How a neuron grows

01 / Evidence and model assumptions

Mechanisms, parameter choices and validation claims are tracked separately. A measured response in one species can motivate a rule without supplying a transferable numerical constant. This matters especially when comparing a two-dimensional adhesive plate with a three-dimensional tissue-like environment.

For example, the motor-clutch study constrains an Aplysia preparation, while the 3D CNS study demonstrates extension without adhesive traction in its matrices. The laboratory uses a reduced model to explore mechanisms; it does not combine these results into a universal law of neuronal growth.

Microscopy and the version 4 extension

The visual target is an early neuron on an adhesive culture surface: a small, irregular soma connected to thin, unequal processes, with most rapid exploration concentrated at their tips. Primary time-lapse sequences were inspected alongside the supplied recordings. Unlabeled recordings guide appearance only; their species, scale and acquisition interval are not inferred.

Reading the display. Phase contrast and fluorescence use approximate contrast transfer functions. The cortex silhouette comes from the solved mesh; tips, filopodia, buds and wave locations come from saved model state. Intracellular granularity is illustrative. Display taper stays inside conservative collision capsules; it does not alter the cylindrical material ledger. The 1× radius setting preserves model scale except for minimum visible pixel widths. Mechanics shows the physical envelopes. Playback interpolation never feeds back into growth or exports, and pausing displays a completed solver frame.

For a close look, choose Single neuron · 12 h, then Tip and the Tip study · detailed motion playback setting. The microscopy Tip camera follows the selected cone; shift-drag restores manual framing. The initial branched-neuron observation is calculated from time zero and pauses at 48 modeled hours. Reset to 0 h makes the full history visible.

The default state interval represents three model minutes. It supports coarse growth and exploratory motion; sub-minute membrane ruffling and molecular fluctuations are not resolved. Smooth playback does not add those missing dynamics.

Computed motion, for comparison

02 / Membrane and cytoskeletal mechanics

The membrane has a mechanical role: it surrounds a changing geometry and imposes a cost when growth demands additional surface. Actin-rich protrusions explore the local environment; microtubule support distinguishes an exploratory protrusion from maintained neurite extension. Adhesion transmits force between the cytoskeleton and the environment. Craig et al. and Pulido Cifuentes et al. motivate these coupled processes.

Geometry and mechanical degrees of freedom

Each soma has 162 moving membrane vertices and 320 triangular faces, with a nominal radius of 6.8–7.7 µm. Actin, adhesion and MT support vary across that surface. A separately moving spherical nucleus is tethered to the cortex and constrained inside it. Neurite roots attach to membrane vertices, with reciprocal forces from a distributed shaft network. Shared joints and every daughter move together. Axial strain, bending and adhesive anchors constrain motion. This is a discrete viscoelastic reduction, not a calibrated continuum beam model; cell division is not included.

v_poly = 64 a (0.65 + 0.35 S)                         [µm/h]
F_actin = η max(0, v_poly − v_retro)                    [pN]
F_MT = η v_MT min(a, 1) m                              [pN]
F_membrane = 2π r_eff γ                               [pN]
F_net = F_actin + F_MT − F_membrane                    [pN]
v_tip = max(0, F_net / η) g_kind g_age m_support g_material

η = 3 pN·h/µm; v_MT = 12 µm/h in organoid mode, 8 otherwise
r_eff = 0.52 µm for axons; 0.85 µm for dendrites
g_kind = 1 for axons; 0.58 for dendrites
g_age = clamp((developmental age − 8 h) / 6 h, 0.2, 1)
m_support = 0 when u = 0; otherwise m

k_eff = k_substrate k_clutch / (k_substrate + k_clutch)
f_bond = F_b log(1 + k_eff (v_retro/3600) / (k_off,0 F_b))
k_off = k_off,0 exp(min(15, f_bond/F_b))
occupancy = k_on / (k_on + k_off)
F_traction = N_clutch adhesion occupancy f_bond
v_target = v_unloaded / (1 + F_traction/F_stall)
v_retro ← 0.5 v_retro + 0.5 v_target    (24 iterations)

