diff --git a/elements/neuron/appunti/2026-07-08-verifable-behaviors.md b/elements/neuron/appunti/2026-07-08-verifable-behaviors.md
new file mode 100644
index 0000000..0a5badd
--- /dev/null
+++ b/elements/neuron/appunti/2026-07-08-verifable-behaviors.md
@@ -0,0 +1,125 @@
+# The Logic of Aggregations — v1
+
+*How combinations of components, across scales and timescales and across day and night, make the
+high-level observables — space, time, quantity, and their compounds (flow, frequency, rate) —
+verifiable. This document sits between the two others: the pseudocode says what each component does
+(mechanism); logic_principles says why the model is the kind of object it is and states the general
+principles (the four operations, the ring, the two turnings); this document says what the general
+principles look like when specific components combine. The unit of analysis here is never a single
+component — it is an aggregation.*
+
+*Why a component has no logic of its own. Asking "what is the logic of the presynapse" is the wrong
+question. A single component reads only what integrates in its own stores; the high-level observables
+— flow, frequency, spatial extent — exist only where several components' emissions accumulate
+together. The logic is therefore a property of relationships, not of parts: it lives in
+PRE ⇄ POST ⇄ ASTROSYNAPSE, in the branch that sums its spines, in the territory that sums its
+processes. This document identifies the aggregations worth describing and, for each, what observable
+it makes verifiable, at what scale, over what timescale, and how that changes between day and night.*
+
+---
+
+## The frame: what "verify" means here
+
+The three axes along which we conceptualize any behavior are **time, space, and quantity**. What an
+aggregation *verifies* is a behavior read along one or more of these axes — and such reads compose
+into the high-level observables: a **frequency** (events per time), a **flow** (quantity per time),
+an **elapsed interval** (time between events), a **specific amount at a specific moment** (quantity
+at a time), a **spatial extent** (quantity over space), a **coincidence** (several things at one time
+and place). None of these is emitted by any component; each is *constituted* by many local acts and
+becomes readable only at a site that holds a store whose instantaneous level is that behavior's
+present shadow (see logic_principles §7, Integrate). The verification is not a computation anyone
+performs — it is the automatic consequence of local acts accumulating in a store, here or above.
+
+So to describe an aggregation's logic is to answer four questions:
+
+1. **What behavior along the axes** does this aggregation make verifiable — read along time, space,
+   quantity, or their compounds (frequency, flow, elapsed interval, amount-at-a-moment, extent,
+   coincidence)?
+2. **At what site** is it integrated — where does the store live that carries its shadow?
+3. **Over what timescale** — is it a fast trace (frequency, recency), a medium store (occupancy,
+   participation), or a slow one (structure)?
+4. **How does it change between day and night** — day constitutes the behavior from activity; night
+   reads a re-evoked version of it to decide what structure to keep.
+
+The aggregations below are the ones where these four have interesting, non-obvious answers.
+
+---
+
+## Aggregation 1 — PRE ⇄ POST ⇄ ASTROSYNAPSE (the tripartite unit, one synapse)
+
+This is the smallest complete aggregation and the foundational one. Its logic is that **the three
+axes get three separate owners**, so a single coincidence is not merely detected but *decomposed* —
+each axis verified independently, then recombined at a meeting-site none of the three owns.
+
+**What behavior it verifies: a coincidence, decomposed into its three axes.** A coincidence is
+several things at one time and place. In a two-party synapse the coincidence would be an
+undifferentiated fact — release met response, or it did not. The third party changes this. Because
+each party owns one axis, the coincidence separates:
+- **PRE owns quantity** — how much was released (the amount, set by its own occupancy × drive).
+- **The astrosynapse owns timing and space** — its clearance sets how long transmitter dwells (the
+  temporal window), its coverage sets whether release stays local or spills (spatial isolation).
+- **POST owns the recombination** — its NMDA receptor is where quantity (glutamate), the postsynaptic
+  contribution (depolarization), and the astrosynapse's gain (D-serine) meet. It reads the coincidence
+  *as scaled by* the timing and gain the other two set.
