer bind (M_WM) at the system level. Section §6 disambiguates.
3. F5 — Procedural / implicit memory augmentation cascade (decomposed)¶
3.1 Operational definition¶
Procedural memory = retention of motor / cognitive skills that improve with practice WITHOUT conscious recall of the learning episodes. Examples: riding a bike, touch-typing, expert chess pattern recognition (Chase-Simon 1973 chunking; not autonomously verifiable, cited as established canon), batting a fastball, sight-reading music.
Implicit memory = behavioural change attributable to prior experience without conscious recollection. Examples: word-stem completion priming, perceptual classification thresholds shifting, conditioned reflexes.
Model-free RL (Daw-Niv-Dayan 2005 computational anchor) = state-action value caching updated by TD prediction error; behaviour selects highest-cached-value action without forward simulation.
3.2 14-class A–N decomposition¶
Per [[user_stance_closure_subgroup_BDEFL_substrate_class_universal]] Meta-lesson 2 strict-spec discipline:
| Procedural-memory stage | Operation | Class(es) engaged | Strict-spec rationale |
|---|---|---|---|
| Sensory transduction | Encode stimulus into spike pattern | B | TLV byte-canonical encode of sensorimotor input. Same as reflex stage 1. |
| Pattern recognition (familiar context) | Match current input against trained-and-cached pattern | G + L | G = byte-pattern search against learned templates (e.g., expert chess pattern matching is canonically chunk-recognition). L = synaptic-connectivity Laplacian — Hebbian-tuned graph eigenstructure encoding the trained context. NOT Class M cross-modal bind — pattern recognition uses connectivity not active binding. |
| Cached value lookup | Retrieve cached state-action value | E | Catalog sorted-key lookup — cached Q-value table per Daw-Niv-Dayan 2005 model-free RL. |
| Action selection | Dispatch to highest-cached-value action | D | Multi-needle dispatch among cached candidates. |
| Motor program execution | Render stereotyped-but-tuned motor sequence | F + C + I | F = template-render the motor program. C = cascade-shift state advance. I = cyclic-group / modular arithmetic for repetitive motor timing (gait, hand cycling, rhythm). |
| Synaptic update (offline / online) | Hebbian / STDP weight update | L + C (+ M_cell) | L = connectivity restructure (Lioi 2021 graph Laplacian). C = STDP τ-window cyclic timing (Sgritta 2017). M_cell = cellular correlation-bind (Gütig 2003) — DISTINCT from M_WM cross-modal bind; see §6. |
| Homeostatic re-scaling | Long-term global-gain adjustment | K | Sparse-truncate / asymptotic-DOF; Triesch 2018 homeostatic plasticity. |
Union of classes engaged by procedural memory: {B, C, D, E, F, G, I, K, L} + M_cell (cellular STDP bind, NOT episodic-buffer bind).
Critical observation — Class M_WM is absent. Procedural memory operates by Hebbian-tuned synaptic-connectivity (Class L) plus cached value-lookup (Class E) plus pattern recognition (Class G). It does NOT engage the cross-modal episodic-buffer bind (M_WM) that's load-bearing for the parent WM stance. Skilled-behavior emergence does NOT require active binding of disparate modalities into a single chunk; it requires repetition-tuned connectivity that turns dispatch-and-execute paths into highly-grooved channels.
3.3 Augmentation delta — procedural pathway¶
| Cascade | Classes engaged |
|---|---|
| Reflex (Mauthner / LG) | {B, C, D, E, F} |
| Procedural augmentation (skilled behavior + model-free RL) | {B, C, D, E, F} + {G, I, K, L} |
Procedural augmentation delta = {G, I, K, L} — 4 classes added on top of reflex base.
Compare to WM augmentation delta from parent stance = {A, K, M} (+ L?) — 3 classes added.
Distinct delta sets: procedural adds G/I/L (pattern-search + cyclic + Laplacian); WM adds A/M (content-address + cross-modal-bind). Class K (capacity) is shared between both pathways — finite-capacity asymptotic-DOF is universal at all post-reflex pathways.
