Create 2026-06-19-abstract-pattern-expression.md

2026-07-08 16:35:04 +02:00
parent dd8a55c4d5
commit 044c6fb875
1 changed files with 514 additions and 0 deletions
@@ -0,0 +1,514 @@
 # The Abstract Pattern
 A component operates within two ceilings set by the previous NIGHT: a **structure** that
 bounds how strongly each behavior can act, and a **budget capacity** that bounds how much fuel
 it can hold. Both ceilings bound an active, competitive DAY process and are raised at NIGHT by
 two kinds of evidence — a **tag** (local eligibility coinciding with non-local validation,
 driving strength) and an **endurance need** (a *fuel* shortfall interrupting a locally
 successful trajectory, driving endurance). Strength and endurance draw on the same finite
 material and energy, so they compete; what is not committed decays for lack of maintenance,
 and the freed material partially funds what was.
 Everything a component does — in DAY and in NIGHT alike — is an instance of one grammar of
 seven functional groups. This is the common template. Components differ only in how they fill
 the groups, never in their structure.
 ---
 ## The Grammar
 ```
 RECEIVE   take in resources + signals that arrived from outside      (boundary: in)
 TRACE     maintain the trace hierarchy — deposit fast trace; accumulate
          possible_tag + endurance_need; stabilize tag on coincidence
 ADJUST    compute local operating parameters from structure + traces + modulators
 BEHAVE    the component's defining action, within both ceilings
 EMIT      send out — signals (messages) + resources (shipments)      (boundary: out)
 RECOVER   refill own private pools consumed by behaving
 DECAY     let traces recede, closing their windows
 ```
 **TRACE absorbs evaluation.** Judging a behavior is always maintaining a trace — whether or
 not a trace is written — so the two evidence streams (strength, endurance) live inside TRACE
 alongside the fast-trace deposit. TRACE spans all timescales: the soma's inactivation,
 adaptation, and nuclear-calcium deposits are all TRACE, three records of one firing.
 **BEHAVE and EMIT stay separate.** EMIT is the output half of the locality interface. RECEIVE
 and EMIT are the *only* boundary crossings; every other group is strictly local. Keeping EMIT
 named makes the boundary auditable — sometimes EMIT coincides with the behavior (presynaptic
 release, somatic firing), often it is a separable consequence (postsynaptic retrograde
 messages) or a different operation entirely (resource shipments).
 The grammar runs at two timescales. In **DAY** it operates on occupancy *within* the ceilings
 (fast traces, budgets, behaviors). In **NIGHT** it operates on the *ceilings themselves*
 (material, energy, structure, budget capacity). Same seven groups, two scopes — the
 capacity-versus-occupancy principle written into the shape of the process.
 **Every flow has a timescale.** Decay relaxes a quantity toward 0 over τ; creation and arrival
 relax it toward a target over τ — the same first-order operator, differing only in the target.
 A within-step write is the special case τ ≪ Δt. Rate-limited inflows (private refill, contested
 replenishment, flux × Δt) carry their τ implicitly; shipment carries an explicit transit delay,
 so distal targets receive later and more spread out than near ones.
 **Strict locality.** Every group uses only the component's own state and signals that have
 physically arrived. No component reads another's interior. All cross-compartment influence
 enters through RECEIVE and leaves through EMIT.
 ---
 ## DAY — The Grammar on Occupancy
 ### RECEIVE — what arrived becomes local
 Resources arrive by three disciplines, each bounded by a ceiling through the *gap* it leaves,
 never by a post-hoc clamp:
 ```
 // CONTESTED supply (astrocyte lactate, arrived shipments) — rationed by competing demands
 demand = BUDGET_CEILING - BUDGET                       // the gap is the claim
 factor = min(1, S / (Σ demand on S + ε))
 BUDGET += demand × factor;  S -= demand × factor       // never exceeds the ceiling
 // PRIVATE reserve (own vesicle pool, own mitochondria) — uncontested, rate-limited
 POOL += min(rate, CEILING - POOL) × Δt
 // TRANSIT (shipment arriving from upstream) — delivered over a transport time
 BUDGET += transit(shipment_channel, τ_transport)       // fraction of in-flight cargo lands
 ```
 Signals arrive as channel reads — forward transmitter, the co-agonist gate, retrograde
 messengers, neuromodulatory broadcast — latched into a local copy while the channel decays. A
 high budget ceiling is not free even in DAY: it makes a large standing claim on contested
 supply, satisfiable only by out-competing neighbours.
