Create 2026-06-29-biological-reference_v15.md

2026-06-29 21:32:16 +02:00
parent 4c9bb9e48e
commit 8a577b4fad
1 changed files with 407 additions and 0 deletions
@@ -0,0 +1,407 @@
 # Tripartite Synapse — Biological Reference (companion to v15 pseudocode)
 > Companion to `tripartite_synapse_v15_pseudocode.md` · principle: `logic_principles_v3`.
 > v15 distributes NIGHT into each component as an iterated cycle. Biologically this reflects
 > that consolidation is not a single event but proceeds across the repeated slow-wave cycles of
 > NREM sleep: each cycle the producers (soma protein synthesis, astrocyte glycolysis/ECM
 > synthesis) make a bounded batch of material and energy, it is transported one hop down the
 > dendrite/axon/astrocytic processes, and tagged synapses incorporate what arrives. Distal sites
 > consolidate later because their material arrives over more cycles. The night ends when there is
 > nothing left worth consolidating (tags spent) or the night's metabolic supply is exhausted —
 > a heavier learning day therefore demands a longer or fuller night, and unfinished consolidation
 > carries to the following night. Energy spent on construction is irreversible (the system's one
 > one-way flow); material released by pruned structure is recycled.
 ---
 ## The three synaptic components and their support structures
 A SYNAPSE is composed of three first-class components:
 - **PRE** — presynaptic bouton (the axon's terminal at this synapse)
 - **POST** — postsynaptic spine (the dendrite's terminal at this synapse)
 - **ASTRO** — astrosynapse, the perisynaptic astrocytic process (the astrocyte's terminal)
 Each has an upstream support structure that supplies it:
 - **AXON** supplies PRE (transmission + transport from soma)
 - **DEND** supplies POST (integration + transport from soma)
 - the **astrocyte cell body** supplies ASTRO (energy + ECM material)
 - **SOMA** is the integrating center and the root of neuronal material
 The compartment analogy: AXON:PRE = DEND:POST = astrocyte-body:ASTRO = supply line : terminal.
 ---
 ## Resource variables
 ### DAY budget (one per component)
 Aggregates fast energy AND fast consumables — everything needed to run moment-to-moment.
 - **pre_budget** — ATP for VGCC gating, vesicle fusion (SNARE), VATPase vesicle refill,
  plus fast consumables: vesicle membrane lipids, synaptotagmin recycling.
 - **post_budget** — ATP for the NaK pump (membrane reset after current), NMDA current
  handling, plus fast actin monomers for transient spine changes and receptor-recycling lipids.
 - **dend_budget** — ATP for bAP propagation (NaK reset along branch), local translation
  (ribosome running cost), SERCA Ca²⁺ resequestration, plus fast mRNA consumed by translation.
 - **soma_budget** — ATP for AP generation (Na⁺/K⁺ currents + NaK reset), CREB
  phosphorylation, nuclear Ca²⁺ handling, plus shipping running costs.
 - **axon_budget** — ATP for AP propagation at nodes of Ranvier, kinesin/dynein motor
  running cost, fast myelin maintenance.
 - **astro_central_budget** — ATP from glycolysis at the astrocyte cell body; funds EAAT
  clearance, serine→D-serine synthesis, lactate export, fast process motility.
 ### astro_lactate[i]
 Lactate exported from the astrocyte cell body to synapse i. Biologically: glucose →
 (glycolysis) → lactate, released into extracellular space, absorbed by neuronal MCT2
 transporters, converted to pyruvate → TCA → ATP in the neuron's mitochondria. The astrocyte
 is the primary fast-energy supplier to pre, post, and dend.
 ### NIGHT energy (one per component) — NOT recoverable
 ATP for structural assembly. Distinct from DAY budget because it is spent on building, and
 the work of assembly is thermodynamically gone once done (cannot be recovered by disassembly).
 - pre_energy: RIM/Munc13 incorporation, VGCC clustering.
 - post_energy: CaMKII anchoring, actin polymerization, PSD scaffold remodeling.
 - dend_energy: mitochondria incorporation, cytoskeletal reinforcement.
 - soma_energy: ribosome biogenesis, ion-channel incorporation.
 - axon_energy: myelination, microtubule stabilization.
 - astro_energy: process retraction, ECM secretion, racemase upregulation.
