Update 2026-06-29-tripartite-synapse_v17.md

elements/neuron/appunti/2026-06-29-tripartite-synapse_v17.md

@@ -476,7 +476,10 @@ rather than spike-punctate, so POST's action-context is "no descending AP arrivi
 channels and responds to whatever glutamate has arrived. Three calcium sources feed the fast trace —
 AMPA current (small Ca, begins ejecting the NMDA Mg block) and NMDA (large Ca, scaled by the local
 coincidence of depolarization + astrocyte D-serine + glutamate). The action deposits the calcium
-trace and emits the two retrograde messages. Crucially POST's ACTION **continues through the soma's
+trace and emits **NO** — the fast retrograde, tracking the NMDA calcium transient, confirming the
+response reached a responsive target. (The other retrograde, **eCB**, is slow — produced from
+*accumulated* depolarization and emitted in PREPARATION as an integrated brake on the next release,
+not per-response.) Crucially POST's ACTION **continues through the soma's
 refractory period**: the refractory belongs to the soma, not to POST — from POST it is simply NOT_AP,
 so the spine keeps responding and integrating, charging up so the neuron is ready to fire again once
 its refractory ends.
@@ -538,9 +541,8 @@ DAY | NOT_AP:                                             // respond to arrived
     if Vm > Mg_eject and glutamate > 0:                                              // SOURCE 2 NMDA (Mg relieved)
         post_fast_trace += NMDA_Ca(glutamate) × sat(astro_Dserine, K_Ds)·rise_speed();  post_budget -= NMDA_cost   // @cut affords: THE pre-post-astro COINCIDENCE — joint three-component read (glutamate × depolarisation × D-serine gain); the meeting-site owning none of its inputs [ms, three-component space]
         // D-serine sets the GAIN (how strongly POST responds), not a coincidence gate — dialed by the astrocyte
-        retro_NO += NO_emit(post_fast_trace);  post_budget -= NO_synth_cost           // EMIT + "responsive target"
+        retro_NO += NO_emit(post_fast_trace);  post_budget -= NO_synth_cost           // EMIT + "responsive target" (fast: tracks the NMDA Ca transient)
-    if Vm > eCB_thr:
+    // (eCB is NOT emitted here — it is slow/integrated, produced in PREPARATION from accumulated depolarisation)
-        retro_eCB += eCB_emit(Vm);  post_budget -= eCB_synth_cost                     // EMIT − brake
 
 // ===== RECOVERY (AP: brief descending-spike arrival — restore + boost traces) =====
 DAY | AP:                                                  // soma's spike arrives (ms event), driven from above
@@ -551,6 +553,10 @@ DAY | AP:                                                  // soma's spike arriv
 
 // ===== PREPARATION (NOT_SPIKE_TRAIN: shape the next response AND the NIGHT) =====
 DAY | NOT_SPIKE_TRAIN:                                     // sustained quiet (emitted by NEURON — above SOMA)
+    // slow retrograde: eCB is produced from ACCUMULATED depolarisation (integrated, not per-response) —
+    // a brake on the NEXT release, so it is prepared here, not deposited in the action
+    if post_fast_trace > eCB_thr:
+        retro_eCB += eCB_emit(post_fast_trace);  post_budget -= eCB_synth_cost         // EMIT − brake (slow, integrated)
     // for NIGHT: climb the tag; dopamine is the integrable coincidence (#3)
     if post_fast_trace > Ca_TAG: post_possible_tag += post_fast_trace;  post_budget -= pka_cost   // @cut affords: participation / consistency of co-activity (running average) [min]
     if dopamine > dop_thr and post_possible_tag > tag_thr:
@@ -784,20 +790,26 @@ downstream components receive depends on the soma having been tagged.
 //   NOTE      SOMA endurance fires only on FUEL shortfall (budget < ap_cost);
 //             refractory / sub-threshold are timing limits, not endurance. Own-state proxy.
 