 ### TRACE — maintain the record hierarchy (and the two evidence streams)
 One group deposits and updates every trace, on every timescale.
 **Deposit** the fast trace from the behavior (and, for components with multi-timescale
 dynamics, the slower records too — soma inactivation and adaptation alongside nuclear calcium).
 Where a fast trace both *drives* and *records* a behavior (residual calcium at the presynapse),
 the deposit precedes ADJUST; elsewhere it follows BEHAVE.
 **Strength evidence (associative, needs dopamine).** A tag forms when local eligibility
 coincides with non-local validation; the number of coincidences set by the component's spatial
 scale.
 ```
 // most components — one non-local coincidence
 if FAST_TRACE > elig and dopamine > dop_thr:
    TAG += dopamine × possible_tag
 // POST — three coincidences (astrosynapse gate, soma bAP, organism dopamine)
 if FAST_TRACE > Ca_TAG and astro_Dserine > thr: possible_tag += FAST_TRACE   // CANDIDATE
 if possible_tag > thr and bAP arrives:          FAST_TRACE  += bAP_boost      // confirm
 if possible_tag > thr and dopamine > dop_thr:   TAG         += dopamine × possible_tag  // STABLE
 ```
 **Endurance evidence (homeostatic, no dopamine).** A need forms only when a **fuel** shortfall
 interrupts a *locally* successful trajectory. The success proxy is the component's own state,
 amplified by a retrograde signal *only where one actually arrives*.
 ```
 if BUDGET < cost and LOCAL_SUCCESS_PROXY > traj_thr:
    ENDURANCE_NEED += LOCAL_SUCCESS_PROXY × (1 + arrived_feedback)   // feedback only if it arrived
 ```
 Local proxies (own-state): PRE — own strong release (amplified by retrograde NO that POST
 emitted, the one component with arrived feedback); POST — own calcium climbing toward the tag;
 DEND — own branch strongly active when propagation fell short; SOMA — own nuclear calcium
 approaching CREB; AXON — own strong propagation load; ASTRO — own high clearance demand when
 synthesis ran short of budget. Same shape everywhere — fuel ran out at the verge of the
 component's *own* success — only the proxy differs.
 ### ADJUST — set the operating parameters
 From structure (the ceiling), the current traces, and arrived modulators, compute the
 parameters that govern this step's behavior:
 ```
 parameter = f(STRUCTURE, FAST_TRACE, modulators, arrived_signals)
 ```
 PRE — release drive from residual calcium and the received DSE brake; POST — AMPA drive from
 arrived glutamate; SOMA — firing threshold from baseline × adaptation × neuromodulators, plus
 the refractory gate; AXON — propagation reliability from structure minus load-driven failure;
 ASTRO — lactate-allocation weights across the territory. ADJUST is where the NIGHT-built ceiling
 silently shapes the moment: the same structure that bounds the maximum also tunes the transfer
 function.
 ### BEHAVE — the defining action, within both ceilings, with two failure modes
 ```
 if BUDGET < cost:
    suppress behavior
    // FUEL shortfall → endurance evidence (handled in TRACE, if the trajectory was succeeding)
 else if OCCUPANCY/STRUCTURE exhausted:        // pool empty, surface already at ceiling, refractory
    suppress behavior
    // → short-term depression, or a structural/timing limit — NOT endurance (more fuel wouldn't help)
 else:
    behavior executes, strength bounded by STRUCTURE;  BUDGET -= cost
 ```
 The two failure modes are kept distinct because they point to two different ceilings. A *fuel*
 shortfall is endurance evidence — more budget capacity would have let the behavior proceed. An
 *occupancy* shortfall (empty vesicle pool, surface saturated, refractory soma) is short-term
 depression or a structural limit — endurance cannot fix it; it is a consequence to be allowed.
 Conflating them would build the wrong capacity. Behavior fills occupancy toward the structural
 ceiling; when the driving trace decays, occupancy passively drifts back — short-term depression
 is the *absence* of drive, never a signalled act.