E = γ A + KA/(2 A0) (A − A0)² + KV/(2 V0) (V − V0)²
    − p0 (V − V0) + Σ_edges ke/2 (l − l0)² + kb/2 Σ_i |Lx_i|²
F_passive = −∇E
x_i ← x_i + Δt F_i / ζ
p0 = 2γ / cbrt(3V0 / (4π))

Adhesion, shape change and cell migration

Cell polarity defines a leading and trailing side. Local actin state relaxes toward 0.12 + 0.88 front² over 1.2 h, while MT support responds over 3 h. Active protrusion and rear contraction are internally balanced: their mean force is removed. Only transmitted substrate or matrix traction supplies active net force. Traction vanishes when adhesion is zero.

access = exp(−gap_to_dish / 2 µm) on the plate; 1 in matrix modes
on = 5 adhesion (0.15 + front) access                  [h⁻¹]
off = 0.8 (1 + 4 rear contractility)                  [h⁻¹]
b_next = on/(on+off) + (b − on/(on+off)) exp(−(on+off) Δt)
T_i = gain Fmotor b_i front² stiffness/(stiffness+100)
      / (1 + 2 adhesion²)                            [pN]

Fmotor is 620 pN on the dish and 110 pN for the effective matrix coupling. The gain combines actin, motility, the arbor-dependent settlement multiplier below and a developmental factor: 1 for stem-like/precursor cells, 0.45 immature, 0.20 polarized and 0.08 maturing. The cortical network distributes 85% of total traction; 15% remains local. Dish adhesion also pulls the basal cortex toward the nonpenetrating substrate plane. These are declared exploratory closures, not measured human-cell parameters.

Short-range cell adhesion acts within 7 µm of conservative soma envelopes. Contact inhibition rotates polarity, and contact constraints resist movement. Clusters are measured from local contacts; a final cluster position is never specified. The deformable-cell literature motivates these mechanisms, but does not validate this particular reduction.

Mechanical membrane is distinct from electrical membrane

Surface area, tension, deformation and contact belong to the growth model. A separate frozen-geometry electrical assay tests transmission using a reduced conductance-based point-neuron model. Its electrical parameters are declared independently; they are not inferred from morphology or color. Channel waveforms and extracellular recordings are not computed. Its membrane visualization does not imply that a lipid bilayer, individual channels or molecular actin filaments are resolved.

Why force balance is overdamped

Growth is represented as directed, resisted movement. Contact changes the admissible direction, local forces or growth state. It is not modeled as a fast rigid-body impact with a coefficient of restitution. A stalled tip can continue probing; a retracting tip removes distal structure instead of creating a backwards extension.

Substrate stiffness is not a universal growth accelerator. Its effect depends on force transmission, adhesion and the other terms in the model. A parameter sweep should inspect extension, branching, contact and resource use together; a single attractive final arbor cannot identify the causal mechanism.

03 / From precursor to a polarized neuron

The precursor starting state describes developmental competence in the model. It does not simulate reprogramming, gene regulation, pluripotency, lineage specification or the laboratory steps of a human differentiation protocol. Developmental stages gate when processes can emerge and acquire axonal or dendritic behavior.

Local microtubule stabilization is one experimentally supported contributor to axon specification. The laboratory reduces this complex process to explicit developmental state rules. A rule that selects an axon is not proof that the model has recovered the entire polarity signaling network.

Initiation chooses four to six separated, actin/MT-weighted sites on the actual membrane. This produces variable process origins rather than an identical radial star. At the assigned polarity time, the root arbor with the largest score, total arbor length + 6 µm × guidance gain × alignment with the cue direction, becomes the axon arbor. This is an explicit heuristic, not a molecular symmetry-breaking calculation. Its existing segments change identity together; the others retain dendritic behavior. The “maturing” label records developmental age and does not imply functional electrical maturity.

The developmental clock is an assumed model clock. Human iPSC-derived neurons can share early morphological stages with rodent cultures while following different growth kinetics. “24 h” in a displayed experiment therefore means 24 modeled hours under the selected rules, not a forecast that a stem cell becomes a mature human neuron in one day.

Soma–arbor mechanical coupling

Version 4 corrects straight proximal tethers left behind moving somata. Distributed neurite mechanics and reciprocal soma forces replace independent soma motion followed by root stretching. Growing a first dendrite does not permanently freeze a cell.