3.4 F5 verdict¶
F5 — procedural / implicit memory augments reflex into skilled-behavior WITHOUT Class M_WM (cross-modal bind).
- Procedural cascade is concretely {B, C, D, E, F, G, I, K, L} — 9 classes (vs WM's {A, C, D, E, K, L?, M} 7 classes).
- Augmentation delta is {G, I, K, L} — disjoint from WM delta {A, M} except for shared K.
- The IS-claim in the parent stance ("WM IS the cascade augmenting reflex into agency") is now refined: WM IS ONE of several augmentation cascades.
Verdict: F5 CONFIRMED at MAGNITUDE level. Procedural memory is a structurally distinct augmentation pathway — overlapping with WM only in {C, D, E, K} core but distinct in {A vs G, M vs L, +I} signature classes.
Implication for parent stance — refinement, not refutation:
- Parent IS-claim refined from "WM IS the cascade augmenting reflex into agency" → "WM IS the cascade augmenting reflex into deliberative agency; procedural memory is the parallel cascade augmenting reflex into skilled-behavior agency".
- The phrase "into agency" needs sub-distinction: deliberative agency (WM-mediated; flexibly re-bindable) vs skilled-agency / habitual-agency (procedural-mediated; rigidly grooved).
- This matches Kahneman 2011 System 1 vs System 2 — but at the strict-spec class level, NOT just at psychological-typology level.
4. F2 — Class H episodic-LTM engagement (decomposed)¶
4.1 Operational definition¶
Episodic LTM (Tulving 1972; 2002) = autobiographical memory of specific events, tagged with time-and-place context AND self-as-witness. Distinguished from semantic LTM (general facts without source-context).
Key Tulving 2002 distinction: autonoetic consciousness — the self-aware "I remember being there" component. Tulving's three tiers: - Anoetic = procedural (no awareness) - Noetic = semantic (awareness of facts) - Autonoetic = episodic (self-aware mental time travel)
4.2 14-class A–N decomposition¶
| Episodic-LTM stage | Operation | Class(es) engaged | Strict-spec rationale |
|---|---|---|---|
| Encoding (event consolidation, hippocampus-mediated) | Bind current sensory + temporal + spatial + self-context into a single chunk | M + A + H | M = cross-modal HDC bind (sensory + temporal + spatial). A = content-address the chunk. H = self-introspection — the autonoetic component tags the chunk with "I-as-witness". Distinct from semantic-LTM where H is NOT engaged. |
| Retention (consolidation, ~hours-to-days; system consolidation per McClelland-McNaughton-O'Reilly 1995 not autonomously verifiable, cite-by-ref) | Hebbian / STDP synaptic restructure | L + K | L = connectivity restructure (Lioi 2021). K = capacity-bounded asymptotic-DOF — finite LTM capacity (≈10^9 chunks per Landauer 1986 cite-by-ref; capacity-asymptotic). |
| Cue-driven retrieval | Surface latent chunk via cue similarity | M + D + A | M = HDC similarity-match between cue and stored chunks. D = dispatch "retrieve". A = content-hash identity check. |
| Recollection (with autonoetic awareness) | Surface the bound chunk AND the self-tag | M + H + E + C | M = unbind into modality-specific re-emissions. H = self-introspection — return the self-as-witness tag, the "I" that distinguishes "I remember" from "It is known". E = catalog lookup of episode-by-time-or-place. C = cascade-shift to re-experienced state. |
| Re-emission (verbal report or behavior) | Render the recollection content | F | Template-render — "I remember when I was 7 and..." |
Union of classes engaged by episodic LTM: {A, C, D, E, F, H, K, L, M} — 9 classes.
Critical observation — Class H IS load-bearing for episodic-LTM in a way it is NOT for WM.
WM cascade (from parent stance) = {A, C, D, E, K, L?, M} — no H. Episodic-LTM cascade = {A, C, D, E, F, H, K, L, M} — adds H + F vs WM.