 ### NIGHT material (one per component) — RECOVERABLE
 Slow structural proteins. Recoverable because disassembly (LTD) returns the proteins to a
 reusable pool (ubiquitin-proteasome → amino acids; internalized receptors → endosomal reserve).
 - **soma_material** (root) — all neuronal structural proteins from CREB-driven synthesis:
  AMPA subunits, PSD scaffold, AZ scaffold, mRNA transcripts (Arc, BDNF), organelles.
 - **dend_material** — from soma: Arc/plasticity mRNA, mitochondria, cytoskeletal proteins,
  AMPA subunits in transit to spines.
 - **post_material** — from dend: AMPA receptor subunits (GluA1/2), PSD scaffold (PSD-95,
  SHANK, Homer), structural actin, CaMKII.
 - **axon_material** — from soma: kinesin/dynein motors, microtubule components, myelin proteins.
 - **pre_material** — from axon: RIM, Munc13, VGCC subunits, structural vesicle proteins.
 - **astro_material** (root: astrocyte cell body) — EAAT proteins, serine racemase, ECM
  proteins (Glypicans, Thrombospondins), process cytoskeleton.
 **Why energy and material are separate in NIGHT but combined in DAY:** during DAY both are
 fast consumables replenished on the same timescale, so one `budget` variable suffices. During
 NIGHT they diverge — material is recoverable after LTD, energy is not — so they must be two
 variables. This asymmetry (material returns to the pool, energy is gone) is what makes one
 synapse's depression genuinely fund another's potentiation.
 ---
 ## Structural variables (strength ceilings — written in NIGHT)
 Each aggregates several correlated structural properties into one capacity.
 - **pre_structure** — active zone capacity:
  slot_ceiling (number of vesicle docking slots) + VGCC_coupling (Ca²⁺-channel proximity to
  slots, sets release efficiency) + refill_ceiling (max RRP replenishment rate).
 - **post_structure** — spine sensitivity capacity:
  slot_ceiling (number of PSD anchoring slots for AMPA) + spine_volume (local reserve and
  actin machinery) + reserve_ceiling (endosomal AMPA pool size).
 - **dend_structure** — branch capacity:
  bAP_fidelity(position) (mitochondrial density sets propagation strength, attenuates with
  distance) + translation_ceiling (local mRNA capacity) + transport_speed (cytoskeletal integrity).
 - **soma_structure** — somatic output capacity:
  baseline_threshold (inverse: ion-channel density at axon initial segment) + AP_reliability
  (Na⁺ channel density) + synthesis_ceiling (ribosome density + CREB machinery).
 - **axon_structure** — axonal capacity:
  propagation reliability (myelination density) + transport_ceiling (motor density + microtubule
  integrity) + mitochondrial density.
 - **astro_structure** — astrosynaptic environmental capacity:
  perisynaptic_distance⁻¹ (wall proximity — closer = more glutamate contained) + EAAT_density
  (clearance ceiling) + Dserine_tonic (baseline co-agonist) + ECM_integrity.
  **Self-reinforcing both directions:** tighter wrap + more tonic D-serine make future
  potentiation easier; looser wrap + zero tonic D-serine make future depression easier.
 ---
 ## Budget ceilings (endurance ceilings — written in NIGHT)
 - **{component}_budget_ceiling** — the maximum fuel the component can hold / the maximum
  duration of sustained behavior. Biologically: mitochondrial density and local fuel-storage
  capacity. Built by activity-driven mitochondrial biogenesis; lost by mitophagy when idle.
  Parallel to structure: structure is strength capacity, budget_ceiling is endurance capacity.
 ---
 ## Trace variables
 ### fast_trace (one per component) — DAY only, decays automatically
 The local record of recent activity that biases the next behavior.
 - **pre_fast_trace** — residual presynaptic Ca²⁺ after spikes (τ≈100ms). Biases NT release
  (facilitation) and provides tagging eligibility.
 - **post_fast_trace** — spine Ca²⁺ amplitude × rise-speed (τ≈tens ms). Encodes the LTP-vs-LTD
  instruction (fast rise → CaMKII → potentiation; slow rise → phosphatase → depression).
 - **dend_fast_trace** — branch Ca²⁺ from bAP + spine spillover (τ≈300ms). Integrates branch co-activity.
 - **soma_fast_trace** — nuclear Ca²⁺ from each AP (τ≈seconds). Drives toward CREB activation.