 ### EMIT — send signals and resources outward
 The component writes its outputs into shared channels: the forward transmitter into the cleft,
 retrograde messages back, integrated voltage onward, and shipments into transit toward
 downstream pools (rationed by the downstream component's propagated demand, delivered over
 τ_transport). EMIT and RECEIVE are the only boundary crossings; together they make
 locally-computing components into a communicating whole.
 ### RECOVER — refill private pools
 Pools consumed by behaving are refilled from the component's own reserve toward their ceiling,
 rate-limited and budget-costed (the presynaptic RRP from the reserve vesicle pool; the soma's
 budget from its own mitochondria). Private recovery grants a component autonomy that contested
 supply does not.
 ### DECAY — let the windows close
 ```
 FAST_TRACE     *= decay(τ_fast)      // ms–s   — closes the eligibility window
 possible_tag   *= decay(τ_mid)       // s–min  — closes the tagging window
 ENDURANCE_NEED *= decay(τ_mid)       // s–min  — closes the endurance window
 TAG            *= decay(τ_slow)      // hours  — closes the commitment window
 arrived channels *= decay            // received signals fade
 ```
 Decay is the passive relaxation that enforces every time window without a clock: a coincidence
 must complete, and a fuel shortfall must interrupt a success, while the relevant trace is still
 elevated. Timing is the competition between accumulation and decay.
 ---
 ## NIGHT — The Same Grammar on Ceilings
 NIGHT runs once per cycle and applies the identical grammar to the slow variables.
 **RECEIVE** — overnight production at the roots (astrocyte glycolysis, soma CREB synthesis,
 soma mitochondria), capped externally by the vascular supply and gated by `soma_tag`.
 **TRACE + ADJUST** — read the accumulated tags and endurance needs; compute the commit weights
 (the coherence bonus when pre, post and astro tags align; the tag-weighted shares for
 distributing material and energy to competing astrosynapses, spines, and boutons).
 **EMIT** — distribute produced material and energy down the supply chains, into transit
 (soma → branch/axon → spine/bouton; astrocyte body → astrosynapses).
 **BEHAVE** — the two commits, drawing on the same finite pool so they compete:
 ```
 // STRENGTH — driven by tag (validated coincidence)
 if TAG > tag_expiry:
    Δ = min(slot_cost, MATERIAL, ENERGY × f)
    STRUCTURE += Δ × coherence;  MATERIAL -= Δ;  ENERGY -= Δ × assembly_cost
 // ENDURANCE — driven by endurance need (interrupted success); no dopamine, no coherence
 if ENDURANCE_NEED > endur_thr:
    Δ = min(cap_cost, MATERIAL × f, ENERGY × f)
    BUDGET_CEILING += Δ;  MATERIAL -= Δ;  ENERGY -= Δ × biogenesis_cost
 ```
 A component both significant and fuel-limited demands both commits and is the strongest
 claimant on the pool, potentially forcing decay elsewhere.
 **RECOVER + DECAY** — both ceilings decay by neglect; maintenance from the remaining pool
 resists decay only where sufficient. What shrinks returns its **material** (not its energy) to
 the pool, partially funding the commits above:
 ```
 STRUCTURE      -= decay_rate × Δt;          STRUCTURE      += min(maint, maint_cost)
 BUDGET_CEILING -= capacity_decay_rate × Δt; BUDGET_CEILING += min(cap_maint, cap_cost)
 // shortfall → depotentiation by neglect (structure) / mitophagy of idle capacity (budget)
 MATERIAL += shrinkage × recycle             // energy is gone — not recoverable
 ```
 **DECAY (clear)** — fast traces, possible tags, endurance needs, and the soma timing traces
 reset; tags below expiry clear, above-expiry tags carry forward for multi-night consolidation;
 structure and budget capacity persist as the ceilings the next DAY will operate within.