+
+So the aggregation verifies not "did they coincide" but "how much, how sharply timed, how spatially
+contained" — three separable reads, because three owners. This is why the synapse is tripartite
+rather than bipartite: three independent axes want three independent controllers (see
+logic_principles §2), and here that abstract claim is a concrete division of who-holds-what.
+
+**At what site: one site per axis, plus a fourth for the meeting.** There is no single integration
+site. Quantity is integrated in PRE's release; the timing/space read is held in the astrosynapse's
+clearance and coverage state; the gain is held in its D-serine level. The *coincidence event* is
+integrated at a fourth site — POST's NMDA receptor — which holds none of the three axis-stores but is
+where their present shadows overlap. This is the meeting-site principle (logic_principles §7,
+Coincide): the site that reads the coincidence owns none of the signals it compares. The astrosynapse
+is the party that makes the coincidence *witnessable* under locality — it supplies the third input
+(D-serine gain) that neither coinciding party owns, so the event can be registered without any party
+reading another's interior.
+
+**Over what timescale: the same three parties read at all three rungs at once.** The coincidence
+itself is instantaneous (fast) — glutamate, depolarization, D-serine present together for
+milliseconds. But the same aggregation simultaneously constitutes slower reads: over minutes, the
+running participation of this synapse (how often the coincidence recurred) accumulates as occupancy
+and, dopamine-gated at PRE and POST, as a tag; overnight, the astrosynapse's coverage — the timing
+axis itself — is rebuilt against accumulated spillover. So one tripartite unit verifies an
+instantaneous coincidence, a minutes-long participation, and an overnight structural need,
+*with the same three parties*, each axis-owner reading its axis at every timescale it matters.
+
+**Day vs night: detect-and-transmit becomes re-verify-and-consolidate, same machinery.** By day the
+coincidence is the event that drives transmission and deposits evidence (the tag at PRE/POST, the
+coverage-need at the astrosynapse). By night the same three parties re-run the coincidence as a probe
+— PRE's replay release, POST's replay response, the astrosynapse's replay clearance — not to transmit
+but to *verify whether the coincidence still holds* when the pattern is re-evoked without the world
+driving it. The participation this re-verification measures is what decides whether each party keeps
+its structure. So the aggregation's day-role (constitute the coincidence from behavior) and
+night-role (re-verify it to arbitrate structure) are the identical three-party mechanism read for two
+purposes — which is the rotation of logic_principles §3, seen at the level of the unit rather than the
+component.
+
+**The one-line logic of this aggregation:** *three axes, three owners, one meeting-site — so a
+coincidence is decomposed into independently-verifiable amount, timing, and extent, and the same
+decomposition is read fast (the event), medium (participation), and slow (structure), by day to
+transmit and by night to consolidate.*
+
+---
+
+## The queue — aggregations still to build
+
+- **The spine's channels → POST (sub-synaptic).** How many ion channels integrate into one
+  postsynaptic response; the smallest integration, and where "quantity at a moment" first becomes a
+  single readable value.
+- **The branch's spines → DEND (the clearest spatial integrator).** How the dendrite sums spatially
+  distributed spine voltages into one branch signal — space made present at a site; and how attention
+  (ACh) reweights the spatial sum.
+- **The soma's branches → SOMA (convergence to a decision).** How dendritic integrals converge into
+  one fire/no-fire; where frequency (the soma's own firing rate) and flow (summed input) are read;
+  the point at which continuous integration becomes a discrete event.
+- **The astrocyte's processes → ASTROCYTE (territory-scale coincidence).** How local astrosynaptic
+  calcium integrates into the regenerative spike; coincidence detection one scale up (synapse-synapse
+  coincidence, not pre-post), broadcast back to all processes.
+- **The pathway loop (PRE→POST→DEND→SOMA→AXON→PRE).** How flow and timing become verifiable around a
+  recurrent loop; how the night's replay reads the loop's coherence as a mechanical all-or-nothing
+  (every link primed or the pattern breaks).
+- **Across timescales within one aggregation** (general note, once several are built).
+- **Across day and night for one aggregation** (general note, once several are built).