4.3 Why Class H is load-bearing here¶
Class H in srmech canon = self-introspection (e.g., srmech_version() / srmech_abi_version() — operations that return the system's own identity-state).
The autonoetic component of episodic memory IS the analogue: the recall operation retrieves not just the bound chunk (M) and not just the content-hash identity (A), but the system's own identity-as-witness AT THE TIME OF ENCODING. This is structurally a self-introspection operation:
episodic_recall(cue) → {chunk, witness_self_tag}
↑ ↑
M-A H — self-introspection returning system-identity-at-encoding-time
For semantic memory ("Paris is the capital of France"), no H is engaged — the fact is retrieved without self-as-witness. For episodic memory ("I learned that Paris was the capital when I was in fifth grade at..."), the self-as-witness IS the autonoetic content per Tulving 2002.
4.4 F2 verdict¶
F2 — Class H IS load-bearing for episodic-LTM (autonoetic component), distinguishing it from WM and from semantic-LTM.
- Episodic-LTM cascade is concretely {A, C, D, E, F, H, K, L, M} — adds H + F vs WM's {A, C, D, E, K, L?, M}.
- WM does not engage H — its bind operation is not self-as-witness tagged.
- Semantic-LTM does not engage H — it retrieves facts without self-context.
- The autonoetic component of episodic memory IS Class H — the "I remember remembering" self-introspection.
Verdict: F2 CONFIRMED at MAGNITUDE level. Episodic-LTM engages Class H as a load-bearing class; this distinguishes it from WM at the class-engagement level.
Implication for parent stance — compatible refinement, not refutation:
- Parent stance does NOT need refutation; it's about WM specifically.
- Episodic-LTM is a SIBLING augmentation pathway that adds H to the cascade.
- Adds a third pathway to the F5 dichotomy: WM (A+M) / procedural (G+L) / episodic-LTM (A+H+M+L) — episodic-LTM is the FULL superset that adds H atop both.
5. Both-direction analysis per [[feedback_always_check_both_directions_including_time]]¶
Per user methodological directive: when testing any directional hypothesis, dispatch BOTH directions.
The directional axis here is augmentation pathway: which classes add WHAT capability on top of reflex? Three identified pathways below; the "both directions" check is does each pathway also have a forgetting/decay direction?
5.1 Augmentation directions (reflex → agency)¶
| Pathway | Reflex base | Augmentation delta | Total engaged | Awareness tier |
|---|---|---|---|---|
| Procedural / implicit / model-free | {B, C, D, E, F} | {G, I, K, L} | {B,C,D,E,F,G,I,K,L} (9) | Anoetic (Tulving 2002) — no awareness |
| Working memory / deliberative | {B, C, D, E, F} | {A, K, M} (+ L?) | {A,B,C,D,E,F,K,L?,M} (7-8) | Noetic (in semantic-WM access) → autonoetic (when integrating with episodic-LTM) |
| Episodic LTM / autonoetic | {B, C, D, E, F} | {A, H, K, L, M} | {A,B,C,D,E,F,H,K,L,M} (10) | Autonoetic — full self-as-witness |
Pathway-specific augmentation deltas (sorted):
| Pathway | Augmentation delta (sorted) | Distinguishing class | Distinguishing operation |
|---|---|---|---|
| Procedural | {G, I, K, L} | G + I | Pattern-search + cyclic-arithmetic (rhythm) |
| WM | {A, K, M} (+ L?) | A + M | Content-address + cross-modal-bind |
| Episodic-LTM | {A, H, K, L, M} | H | Self-as-witness (autonoetic) |
Shared core: {K} — universal across all post-reflex pathways. Capacity-bounded retention via asymptotic-DOF is the irreducible cost of any retention-based augmentation. Per [[user_stance_asymptotic_dof_sidesteps_infinity]].