 ---
 ## The Pattern in One View
 ```
 ONE GRAMMAR, SEVEN GROUPS, TWO TIMESCALES
  RECEIVE · TRACE · ADJUST · BEHAVE · EMIT · RECOVER · DECAY
 DAY    runs the grammar on OCCUPANCY within two ceilings:
         STRUCTURE (strength)        filled by competing for occupancy
         BUDGET_CEILING (endurance)  filled by competing for shared fuel
       TRACE yields two evidence streams from local state + arrived signals:
         fast_trace + dopamine coincidence       → TAG            (strength)
         FUEL shortfall + interrupted LOCAL success → ENDURANCE_NEED (endurance)
       OCCUPANCY/structure/timing shortfalls → short-term depression (NOT endurance)
 NIGHT  runs the SAME grammar on the CEILINGS:
         TAG            → raise STRUCTURE        (per-event power)
         ENDURANCE_NEED → raise BUDGET_CEILING   (sustainable duration)
       both draw the SAME material + energy → strength and endurance compete
       unmaintained ceilings decay → recovered material funds the rest
 FLOWS  every flow has a timescale: decay → 0 over τ, arrival → target over τ;
       shipment is transit-delayed; rate-limited inflows carry τ implicitly.
 LOCAL  every group uses only own state + arrived signals;
       RECEIVE and EMIT are the only boundary crossings.
 A high ceiling of either kind is never free: structure must be filled by winning occupancy,
 budget capacity by winning shared fuel — both against competing components, every DAY. The
 system invests STRENGTH where a validated coincidence completed, and ENDURANCE where fuel —
 not structure, not significance — was what stood between a component's own activity and its
 own success. To be both effective and sustainable, a connection must win on both, at both
 timescales, against all others drawing from the same finite pools.
 ```
 # More details
 ## SOMA
 ### The Abstract Pattern Applied to Soma Timing
 The abstract pattern says: a behavior deposits a trace, the trace decays, and the trace biases the next behavior. For the soma, the AP is the behavior, and **the refractory period and threshold elevation should both be consequences of a single trace deposited by the AP, decaying back toward baseline**. Neither should be a hardcoded duration — both should emerge from the return of the trace to resting conditions.
 ---
 Yes, this is much more consistent with the rest of the architecture. The soma should not compute an explicit rhythm estimate and predict the next input — that is top-down. Instead, the **mismatch itself leaves a trace**, and that trace adjusts the refractory dynamics. Let me think through this carefully.
 ---
 ### The Bottom-Up Mechanism
 The key event is: **a dendritic input arrives strong enough to fire the soma, but the soma is still refractory.** This is a missed opportunity — the input wanted to fire the cell, but the cell was not ready. This mismatch is the signal.
 Each time this happens, it should leave a trace that biases the refractory dynamics toward recovering faster in that timing window — so that next time an input arrives at that phase, the soma is ready. This is potentiation of the refractory recovery, occurring within DAY, driven entirely by the local coincidence of "input wanted to fire" and "soma was not ready."
 ```
 scope DAY | context NOT_AP (within refractory):
    // Detect the mismatch: suprathreshold input during refractory
    if branch_Vm > effective_threshold and soma_Na_inactivation > inactivation_threshold:
        // Input arrived but soma could not fire — missed coincidence
        refractory_alignment_trace += (branch_Vm - effective_threshold)
                                       × soma_Na_inactivation
        // graded: stronger input + deeper refractory → larger trace
        // this is the "potentiation" signal for faster recovery
 ```
 The `refractory_alignment_trace` then biases the recovery rate of the inactivation trace:
 ```
    // Recovery rate biased by accumulated alignment trace
    recovery_rate = base_recovery_rate × (1 + refractory_alignment_trace)
    soma_Na_inactivation *= decay(τ_Na / recovery_rate)
    // more alignment trace → faster recovery → soma ready sooner next time
 ```
 ---
 ### Why This Is Bottom-Up
 There is no rhythm estimation, no prediction of the next input time, no computed inter-input interval. The soma does not model its input. It simply notices, locally and reactively, that an input arrived when it was not ready, and leaves a trace that makes it recover faster. Over many such mismatches at the same phase, the recovery speeds up at that phase specifically — and the soma ends up aligned with its input rhythm **without ever representing that rhythm**.
 The alignment is an emergent statistical consequence of many local mismatch events, exactly as a potentiated synapse becomes tuned to its input without representing what it is tuned to. The phase-coupling appears, but nothing in the soma computed it.