-// THE THREE CATEGORIES (same at DAY and NIGHT; here DAY, subject = the branch input / spike):
+// THE THREE CATEGORIES (SOMA is an INITIATOR — it fires from its OWN integration crossing its OWN
-//   ACTION      fire (the defining deed) — context AP
+//   threshold, so nothing external licenses its ACTION. It runs CONTINUOUS and the ACTION⇄RECOVERY
-//   RECOVERY    restore the ability to fire again: refractory de-inactivation + mito replenish —
+//   alternation is decided INTERNALLY by its own condition. It EMITS AP and REFRACTORY as signals of
-//               the alter-ego of the spike, runs in NOT_AP
+//   these internal phases — it cannot be contextualized by them, since it is their source. Only its
-//   PREPARATION shape the next spike: alignment, adaptation, tag-climb (for NIGHT), ship, decay
+//   PREPARATION is externally contextualized, by the NEURON's NOT_SPIKE_TRAIN. See logic_principles §1.):
+//   ACTION      (CONTINUOUS; internal guard: branch_Vm > threshold and can_fire) fire — the defining
+//               deed. Self-triggered. Emits AP.
+//   RECOVERY    (CONTINUOUS; internal: not firing) restore the ability to fire again: refractory
+//               de-inactivation + mito replenish — the alter-ego of the spike. Emits REFRACTORY.
+//   PREPARATION (NOT_SPIKE_TRAIN, from the NEURON) shape the next spike: alignment, adaptation,
+//               tag-climb (for NIGHT), ship, decay.
 //   Pattern: (ACTION ⇄ RECOVERY) × many spikes, then PREPARATION.
 //   ONE spike's fast trace feeds TWO preparation destinations: nuclear-Ca → tag (for NIGHT),
 //   inactivation/adaptation/alignment → next-spike timing (this scope).
 
-// ===== ACTION =====
+// ===== ACTION (CONTINUOUS — self-triggered by internal threshold-crossing; emits AP) =====
-DAY | AP:                                                  // integrate + fire (the defining deed)
+DAY | CONTINUOUS · action:                                 // integrate + fire (the defining deed)
     threshold = soma_structure.baseline_threshold × (1 + soma_adaptation) × neuromod(NE, ACh)
     can_fire  = soma_Na_inactivation < inactivation
-    if branch_Vm > threshold and can_fire:
+    if branch_Vm > threshold and can_fire:                 // INTERNAL guard licenses the action (no external context)
         if soma_budget < ap_cost:                          // FUEL shortfall → endurance (a PREPARATION deposit)
             if soma_fast_trace > traj_thr and soma_fast_trace_rising:
                 soma_endurance_need += soma_fast_trace
@@ -813,8 +825,8 @@ DAY | AP:                                                  // integrate + fire (
         soma_budget -= creb_cost
         soma_emitted_activity += 1;  soma_emitted_structure = soma_structure   // NEURON sums these
 
-// ===== RECOVERY (restore the ability to fire; alter-ego of AP; runs in NOT_AP) =====
+// ===== RECOVERY (CONTINUOUS — self-entered when not firing; emits REFRACTORY) =====
-DAY | NOT_AP:                                              // de-inactivate + replenish
+DAY | CONTINUOUS · recovery:                               // de-inactivate + replenish (soma's own refractory dynamics)
     soma_budget += fill(soma_budget, soma_budget_ceiling, mito_output, 0, soma_budget)   // mito replenish
     recovery = base_recovery × (1 + soma_refractory_alignment)
     soma_Na_inactivation *= decay(τ_Na / recovery)         // refractory recovery (sped by alignment)
@@ -840,7 +852,7 @@ DAY | NOT_SPIKE_TRAIN:
 //    contend WITHIN (participation arbitrates); material contends BETWEEN components (recovery).
 
 // ===== ACTION (build ⇄ release; participation gates direction; tag funds build, persists across cycles) =====
-NIGHT | build or release:
+NIGHT | NON_REM_1:
     if soma_tag > tag_expiry and soma_participation ≥ MEDIUM:      // BUILD (slice)
         Δ = min(slot_batch, soma_material, soma_energy·f_cap, soma_tag) × soma_participation
         soma_structure += Δ;  soma_material -= Δ;  soma_energy -= Δ·assembly_cost
@@ -856,7 +868,7 @@ NIGHT | build or release:
     // else: HOLD — tag waits for its pattern
 
 // ===== RECOVERY (ROOT production + acquire/free material in CONTENTION; ship to feed the pattern) =====
-NIGHT | produce + import + free:
+NIGHT | NON_REM_2:                                          // (SOMA also ROOT-produces here)
     soma_material += CREB_synth(soma_tag)·Δt_cycle         // ROOT: produce material — recoverable
     soma_energy   += mito_synth()·Δt_cycle                 // ROOT: produce energy — NOT recoverable
     night_energy_spent += mito_synth()·Δt_cycle
@@ -869,7 +881,7 @@ NIGHT | produce + import + free:
     soma_structure += min(soma_maint, maint_cost)          // hold up what is used
 
 // ===== PREPARATION (REPLAY the fire — SAME machinery as day ACTION; ignites the pattern; multi-timescale) =====
-NIGHT | replay-fire + measure + prime:
+NIGHT | REM:                                               // (SOMA replay-fires: ignites the pattern)
     // FAST: replay-fire (probe); ignites the pattern down the day pathway
     spont = intrinsic_fluctuation()
     if spont > soma_spont_thr:                             // ignite (SAME threshold logic as day fire)