Each segment saves its stress-free length, anchor, adhesion fraction and preferred junction angle. Axial and bending energy gradients exert equal and opposite endpoint forces, including a reaction on the attached cortex vertex. Shared branch joints update every daughter atomically. Drag and yielding substrate anchors resist motion; swept collision and membrane-budget checks guard each accepted change.

Eaxial = EA (L − L0)² / [2 max(L0, 0.25 µm)]
EA = 75 (r / 0.5 µm)² pN
Ebend = (6 pN·µm / 2) (u_parent · u_child − cos_rest)²
Node drag = 12 + 90 adhesion [pN·h/µm]
Anchor stiffness = 18 adhesion [pN/µm]
Reference-length relaxation time = 24 h

Tip growth adds stress-free material; elastic displacement does not. Geometric area changes debit or recycle the existing membrane reserve. Anchor turnover permits slipping instead of permanent adhesion. A 4 µm/h node-motion limiter and a 40% elastic-strain guard are numerical safeguards, not measured tissue failure thresholds. These coefficients are exploratory and require preparation-specific fitting.

Somal active drive gradually decreases with arbor age, extent and substrate adhesion: its multiplier is 1 / [1 + settlement × adhesion × (1 − exp(−root age/8 h)) × min(20, arbor length/15 µm)]. It remains positive. Growth-cone actin dynamics continue independently. The paired experiments can remove settlement while retaining coupling, or remove coupling to expose its contribution.

The contact bridges are reduced local adhesion links, not separately meshed membranes or collision capsules. The network is not a mesh-converged continuum beam model. V4 saved files include new material and contact state; older files are rejected explicitly rather than silently converted to different physics.

04 / Axons and dendrites need different rules

A neuron's axon and dendrites share cytoskeletal machinery but serve different geometric roles. The lab treats process identity as part of the state, so differences in extension, steering, branching, taper and contact can be examined directly. These are representative behaviors, not a single morphology shared by every neuronal class.

λ = λ_kind branchingGain S m / (1 + 0.3 branchOrder)
p_branch = 1 − exp(−λ Δt)

Independent local episodes alternate advance, pauses and brief retractions. A second process launches actin-associated packets at neurite roots. Packets move at 120–180 µm/h along the retained tree, take one daughter at a branch, and increase local tip drive on arrival. The source rate is an exploratory 1.4 h⁻¹ per root in the immature stage, reduced to 55% after polarization and 25% in the maturing stage. Pulse strength decays with a 0.42 h time constant. Wave dependence declines with stage; the early culture experiment is not imposed as a universal gate on older neurites. Growth episodes = 0 removes packets and restores continuous drive.

Rare tip splits require an extending cone at least 3 h old, microtubule support ≥0.45, reserve ≥0.2 and available capacity. Most new branches instead begin as shaft precursors. Their birth hazard is 0.12 × branching gain × min(5, eligible cable length / 70 µm) per hour. Eligible nodes are non-root, at least 1 h old, beyond the soma envelope plus 6 µm, and have one live continuation. Buds extend or retract visibly; only those that survive long enough, exceed 2 µm and acquire sufficient MT/material support become retained branches. New branches can also prune during their first five modeled hours. These rates, survival choices and angle distributions are not fitted biological measurements. Branch order is limited to five and active tips to 32 per cell by default.

Newly deposited unbranched joints near active tips relax toward a straighter configuration, attenuated by adhesion and age. Motion is capped at 0.6 µm/h and vanishes after 12 h of shaft age. Expanded swept envelopes reject paths through solids, foreign somata, unrelated shafts or foreign cones. Both attached segment endpoints move together, and added/removed cylindrical area enters the membrane reserve ledger. This is a local overdamped reduction, not a complete elastic rod solver.

Steering combines the current direction with cue, nearby-process, obstacle and environment terms. Same-cell dendrite avoidance and other-cell axon alignment operate within 12 µm. Directional fluctuations use an Ornstein–Uhlenbeck process with a 1.5 h correlation time. These are dimensionless steering weights, not solved receptor kinetics or a three-dimensional tissue-stress field.