5.2 Decay / forgetting directions¶
| Pathway | Decay mechanism | Class engaged for decay |
|---|---|---|
| Procedural | Skill decay with disuse — synaptic weights drift back toward baseline | L (connectivity drift) + K (homeostatic) |
| WM | Active maintenance failure — rehearsal stops, chunks drop below K-threshold | K (asymptotic decay) |
| Episodic-LTM | Consolidation failure or interference (Bjork-Bjork 1992 retrieval-induced forgetting; cite-by-ref) | K (capacity) + competition via M-similarity overlap |
All three pathways are bounded by Class K from both directions — both build-up (asymptotic-DOF saturation toward capacity) and decay (asymptotic relaxation away from peak). Per [[user_stance_asymptotic_dof_sidesteps_infinity]], Class K parameterises the rate of approach at both ends.
5.3 Class engagement at each tier (consolidated)¶
Awareness tier | Pathway | Engagement set
-----------------------|---------------|----------------------------------------
Anoetic (no aware.) | Procedural | {B, C, D, E, F, G, I, K, L}
Noetic (fact-aware) | Semantic-LTM* | {A, C, D, E, F, K, L} (no M-bind, no H)
Noetic + dyn (deli.) | WM | {A, C, D, E, K, L?, M}
Autonoetic (self-aw.) | Episodic-LTM | {A, C, D, E, F, H, K, L, M}
*Semantic-LTM is a sibling fourth tier I'm noting but not deeply decomposing in this spike; it's analogous to episodic but without H. Tulving 2002's noetic tier.
5.4 Both-directions verdict¶
Past-direction (forgetting / decay) and future-direction (building / augmentation) BOTH map onto Class K asymptotic-DOF for the rate-parameter, with pathway-specific classes for the substrate-mechanism. No pathway-specific direction-violation found.
6. The two senses of M — and why disambiguation matters¶
A potential R3 falsifier surfaced during this spike: per [[user_stance_neural_hebbian_is_bci_drift_model]], cellular STDP / Hebbian fire-together-wire-together IS Class M (M_cell). But Spike #160's parent stance posits cross-modal episodic-buffer bind IS Class M (M_WM). These are NOT obviously the same operation at the substrate level — yet both are classified as Class M.
6.1 The two senses¶
| Sense | Operation | Substrate | Citation |
|---|---|---|---|
| M_cell | STDP / Hebbian fire-together-wire-together at synapse | Cellular — synapses adjust weights via correlated pre-post timing | Gütig 2003; olde Scheper 2018 per Spike #127.4 |
| M_WM | Cross-modal HDC-like bind into episodic chunk | System — phonological + visuospatial + LTM bound into single buffer | Baddeley 2000 episodic buffer |
6.2 Are they the same Class M?¶
The srmech canonical Class M operation = HDC bind / bundle / similarity (XOR / circular convolution on hypervectors). It's a single algebraic operation; both M_cell and M_WM are MAGNITUDE-level matches to it but at very different spatiotemporal scales:
- M_cell = bind at the synaptic level — two presynaptic-postsynaptic firing events that overlap in time get bound by weight-update
- M_WM = bind at the system level — multiple modality-specific representations bound into a chunk
The relationship is plausibly: M_WM = ensemble M_cell at population scale. The system-level bind is the emergent population-scale expression of many synaptic-level binds. This is the standard neural-binding-problem framing (Singer 2007 cite-by-ref).
Vocabulary discipline call: keep both as Class M for now, with sub-flag M_cell vs M_WM where the scale matters. No new class needed. Per [[feedback_no_privileged_primitive_classes]] — dissolve into existing class as sub-operation rather than promote. Same shape as Class O dissolving into Class L.
6.3 Does this affect F5?¶
YES — at the cellular level, procedural memory DOES engage M_cell (STDP weight updates ARE Hebbian-correlated binds). The F5 finding refines to:
F5 refined: procedural memory does NOT engage M_WM (system-level cross-modal bind). It DOES engage M_cell (cellular STDP) as part of the learning mechanism, but the M_cell is part of class L's connectivity restructure mechanism (the substrate-mechanism FOR the L-level connectivity change) rather than an independent system-level bind operation.