 - **axon_fast_trace** — propagation load (τ≈seconds). High load → Na⁺ inactivation at branch
  points → propagation failure (this is axonal short-term depression).
 - **astro_fast_trace** — perisynaptic Ca²⁺ from mGluR5 activation by glutamate spillover
  (τ≈seconds). Drives D-serine release.
 ### soma timing traces (emergent refractory + adaptation + alignment)
 - **soma_Na_inactivation** (τ≈ms) — sodium-channel inactivation after an AP. Its recovery IS
  the refractory period (emergent, not a hardcoded timer). High → absolute refractory; decaying
  → relative refractory; recovered → normal.
 - **soma_adaptation** (τ≈100s of ms) — slow K⁺ channel (SK/M-type) activation accumulating
  over a spike train, raising threshold. This is spike-frequency adaptation.
 - **soma_refractory_alignment** — deposited when a suprathreshold input arrives during
  refractoriness (a missed coincidence). Speeds future recovery so the soma aligns to its input
  rhythm. Bottom-up: no rhythm is represented; alignment emerges from accumulated local
  mismatches and decays when mismatches stop (self-limiting).
 ### possible_tag (one per component) — intermediate, τ≈s–min
 Graded accumulation of tagging eligibility. For POST, this is the CANDIDATE tag lifetime.
 ### endurance_need (one per component) — intermediate, τ≈s–min
 Deposited when budget depletion interrupts a behavior that was on a LOCALLY successful
 trajectory. Records that fuel — not structure, not significance — was the binding constraint
 on a forming success. Requires NO dopamine (homeostatic, not associative).
 **Local success proxy per component** (each uses only its own state + arrived signals):
 - PRE: own fast_trace high (was releasing strongly), optionally amplified by retrograde
  messenger (endocannabinoid / NO / BDNF) that has arrived.
 - POST: own Ca²⁺ climbing toward tagging threshold (naturally local).
 - DEND: own branch strongly active (high branch voltage/Ca²⁺) when propagation fell short.
 - SOMA: own nuclear Ca²⁺ climbing toward CREB.
 - AXON: own propagation load high (was carrying a strong train).
 - ASTRO: own local glutamate/Ca²⁺ high (was under heavy clearance/D-serine demand).
 ### tag (one per component) — DAY→NIGHT bridge, τ≈hours
 The validated record of significance that survives to NIGHT and gates strength commits.
 Formed by coincidence of local eligibility + non-local validation (dopamine).
 **POST is special — two-phase, three coincidences:**
 - CANDIDATE: local Ca²⁺ above threshold + astrosynapse D-serine present (coincidence 1).
 - amplified when bAP confirms soma fired (coincidence 2).
 - STABLE: CANDIDATE + dopamine within stabilization window (coincidence 3).
 Biologically: early CaMKII creates a labile tag (early-LTP); PKA driven by dopamine via D1R
 stabilizes it (late-LTP). Without dopamine, the candidate degrades — early-LTP reverses.
 ---
 ## Behaviors — biological meaning
 ### PRE | AP — neurotransmitter release
 `NT_flux = RRP × sat(pre_fast_trace, K_release)` models continuous NT release proportional to
 the readily-releasable pool and a saturating Ca²⁺ drive (synaptotagmin's cooperative Ca²⁺
 sensitivity, simplified to a saturating curve). RRP depletes as released (short-term depression
 as a consequence) and refills via VATPase (energy-throttled, so low budget deepens depression).
 The mGluR2/3 brake is presynaptic autoinhibition by spillover (Gi → reduced VGCC opening).
 ### POST | NOT_bAP — three calcium sources, two plasticity cases
 - **Source 1 (AMPA):** glutamate opens AMPA → depolarizing current + small Ca²⁺; the
  depolarization begins ejecting the NMDA Mg²⁺ block.
 - **Source 2 (NMDA):** if depolarized enough (Mg²⁺ ejected) AND D-serine present (astrocyte
  co-agonist) AND glutamate bound → large Ca²⁺ influx. This is the coincidence detector.
 - **Source 3 (bAP, separate context):** back-propagating AP adds depolarization + Ca²⁺,
  amplifying an existing signal supralinearly.
 - **Case 1 (STP):** high Ca²⁺ drives AMPA receptors from the local reserve to the surface,
  bounded by the anchoring-slot ceiling. Fast, reversible, NO dopamine. When Ca²⁺ falls,
  receptors drift back — short-term depression as a passive consequence, never signaled.