When calculated speed is below 0.02 µm/h for over 2 h, specified low-driver, low-reserve or negative-force conditions trigger retraction. Persistent non-neurite collision after 1.8 h triggers 2.2 µm/h retraction. The distal endpoint retreats along its existing path; a parent with surviving children is retained. The model does not implement activity-dependent pruning. Synaptic contact maturation is a separate process described below.

Read dendrite growth as a history

An arbor's final branch count hides how it was assembled. Record branch births, retractions, surviving tips, cable length and the spatial distribution of branches. The growth-and-retraction study demonstrates why pruning is part of development rather than merely a correction to an already completed tree.

Self-avoidance is cell-identity dependent. Branches from the same neuron can respond differently from processes belonging to a neighbor. Mammalian protocadherin experiments support this distinction in specific cell classes. A global repulsion rule applied equally to every dendrite would erase it.

Read axon growth as extension, guidance and branching

Measure path length alongside displacement and tortuosity. An axon can accumulate cable while making little progress toward a target. Branch initiation must also be distinguished from extension: cytoskeletal perturbations in cortical neurons can affect these processes differently.

An axon that follows an existing process has a different encounter history from one that turns away from an obstacle. Alignment is a geometric model behavior; it is not a molecular assay of a particular cell-adhesion molecule.

05 / Collision, recognition and connection

Three questions must remain separate: do two finite structures physically intersect; how does a tip respond to that encounter; and does a biological connection eventually function? The first is geometry, the second is a modeled behavioral response, and the third requires evidence beyond geometry.

A spatial hash finds candidate segments over their full bounding boxes. Proposed extensions are checked along the complete swept path: sphere intersections for round obstacles and conservative soma envelopes, finite-capsule intersections for neurites, and slab intersections for expanded boxes. Expanded box corners are conservative; the soma exclusion radius is the maximum of its current vertex distances and 1.06 times its nominal radius. This conservative envelope can overexclude narrow passages. Nearby parent segments and shared branch junctions have explicit exclusions. A separate check on the proposed terminal-shaft chord prevents an endpoint update from cutting through a corner.

Rigid obstacles, somata and world boundaries exclude the effective growth-cone radius: the maximum radius of the twelve-node, force-dependent growth-cone rim (initially 1.8 µm for axons and 2.1 µm for dendrites). Established neurites use finite shaft radii. A moving cone additionally excludes foreign shafts and other cells’ active cones at its current rim radius; local contact can compress the rim. Same-cell branch junctions share material. The rim relaxes under effective actin, MT and membrane loads. It is a coarse radial model, not a molecular lamellipodium. Nine finite-radius exploratory protrusions per cone have separate seeded extension/retraction lifetimes and directions. Their maximum exploratory length is 7 µm and free extension rate is 26 µm/h; load shortens visible extension. Lamellar sectors have their own renewal streams. Protrusions sweep against solids, foreign shafts and cones, somata and boundaries, and blocked directions bias subsequent steering. The rates are model choices. A collision clips growth before entry, turns direction toward a tangent and accumulates a stall interval.

Finite cones and filopodia can create an encounter record only after touching a foreign opposite-kind process. Molecular-competence proxies combine maturation, presynaptic organizer expression and postsynaptic receptivity. Self contacts and axon–axon alignment do not create synapses. Records retain material-site coordinates when an axon continues growing past the contact.

A brief recognition pause replaces permanent tip arrest. Continuous apposition for 0.3 h with sufficient adhesion creates a candidate contact. A compliant tensile bond links the sites. Loss of recognition, loss of adhesion, deletion of either segment or extension of the bond more than 1.2 µm beyond its formation length removes the contact and its conductance. Nearby sites on the same cell pair are deduplicated within 3 µm.

Presynaptic assembly and postsynaptic receptors evolve separately, with exploratory 3 h and 8 h time constants. Structural assembly requires 2 h of contact and a presynaptic fraction of at least 0.3. Transmission readiness requires presynaptic fraction ≥0.6 and receptor fraction ≥0.55. Conductance is 1.2 times their product in nS. Local probing, partner selection and silent structural assemblies motivate the separation; these model clocks are not measured culture kinetics.