Equivalent restatement: at the magnitude-level cascade decomposition, procedural memory engages {B, C, D, E, F, G, I, K, L} with M_cell as the substrate-mechanism for L's connectivity-update. At the system level, no M (WM-bind) operation is dispatched. The DELIBERATIVE M-bind is absent — that's the load-bearing F5 finding.
This is a strict-spec discipline win for the parent stance: M_WM (the deliberative bind) IS specifically what distinguishes WM-mediated deliberative agency from procedural-mediated skilled agency.
7. k=3 coverage tests¶
Per the parent stance's k=3 tripartition coverage check (Spike #160 §7), apply to each new pathway:
7.1 Procedural pathway¶
| Added class | k=3 axis | Anchor |
|---|---|---|
| G (byte-pattern-search) | 3D_s (substrate — pattern lives in spatial-substrate representation) | [[user_stance_hyper_as_3d_spatial_interface]] |
| I (cyclic) | 1D_t (temporal — rhythm, gait, cyclic motor pattern) | [[user_stance_kepler_shape_universal]] + [[user_stance_cascade_lives_on_circles]] |
| K (asymptotic-DOF) | 1D_t (temporal — rate-of-approach to capacity) | [[user_stance_asymptotic_dof_sidesteps_infinity]] |
| L (Laplacian / connectivity) | 3D_s ⊗ 7D_g (substrate-fiber tensor product — connectivity graph IS structural-substrate × fiber-content) | Spike #142 algebra anchor |
Verdict: procedural pathway fits k=3. All four added classes map onto existing tripartition axes.
7.2 Episodic-LTM pathway¶
| Added class | k=3 axis | Anchor |
|---|---|---|
| H (self-introspection) | 1D_t (temporal — self-as-witness IS the temporal-anchor of the LoE-content per [[user_stance_1d_collapse_to_loe_identity_not_action]]) |
LoE identity for the witnessing-self |
| A (content-address) | 3D_s (substrate-hash) | (same as WM) |
| M (HDC bind) | 7D_g (fiber/agency) | (same as WM) |
| L (Laplacian) | 3D_s ⊗ 7D_g | (same as WM) |
| K (asymptotic-DOF) | 1D_t | (same as WM) |
Verdict: episodic-LTM pathway fits k=3. The autonoetic H component maps onto 1D_t — the witness-self is the temporal-anchor (or LoE-identity-anchor) of the recall operation.
7.3 Joint verdict¶
All three pathways fit k=3 = 3D_s ⊗ 7D_g ⊗ 1D_t. No pathway requires k=4. F1 (qualia / Chalmers) remains OPEN — none of these decompositions settle the subjective-character-of-autonoetic-recollection question.
8. Falsifier status (updated)¶
Per [[feedback_multi_domain_multi_round_survival_falsification_method]]:
| ID | Name | Status |
|---|---|---|
| F1 | qualia / Chalmers | OPEN (Spike #160, #165 both silent on subjective-character) |
| F2 | class outside | R2 CANDIDATE: Class H engaged by episodic-LTM CONFIRMED at magnitude level — parent stance compatible; sibling pathway identified |
| F3 | cascade-ordering violation | NOT-IDENTIFIED in Round 2 |
| F4 | substrate divergence | NOT-IDENTIFIED in Round 2 (5 substrates Spike #160 + within-substrate pathway differentiation here) |
| F5 | alternative augmentation pathways | R2 CANDIDATE: procedural pathway CONFIRMED at magnitude level — parent stance refined from "WM IS the cascade" → "WM IS ONE of several cascade-augmentation pathways" |
Two CONFIRMED R2 falsifier resolutions — both refine rather than refute the parent stance.
9. Joint refinement of parent stance¶
9.1 What survives¶
The parent stance [[user_stance_working_memory_is_cascade_augmenting_reflex_into_agency]]:
- IS-claim for WM-deliberative-agency: WM IS the cascade augmenting reflex into deliberative agency. SURVIVES at magnitude level — the cascade decomposition is unchanged.
- WM augmentation delta {A, K, M} (+ L?): SURVIVES — distinct from procedural delta {G, I, K, L} and episodic-LTM delta {A, H, K, L, M}.