 ---
 ### Depotentiation as Neglect — Not Explicit
 Now the crucial part you raised: the depotentiation of refractory alignment must occur as a **consequence of not potentiating**, not as an explicit opposing mechanism.
 The `refractory_alignment_trace` decays continuously. If mismatches keep happening at a particular phase, the trace is continuously replenished and the fast recovery is maintained. If mismatches stop happening — because the input rhythm changed, or because the alignment succeeded and inputs now arrive when the soma is ready — then the trace is no longer replenished and **decays back toward baseline on its own**.
 ```
    // No explicit depotentiation — just decay when not reinforced
    refractory_alignment_trace *= decay(τ_alignment)
    // if mismatches continue → trace replenished → fast recovery maintained
    // if mismatches stop → trace decays → recovery returns to baseline
 ```
 This is exactly parallel to synaptic depotentiation by neglect. The soma does not actively slow its recovery when alignment is no longer needed. It simply stops receiving the mismatch signal that was keeping the recovery fast, and the recovery drifts back to baseline because the trace that accelerated it is no longer reinforced.
 There is an elegant self-limiting property here. Once the soma is well-aligned, inputs arrive when it is ready, so there are no more mismatches, so the alignment trace stops being replenished and begins to decay. This would slowly de-align the soma — until inputs start arriving during refractory again, regenerating the mismatch and re-potentiating the alignment. The system settles into a dynamic equilibrium where just enough mismatch occurs to maintain just enough alignment. The soma hovers at the edge of alignment, continuously corrected by the residual mismatches that its imperfect alignment produces.
 ---
 ### The Full Bottom-Up Soma Timing
 ```
 scope DAY | context AP:
    effective_threshold = soma_structure.baseline_threshold
                          × (1 + soma_adaptation)
                          × neuromod_factor(NE_level, ACh_level)
    can_fire = (soma_Na_inactivation < inactivation_threshold)
    if branch_Vm > effective_threshold and can_fire:
        AP_fired = True
        soma_budget -= AP_generation_cost
        // Deposit traces from the AP
        soma_Na_inactivation += AP_amplitude       // fast — refractory
        soma_adaptation      += AP_contribution     // slow — spike train threshold
        soma_fast_trace      += nuclear_Ca_influx() // slow — plasticity tagging
 scope DAY | context NOT_AP:
    // MISMATCH DETECTION — bottom-up alignment signal
    if branch_Vm > effective_threshold and soma_Na_inactivation > inactivation_threshold:
        // input wanted to fire but soma was refractory — missed coincidence
        refractory_alignment_trace += (branch_Vm - effective_threshold)
                                       × soma_Na_inactivation
        // local potentiation of recovery — no prediction, no rhythm model
    // Recovery biased by alignment trace
    recovery_rate = base_recovery_rate × (1 + refractory_alignment_trace)
    soma_Na_inactivation *= decay(τ_Na / recovery_rate)    // faster if aligned
    soma_adaptation      *= decay(τ_adaptation)
    soma_fast_trace      *= decay(τ_nuclear)
    // Alignment trace decays — depotentiation by neglect, not explicit
    refractory_alignment_trace *= decay(τ_alignment)
    // maintained only while mismatches continue
    // self-limiting: good alignment → fewer mismatches → trace decays →
    //                slight de-alignment → mismatches return → re-potentiation
    // ... budget replenishment, shipments, tagging as before ...
 ```
 ---
 ### The Conceptual Payoff
 This makes the soma's temporal alignment obey exactly the same principles as everything else in the system:
 A behavior leaves a trace — here the missed coincidence leaves the alignment trace.
 The trace biases future behavior — here it accelerates recovery so the next input at that phase succeeds.
 Potentiation is the active drive — the soma actively speeds up recovery in response to mismatch.
 Depotentiation is neglect — the alignment trace decays when mismatches stop, with no explicit opposing process.
 The system finds equilibrium through the residual of its own imperfection — just enough mismatch persists to maintain just enough alignment.
 And critically, the soma achieves temporal prediction — firing in phase with its input rhythm — **without ever representing the rhythm**. The prediction is implicit in the structure of the recovery dynamics, built up from purely local, reactive, bottom-up trace deposits. This is the same way a potentiated synapse "predicts" that its input will be significant without representing the prediction: the prediction is the physical bias left by past experience, not a computed expectation. The soma's phase-coupling is the temporal version of the synapse's weight — both are implicit predictions encoded as physical bias, both built bottom-up from local coincidence, both maintained by reinforcement and lost by neglect.