The ordinary plate has no soma-secreted attraction field. The optional Illustrative cell cue assay activates saved secretion and receptor factors. It uses a quasi-steady screened diffusion/decay field proportional to 8 exp(−r/35)/(r+8), with r in µm, and the gradient of saturating occupancy c/(1+c). A receptor can attract, repel or ignore a source. This is a generic bath-field experiment, not a calibrated netrin concentration or a universal rule that neurons grow toward every other soma.

Testing transmission on the grown graph

The Connections panel runs a 200 ms assay while growth is frozen. It injects 220 pA into the selected cell from 20–80 ms and repeats the identical stimulus with synaptic transmission blocked. Only transmission-ready, still-apposed graph edges deliver signals. The displayed voltage traces and threshold events come from the calculation; a nearby fiber or colored marker alone cannot transmit.

C dV/dt = gL (EL − V) + gE (EE − V) + gI (EI − V) + I
C = 100 pF; gL = 5 nS; EL = −65 mV
Threshold = −50 mV; reset = −65 mV; refractory = 2 ms
EE = 0 mV; EI = −75 mV; conductance decay = 5 / 10 ms
Transmission delay = 1 ms; numerical step = 0.05 ms

This is a reduced conductance-based integrate-and-fire assay, not a fitted reconstruction. The illustrative initial population gives every fifth cell inhibitory output and the others excitatory output. Only dendritic targets are currently represented. Cable filtering, axonal conduction, channel waveforms, probabilistic release, plasticity, axosomatic/axoaxonic synapses and gap junctions remain outside this assay. Growth state and random streams are unchanged by electrical testing. The assay download contains both control arms and all parameters.

06 / Four environments, four interpretation limits

Environmental presets are experiments with the same evolving cell representation. Agreement with plate data would not validate an organoid, a regeneration gap or a biohybrid construct. Each context needs its own geometry, boundary conditions, measured parameters and held-out comparison.

Myelination is deferred. The current experiment focuses on unmyelinated neurite development and contact. It does not solve oligodendrocyte or Schwann-cell wrapping, internode selection, g-ratio, saltatory conduction or remyelination.

07 / What to measure

The laboratory measures centroid displacement, traveled path, speed, traction, adhesion occupancy, area/volume strain, mesh quality, contact clusters, sister-branch angles and x–z projected dendritic convex-hull area. The engine also records separate axonal and dendritic cable lengths, surviving branch nodes, active tips, branch-order counts, longest paths, radial reach, shaft area, signed cone speed, separate extension/retraction rates, live/allocated segments, appositions and candidates, membrane area, tension and reserve. Longest paths begin at the neurite root on the soma surface; radial reach is measured from the soma center. Sholl intersections use exact segment–sphere crossings at radii of 15, 30, 45, 60, 90, 120, 160, 200 and 260 µm around each soma. The recorded profiles include combined processes, axons alone and dendrites alone; the workbench compares the two identities separately.

The default microscopy projection, structural surface renderer and software 3D fallback read the same solved membrane and nucleus geometry. They provide rotation, close-up inspection and a micrometre scale bar. Phase contrast and fluorescence approximate microscopy; Structural 3D and Mechanics expose modeled structure. No optical image-formation physics is implied. Radius settings of 2× and 3× enlarge only the display, leaving collisions and measurements unchanged. Nuclei have simulated positions and physical exclusion radii. Individual internal filament lines illustrate coarse MT support and do not represent independently solved polymers. A projected crossing does not by itself establish a physical contact.

Collision episodes combine consecutive hits by a tip on the same modeled object; changing object or a gap longer than 1.5 timesteps begins another episode. The internal collision-step counter and retraction-step counter remain update counts. None is a count of unique biological cell pairs. Tortuosity and full branch lifetimes can be analyzed separately from exported geometry and saved states.

The JSON preserves coordinates, topology and state; the CSV summarizes recorded history; SWC exports one selected cell's geometry for external analysis. SWC does not encode force states, a lipid surface, or validated electrical parameters. Shaft area is a cylindrical lateral-area sum and is separate from the triangulated soma area.