 - **Case 2 (LTP tag):** high Ca²⁺ + (later) dopamine sets the tag that NIGHT uses to raise the
  slot ceiling. NIGHT builds slots; DAY fills them.
 ### DEND | bAP — bidirectional signaling
 Propagates the bAP from soma toward spines (fidelity attenuates with distance — distal spines
 get weaker confirmation, are harder to potentiate) and integrates spine signals toward the soma.
 ### SOMA | AP — integration, firing, emergent timing
 Fires when integrated branch input exceeds a threshold that is the baseline (from structure)
 raised by adaptation and modulated by neuromodulators, gated by the emergent refractory state.
 Each AP deposits three traces (inactivation → refractory, adaptation → threshold rise, nuclear
 Ca²⁺ → plasticity). The soma is the coincidence detector at the cellular scale (nuclear Ca²⁺ +
 dopamine → CREB), and the production bottleneck: its tag gates how much material all downstream
 components get in NIGHT.
 ### AXON | AP — reliable propagation with frequency-dependent failure
 Propagation reliability is set by myelination and degraded by high-frequency load (Na⁺
 inactivation at branch points = axonal STD). The axon also transports material to boutons and
 sets the timescale of presynaptic structural commits.
 ### ASTRO | CONTINUOUS — gatekeeper and energy hub
 Clears glutamate (EAAT), supplies D-serine (the NMDA co-agonist that gates postsynaptic LTP),
 and distributes lactate to the territory by demand-weighting (active synapses generating more
 clearance load pull more fuel; slow synapses get less). The same spillover that excites the
 astrocyte (mGluR5 → Ca²⁺ → D-serine) also brakes the presynapse (mGluR2/3 → Gi) — one signal,
 opposite effects via different receptors. The astrocyte is the energy root and the gain control
 of the whole synapse.
 ---
 ## NIGHT operations — biological meaning
 - **Step 1 (replenish/distribute):** overnight protein synthesis peaks (CREB-driven, gated by
  soma_tag — corresponds to slow-wave-sleep replay). Soma material flows to branches/axon then
  spines/boutons; astrocyte material flows to astrosynapses, tag-weighted.
 - **Step 2 (strength commits):** tagged components build structure — more slots, tighter
  coupling, tighter astrosynaptic wrap. Coherence bonus when pre+post+astro all tagged (the
  whole synapse agrees). astro_structure self-reinforces.
 - **Step 2b (endurance commits):** components with high endurance_need build budget_ceiling —
  mitochondrial biogenesis. Competes with step 2 for the same material/energy.
 - **Step 3 (passive decay):** both ceilings decay; maintenance from the remaining pool resists
  decay only where sufficient. Depotentiation and endurance-loss are both by neglect — no
  signal weakens anything; unmaintained capacity simply drifts down. Recovered material (not
  energy) returns to pools.
 - **Step 4 (homeostatic scaling):** if the soma fired too much overall, all synapses scale down
  proportionally (sleep-associated global downscaling), preserving relative differences.
 - **Step 5 (clear traces):** fast traces, possible tags, endurance needs, and soma timing traces
  reset; tags below expiry clear, above-expiry tags carry forward (multi-night consolidation);
  structure and budget_ceiling persist.
 ### Shockwave lockdown
 Emergency global astrocytic Ca²⁺ wave → GABA + ATP release → mass AMPA internalization and
 hyperpolarization. Bypasses budget gates. A circuit breaker against runaway excitation.
 ---
 ## Pool-filling: private reserve vs contested supply
 The pseudocode uses two filling primitives, distinguished by where the resource comes from.
 **`fill` (private reserve).** The pool is replenished from a source the component owns
 outright, uncontested by siblings, bounded by the component's own ceiling and a rate cap.
 - RRP refill — vesicles mobilized from the bouton's own reserve pool toward the docking-slot
  ceiling, rate-limited by VATPase. The reserve is private to the bouton.
 - SOMA self-replenish — the soma fuels itself from its own mitochondria toward its budget
  ceiling. No other component draws on it.
 **`refill` (contested supply).** The pool is replenished from a supply that multiple
 components compete for, rationed by demand (gap to ceiling).
 - pre/post/dend/axon budgets — drawn from astrocytic lactate (shared across all synapses the
  astrocyte wraps) plus shipment from soma/axon/dendrite (shared across downstream targets).