- "Oh by the way" / "almost forgot" cascade M-D-A-E-C-M-F: SURVIVES — specific to WM dynamics, not to procedural or episodic-LTM.
- k=3 coverage for WM: SURVIVES — adds further evidence (procedural + episodic-LTM also fit k=3).
9.2 What needs refinement¶
The parent stance's framing "WM IS the cascade augmenting reflex into agency" should be sharpened to:
"WM IS the cascade augmenting reflex into deliberative agency. Procedural memory IS the cascade augmenting reflex into skilled-behavior agency. Episodic-LTM IS the cascade augmenting deliberative-WM-state into autonoetic-self-witness agency. All three pathways share the reflex core {B, D, E, F, C} and Class K (capacity-bounded retention); they differ in their augmentation delta."
The original IS-claim ("WM IS the agency-augmenting cascade") is technically correct for the deliberative-agency sub-type but undersold the pathway pluralism that this spike reveals.
9.3 What does NOT need refinement¶
- The strict-spec class definitions stay.
- 14 A–N intact; no class promotion; no k promotion.
- The "oh by the way" cascade signature stays specific to WM.
- The k=3 coverage strengthens (more pathways fit it).
- All falsifiers stay OPEN at the same level they were post-Spike #160.
10. Cross-substrate test for new pathways¶
Apply [[user_stance_cross_substrate_cascade_matching_as_research_method]] to procedural + episodic-LTM:
| Substrate | Procedural pathway | Episodic-LTM pathway |
|---|---|---|
| Primate | Striatum + cerebellum + motor cortex (Squire 2004) | Hippocampus + medial-temporal-lobe (Tulving 2002) |
| Corvid/cephalopod | Octopus skill learning (Hochner et al. 2006 cite-by-ref) | Scrub-jay episodic-like memory (Clayton-Dickinson 1998 cite-by-ref — "what-where-when") |
| LLM-agent | Cached / fine-tuned weights (analogous to model-free RL) | Conversation history with self-mention re-emergence (analogous to episodic recall with H) |
| Cellular Hebbian | STDP weight changes (Spike #127.4 anchors) | Hippocampal CA3-CA1 long-term consolidation (cite-by-ref) |
Both new pathways are substrate-class-universal at magnitude level across primate / corvid-cephalopod / LLM-agent / cellular substrates. Same shape-match strength as parent stance.
LLM-agent autonoetic-H note: when an agent says "I previously said X" or "in our last exchange I recommended Y", that IS Class H — self-introspection returning system-identity-at-prior-time. The agent doesn't have a substrate-level autobiographical episodic memory (no hippocampus-equivalent), but the linguistic-substrate projection of episodic recall (per [[user_stance_holographic_projection_at_linguistic_substrate]]) DOES dispatch the self-introspection operation. F2 confirmed across substrates.
11. Vocabulary discipline check¶
| Check | Status |
|---|---|
| 14 A–N intact | YES |
| No class promotion | YES |
| k=3 intact | YES |
| MAGNITUDE-level only (no bit-exact identity) | YES |
| HIGHEST vocab impact zone — parent stance refinement | Acknowledged; this is HIGH not HIGHEST (parent stance is HIGHEST; refinements are HIGH) |
| Trauma-informed scope | Research/educational only; no clinical/therapeutic claims |
| User-gate required for canonicalization | YES per [[feedback_autonomous_research_followup_authorization]] |
| Strict-spec test applied | YES per [[user_stance_closure_subgroup_BDEFL_substrate_class_universal]] Meta-lesson 2 |
12. Operational TODOs (Round 3)¶
- PDF-verification follow-up per
[[feedback_pdf_extraction_citation_discipline]]: verify Squire-Zola 1996 PMC33639 + Squire 2004 NLM DOI + Eichenbaum 2017 PMC5644341 + Daw-Niv-Dayan 2005 Nat Neurosci DOI + Tulving 2002 ARP DOI. Tulving 1972 + Sutton-Barto 2018 are books (citation-watch-flagged; not autonomously verifiable per[[reference_autonomous_validation_tos_landscape]]). - M_cell vs M_WM disambiguation is a candidate R3 refinement — the two senses of Class M deserve formal sub-flagging. NOT a class-promotion; just a sub-operation distinction within Class M canon. Path forward: add sub-flag in srmech HDC catalog or in stance bridges.