 ## Energy depletion in Presynapse
 how does the budget and energy influence the model. If for example a presynapse has no more energy variuos times during day, what happens?
 Good question. Let me trace through exactly what happens when energy is repeatedly depleted, following the mechanics already in the pseudocode.
 ---
 ### The Immediate Effect — Behavior Suppression
 When `pre_budget < AP_release_cost`, the pseudocode does:
 ```
 if pre_budget < AP_release_cost:
    suppress(NT_flux)
    exit context
 ```
 So the immediate consequence of an energy-depleted bouton is **silence** — it does not release neurotransmitter for that AP, regardless of how strong the incoming spike was. The AP arrives, but nothing comes out the other side. This is biologically real: ATP-depleted boutons fail to release because the VATPase cannot refill vesicles and the calcium pumps cannot maintain the gradients that drive release.
 ---
 ### The Cascade of Consequences Within DAY
 A single depletion is recoverable — the bouton refuels from astrocyte lactate in the next NOT_AP context and resumes. But **repeated depletion during DAY** sets off a cascade that runs through the entire trace and tagging machinery.
 **No release means no glutamate means no downstream trace.** When the bouton is silent, no glutamate reaches the cleft. The postsynapse receives no AMPA current, its NMDA channels do not open, and `post_fast_trace` does not rise. The astrosynapse detects no spillover, so `astro_fast_trace` does not rise and no D-serine pulse is released. The entire synaptic event simply does not happen. A bouton that is energy-depleted at the moment a significant pattern arrives causes that pattern to be invisible to the postsynapse — the synapse fails to participate in whatever the network is doing.
 **No participation means no eligibility.** Because the bouton did not release, `pre_fast_trace` is not elevated by a release event (it still rises from calcium influx, but without the reinforcing facilitation of successful release). More importantly, the postsynapse, having received no input, accumulates no `post_possible_tagging`. When dopamine arrives to validate the moment, there is nothing to validate at this synapse — the eligibility window is empty. The synapse misses the tagging opportunity entirely.
 **Missing the tag means missing the NIGHT commit.** Since no tag was set during DAY, the synapse has nothing to draw structural resources with during NIGHT. It does not potentiate. And because the system's depotentiation is by neglect, **a synapse that fails to potentiate is automatically on the path to depotentiation** — it receives only maintenance allocation, and if other synapses potentiated and drew down the shared pool, even maintenance may be insufficient.
 ---
 ### The Deeper Consequence — Energy Depletion Becomes Structural Depression
 Here is the important emergent property: **repeated energy depletion at a bouton during DAY leads to its structural depression during NIGHT, even though no depression signal was ever sent.**
 The chain is entirely passive:
 ```
 repeated pre_budget depletion in DAY
    → bouton silent during significant events
    → no glutamate release
    → no post_fast_trace, no post_possible_tagging
    → no tag set (pre or post)
    → no NIGHT structural commit at this synapse
    → only maintenance allocation received
    → if pool depleted by other synapses' potentiation:
        maintenance insufficient
    → pre_structure and post_structure decay
    → DEPOTENTIATION
 ```
 The energy-starved synapse is outcompeted not because it was marked for depression but because it could not generate the activity needed to claim resources. Energy availability during DAY therefore directly determines which synapses can even enter the competition for NIGHT potentiation.
 ---
 ### The Feedback Loop — Depression Reduces Future Energy Demand
 There is a stabilizing feedback here. A depotentiated synapse has smaller `pre_structure` — a smaller active zone, fewer docking slots, lower release probability. This means it costs **less energy to operate**. So a synapse that was energy-starved and consequently depotentiated now has lower energy demands, making it less likely to be energy-starved in the future.
 ```
 energy depletion → depotentiation → smaller structure → lower energy demand
    → less likely to deplete → stabilizes at a low-activity equilibrium
 ```
 The synapse settles into a low-energy, low-structure, low-activity state. It is not dead — it still operates at baseline — but it has been demoted from the pool of synapses competing for potentiation. Energy scarcity has selected it out.