 **Neither primitive (their own forms).** Some inflows are not fills toward a ceiling:
 - AMPA surface insertion — Ca²⁺-driven rate from the spine's private endosomal reserve, with
  an explicit passive drift-back (short-term depression) when Ca²⁺ is low. Not a steady fill.
 - D-serine release — demand-driven (saturating in astro Ca²⁺) and budget-limited, like NT
  release; a release process, not a pool top-up.
 - Root productions — `glycolysis(glucose)` at the astrocyte and `CREB_synth(soma_tag)` at the
  soma are the system's energy and material roots: raw inflows capped only by the external
  vascular supply, not fills toward an internal ceiling.
 The distinction matters biologically: a private reserve guarantees a component some autonomy
 (the bouton can refill its RRP from its own vesicles even when lactate is scarce), while a
 contested supply couples a component's fate to its neighbours' demands (operational budget
 fails first where many active synapses compete for the same lactate).
 ---
 ## PRE ↔ POST interaction: local computation, message-only coupling
 The presynapse and postsynapse never read each other's internal state. They interact only
 by writing to and reading from shared cleft channels. Each side computes entirely locally on
 what it has: its own variables plus whatever signals have arrived in the cleft. This is the
 message-passing realization of the locality principle.
 **Forward channel — glutamate (PRE → POST and ASTRO).** The presynapse writes glutamate via
 NT_flux. The postsynapse reads it (AMPA, NMDA) and the astrosynapse reads it (clearance,
 mGluR5). The astrosynapse clears it. PRE never knows whether POST responded — it only emits.
 **Gate channel — astro_Dserine (ASTRO → POST).** The astrosynapse writes D-serine; the
 postsynapse reads it as the obligatory NMDA co-agonist. POST cannot open NMDA without this
 arrived signal, but it does not read the astrocyte's state — only the delivered D-serine.
 **Backward channel + — retro_NO (POST → PRE).** When the postsynapse's NMDA opens (Mg²⁺
 ejected, D-serine present, glutamate bound), nNOS — physically tethered to the NMDA receptor
 through PSD-95 — synthesises nitric oxide (and, on a slower timescale, BDNF is released).
 These diffuse retrogradely to the presynapse. Biologically this is the classic retrograde
 messenger of LTP: it tells the bouton that its release landed on a postsynapse that genuinely
 responded. In the model, POST emits `retro_NO` proportional to its own NMDA-driven calcium —
 computed purely from POST's local state — and PRE reads it as `retro_NO_local`.
  `retro_NO_local` is exactly the grounding of the presynaptic endurance signal. The
  presynapse's local success proxy is "I was releasing strongly" (`pre_fast_trace` high). On
  its own that only says the bouton was working hard, not that the work mattered. `retro_NO`
  adds the missing confirmation — that the postsynapse responded — without PRE ever reading
  POST's calcium. So PRE deposits endurance need as `pre_fast_trace × (1 + retro_NO_local)`:
  strong release that was confirmed effective makes the strongest claim that fuel, not
  futility, was what interrupted a forming success. retro_NO is short-lived (NO degrades and
  diffuses within seconds), so the channel decays fast — confirmation must be recent to count.
 **Backward channel − — retro_eCB (POST → PRE).** When the postsynapse is strongly
 depolarised, it synthesises endocannabinoids (2-AG, anandamide) that diffuse retrogradely and
 bind presynaptic CB1 receptors, suppressing release. This is depolarisation-induced
 suppression of excitation (DSE) — a homeostatic negative feedback: an over-driven postsynapse
 tells the presynapse to release less. In the model, POST emits `retro_eCB` from its own
 membrane potential, and PRE reads it as `retro_eCB_local`, which reduces the release drive
 `sat(...) × (1 - retro_eCB_local)`. Again POST computes from its own state; PRE adjusts from
 the arrived signal; neither reads the other's interior.
 The two backward channels are opposite-signed messages the postsynapse sends about its own
 condition: retro_NO says "your input was effective — worth sustaining," retro_eCB says "I am
 saturated — ease off." Together with the forward glutamate and the D-serine gate, they make
 the synapse a fully message-coupled system of locally-computing components.