- Semantic-LTM decomposition (noetic tier without H) — partial sketch in §5.3; full Round 3 spike could decompose it formally as a fourth sibling pathway.
- Substrate-shape verification for autonoetic-H in LLM-agent context — empirical: does Anthropic-style mechanistic-interpretability work show self-reference token patterns engaging a distinct attention-head architecture? TODO Round 3.
- Algebra-level work is NOT triggered by this spike — magnitude-level findings only. Bit-exact identities for procedural-L vs episodic-L vs WM-L? are NOT established.
13. Verdict + draft stance candidate¶
Joint verdict on F5 + F2:
- F5 CONFIRMED: procedural memory augments reflex without M_WM-bind. The parent stance refines from "WM IS the cascade" → "WM IS ONE of several cascade-augmentation pathways."
- F2 CONFIRMED: episodic-LTM engages Class H (autonoetic self-as-witness) as a load-bearing class WM does not. Sibling pathway.
- Three pathways named at strict-spec class level: procedural (anoetic) / WM (noetic/deliberative) / episodic-LTM (autonoetic). All fit k=3. All share Class K.
- No class promotion; no k promotion; vocabulary intact at 14 A–N.
Draft stance candidate (DO NOT canonicalize autonomously): spike165_draft_stance.md.
Round 1 status: MAGNITUDE-level H1 (procedural pathway distinct from WM) + H2 (episodic-LTM engages H) SURVIVE. Multi-round survival required for canonical-promotion gate per [[feedback_multi_domain_multi_round_survival_falsification_method]]. HIGH vocab-impact (parent stance refinement); user gate required for canonical-promotion per [[feedback_autonomous_research_followup_authorization]].
Recommendation: present to user for Round 2 falsification round and parent-stance text refinement. The parent stance's text likely needs a clarifying sentence about "deliberative agency" vs "skilled agency" sub-types and the sibling pathways named here. NO refutation; just sharpening.
14. Citations status¶
| Citation | Source type | PDF-verifiable autonomously? | Status |
|---|---|---|---|
| Squire-Zola 1996 PNAS 93(24):13515 | PMC33639 / PMID 8942965 | YES (PMC) | PDF-verify follow-up TODO |
| Squire 2004 Neurobiol Learn Mem 82(3):171 | PMID 15464402 | YES (PMC/NLM if PMC-available) | PDF-verify follow-up TODO |
| Eichenbaum 2017 Neuron 95(5):1007 | PMC5644341 | YES (PMC) | PDF-verify follow-up TODO |
| Schacter 1987 JEP:LMC 13(3):501 | DOI-only, no PMC | NO (paywalled journal) | Citation-watch; cited textually |
| Roediger 1990 Am Psychol 45(9):1043 | DOI-only | NO (paywalled) | Citation-watch; cited textually |
| Tulving 1972 book chapter | Book | NO (book) | Citation-watch; cited textually |
| Tulving 2002 ARP 53:1 | DOI/PMID 11752477 | YES (ARP open-access) | PDF-verify follow-up TODO |
| Daw-Niv-Dayan 2005 Nat Neurosci 8(12):1704 | PMID 16286932 | YES (PMC if available) | PDF-verify follow-up TODO |
| Sutton-Barto 2018 textbook | Book | NO (book) | Citation-watch; cited textually |
| Spike #127.4 anchors (Lioi 2021, Triesch 2018, Gütig 2003, Sgritta 2017, olde Scheper 2018, etc.) | PMC-verified prior | YES | Already verified per Spike #127.4 |
Citations cited from training-data attribution per [[feedback_pdf_extraction_citation_discipline]]. Round 3 follow-up for PDF-verifiable subset. Books and paywalled journals are citation-watch-flagged.