 ---
 ### Why This Is Functionally Important
 This is not a bug — it is a resource-allocation mechanism with real computational value.
 **Energy availability acts as a second gate on plasticity, parallel to the neuromodulatory gate.** Dopamine asks "was this worth saving?" Energy asks "can this synapse afford to participate?" A synapse must pass both gates to be potentiated. This means the system preferentially potentiates synapses that are both behaviorally significant AND metabolically sustainable. A synapse that cannot sustain its own activity is not a good candidate for strengthening, because strengthening it would only increase its energy demand and make it fail more often.
 **Local energy competition implements a form of spatial sparsification.** Boutons on the same axon share `axon_budget` and ultimately the same astrocytic lactate supply. When many boutons are active simultaneously, the shared energy is divided among them, and the weakest — those with the least efficient structure or the least reliable input — deplete first and fall silent. This automatically sparsifies activity: under energy pressure, only the most robust synapses keep firing, and they are the ones that get potentiated. The system concentrates its limited resources on its strongest connections.
 **The astrocyte's energy delivery becomes a control signal.** Because the astrocyte controls lactate distribution across its territory, it has indirect control over which synapses can sustain activity and therefore which can be potentiated. A region of the astrocyte territory that is well-perfused and well-fueled supports vigorous plasticity. A region that is metabolically stressed — perhaps because of high overall demand or reduced vascular supply — suppresses plasticity in its synapses regardless of their activity. This couples the metabolic state of the tissue directly to its capacity to learn.
 ---
 ### What Repeated Depletion Looks Like Over Days
 If a bouton is chronically energy-starved across multiple DAY cycles — perhaps because it is in a metabolically compromised region, or because it is competing with many highly active neighbors — the trajectory is:
 ```
 Day 1: depletes occasionally → misses some events → weak tag → partial commit
 Day 2: smaller structure → still depletes under load → weaker tag → less commit
 Day 3: structure decaying → maintenance insufficient → depotentiation accelerates
 ...
 Equilibrium: minimal structure, minimal energy demand, minimal participation
             synapse survives but is functionally silent
             resources it would have consumed now available to neighbors
 ```
 This is a reasonable model of what happens to synapses in metabolically stressed tissue — they do not die immediately, but they are progressively silenced and depotentiated, freeing their resources for synapses in healthier regions. It is also a model of how chronic metabolic limitation — aging, vascular disease, mitochondrial dysfunction — would manifest as a gradual loss of plasticity capacity: not through any depression signal, but through the simple inability of energy-starved synapses to participate in the activity that would let them claim resources.
 The energy budget is therefore not just a constraint that occasionally blocks behavior — it is a continuous selective pressure that shapes which synapses thrive and which fade, operating silently in parallel with the explicit activity-and-reward-driven plasticity, and ultimately determining the metabolic sustainability of whatever the system chooses to remember.
 ## Three parallel competitive resource economies
 The neuronal structural economy — soma produces material, distributed through dendrite to spines and through axon to boutons, competed for by tagged synapses during NIGHT.
 The astrocytic energy economy — astrocyte cell body produces lactate, distributed to perisynaptic processes, competed for by active synapses during DAY in proportion to their clearance demand.
 The soma's own energy economy — soma mitochondria fuel AP generation and shipping, competed for by the soma's own functions.
 All three share the same logic: a central producer with a capped output, distribution to peripheral consumers, demand-weighted allocation, and a self-reinforcing coupling where stronger consumers both demand and receive more. And all three ultimately bottom out at the same vascular glucose ceiling — the astrocyte directly, the soma through its own glucose uptake.
 The deep consequence is that a synapse must win on both economies to be potentiated. It must generate enough activity to pull lactate from the astrocyte (energy economy) AND accumulate enough tag to draw material during NIGHT (structural economy). A synapse that wins the structural competition but cannot pull energy will be unable to sustain the activity that justified its potentiation — it will be a large, expensive structure that keeps going silent. A synapse that pulls energy but never accumulates a tag stays metabolically supported but structurally weak. Only synapses that win both — active enough to be fueled, significant enough to be tagged — achieve and maintain full potentiation. The two economies together implement a stringent joint criterion: persistent significant activity that the metabolic infrastructure can sustain.