 **Why RRP refill is in NOT_AP only.** During an AP the bouton releases — RRP depletes. Refill
 (VATPase reloading vesicles from the reserve pool) is a recovery process that proceeds between
 spikes. Placing `fill(RRP, ...)` only in the NOT_AP context makes the AP context pure
 depletion and the NOT_AP context pure recovery. A consequence falls out for free: during
 sustained high-frequency firing there are many AP steps and few NOT_AP steps, so RRP depletes
 faster than it recovers — short-term depression deepens with frequency, with no explicit
 depression rule. The release itself is throttled further when budget is low (VATPase refill
 is energy-limited), coupling metabolic state to the depth of depression.
 ---
 ## Presynaptic short-term potentiation — VGCC coupling occupancy
 `VGCC_active` is the presynaptic parallel to the postsynaptic `AMPA_surface`. Both are MEDIUM-tier
 occupancy variables: a current operating value filled toward a NIGHT-built ceiling, no dopamine,
 reversible, drifting back when undriven.
 Biologically, `VGCC_active` represents the effective coupling between voltage-gated calcium
 channels and the vesicle docking slots — how reliably each calcium influx is converted into
 release. Repeated eligible activity (accumulated `pre_possible_tag`) transiently tightens this
 coupling — through calcium-channel facilitation, active-zone protein phosphorylation, and
 channel-to-sensor proximity changes — raising release efficiency without changing the number of
 channels (which is the structural ceiling `pre_structure.VGCC_coupling`, written only at NIGHT).
 When eligibility falls, the coupling relaxes back to baseline over seconds-to-minutes: presynaptic
 short-term depression as the passive consequence of undriven coupling, never a signalled act.
 This gives the presynapse a genuine intermediate-timescale memory it previously lacked — a
 "this bouton has been reliably active lately" state that outlasts individual spikes and bursts,
 filling the gap between the fast trace (residual calcium, ~100 ms) and the tag (hours). It also
 completes the capacity/occupancy symmetry across the synapse: both PRE and POST now fill a
 MEDIUM occupancy variable toward a PERSISTENT structural ceiling, rather than PRE reading its
 ceiling directly as if capacity and occupancy were the same thing.
 ---
 ## NIGHT as iterated NREM cycles — the biology
 The distributed, cyclic NIGHT models sleep-dependent consolidation more faithfully than a single
 commit step.
 **Why cycles, not one event.** NREM sleep proceeds in repeated cycles (the ultradian ~90-minute
 rhythm, and within it the <1 Hz slow oscillation with its up- and down-states). Protein synthesis,
 hippocampal–cortical replay, and synaptic renormalization all advance incrementally across these
 cycles rather than in a single consolidation moment. Modeling NIGHT as a loop of cycles captures
 this: each cycle is a small, local round of produce → transport → incorporate.
 **Production each cycle (the roots).** The soma's CREB-driven transcription/translation produces a
 batch of structural material per cycle, gated by the soma's own tag (replay-driven activity).
 The astrocyte cell body produces a batch of energy (glycolysis) and ECM material per cycle, capped
 by glucose. These are the two roots; everything downstream lives on what they ship.
 **Transport over cycles (the descent).** Material and energy move one hop down the supply chains
 per cycle — soma → dendrite/axon → spine/bouton; astrocyte body → astrosynapses — by the same
 motor transport that carries cargo by day, now at the consolidation timescale. A distal bouton on
 a long axon therefore receives its material only after several cycles, so its consolidation lags
 a proximal one. This is the NIGHT-scale image of the transit delay.
 **Incorporation and tag consumption (the commit).** A tagged synapse incorporates arrived material
 into structure (more receptor slots, tighter active zone, tighter astrocytic wrap) or into budget
 capacity (mitochondrial biogenesis), spending energy on the assembly. The tag is consumed in
 proportion to what was built — the molecular tag (CaMKII/PKA-maintained eligibility) is discharged
 as capture completes. A strong tag is satisfied early; a marginal one waits for later cycles.
 **Two ways the night ends.** Either the standing tags are all spent (consolidation finished — the
 rested case) or the night's metabolic budget is exhausted (ran out of night — the overloaded
 case). Unspent tags are not discarded: they persist (decaying slowly) into the next day and
 compete again the next night, so importance is re-tested across nights and a marginal memory may
 consolidate over several nights or, if it decays first, never.
 **Energy is the irreversible cost.** Material released when an unmaintained structure is pruned
 returns to the pool and is reused; the energy burned to build or to prune is gone. Across a
 lifetime this energy throughput bounds how much the system can ever consolidate — the metabolic
 arrow of time underlying the whole model.