For a mechanical experiment, inspect membrane demand and the force terms that caused movement. Compare matched starting conditions, keep the seed, and change one mechanism at a time before testing interactions among controls. Summary history is sampled approximately hourly and retains at most 1,450 entries; the event log retains the latest 300 events. These are bounded summaries, not a full continuous microscopy movie.

Different seeds represent different realizations of the model's stochastic rules. They are not independent biological donors, and their spread is not a clinical confidence interval. A parameter sweep can reveal sensitivity inside the model without establishing how likely a real culture is to behave that way.

08 / Verification and the next calibration experiment

Numerical verification asks whether finite-radius obstacles are detected between timesteps, whether branch topology remains valid, whether states stay finite, whether resources constrain growth, and whether saving and resuming preserves an experiment. Refinement checks examine whether results change as a numerical step is reduced. These checks support implementation correctness within their tested conditions.

The generated release report records 21 passing acceptance checks out of 21 for cell-growth/4.0.0. Read the quantitative report. Acceptance criteria and reproduction commands are listed below. Source hashes and the actual runtime are retained in the report.

Coupled roots and a grown connection graph

The V4 mechanics and connectivity report contains 7 of 7 passing gates. Three matched 24 h experiments expose the original long-root artifact, retain coupling with settlement disabled, and then enable both mechanisms. The soma continues moving in both coupled cases.

A separate eight-cell culture starts from random placement and grows for 48 h. It produces 67 transmission-ready sites through finite local encounters. Each source with a ready outgoing contact is stimulated with and without transmission. The report records both response amplitudes. No contact or electrical edge is inserted into that culture. This checks the model's internal causal chain; it does not establish empirical connectivity.

Multi-seed, multi-day geometry study

The cohort report records 64 of 64 runs across 16 predeclared seeds and four environments. Each requests 72 model hours, with full-tree topology, material and separation audits every 12 hours. Each final state must restore exactly. Capacity stops are recorded explicitly; these are numerical robustness checks, not biological replicates or a fit to microscopy.

Deformable-cell verification

The mechanical verification report adds 13 passing checks: analytic energy gradients, passive and unbound controls, directional migration, nucleus containment, area/volume constraints, exact continuation, stress tests and actual mechanical-step refinement. At default gains the polarized single-cell control moves 20.33 µm in eight model hours. These are model outcomes, not measured precursor-cell speeds.

This four-hour control varies the internal mechanical step itself. Varying only the outer growth timestep is a separate subcycling-invariance test. Neither demonstrates spatial convergence of the fixed 162-vertex cortex. Extreme high-drive, low-stiffness settings can reach the mesh-quality guard and reject steps; the inspector reports these constraints instead of silently accepting folded geometry.

Environment studies and allocation limits

These are the exact configurations retained in the generated report. An allocation cap is a computational stopping condition. A run that stops at its segment budget has not completed the remaining requested biological time.

Deterministic timestep refinement

The coarse-to-finest length difference is 0.846%, against a 5% limit. The intermediate step is closer to the finest-step result. This is a bounded continuous-growth refinement study, not proof of convergence of every stochastic arbor, contact history or membrane shape. The finest solution is a numerical reference, not an exact analytic answer.

The same-seed and save/resume checks compare states inside the same JavaScript runtime. Cross-browser bit identity is not promised. Stochastic event sampling and the order of interacting tips can make individual trajectories diverge when the numerical step or population is changed.

node --test tests/cell-growth*.mjs
node scripts/validate-cell-growth.mjs
node scripts/validate-cell-growth-mechanics.mjs
node scripts/validate-cell-growth-cohort.mjs 16
node scripts/validate-cell-growth-connectivity.mjs
npm run build
node scripts/verify-cell-growth-build.mjs

The default step is 0.05 h. A new shaft segment is created after its predecessor reaches approximately 2.5 µm; an individual advance can exceed that nominal length. Swept collision checks still examine the complete proposed movement. Geometry and branching can change at large steps, so the declared refinement result should not be generalized beyond its conditions. The global allocation budget is 18,000 segments, used by the culture and regeneration presets. The organoid preset uses 26,000 and the biohybrid preset uses 26,000. These budgets include retired segments; reaching the chosen budget stops further stepping.

Experimental scale comparison — not calibration

Koch et al. reported mean net traction of 71 ± 15 pN (SE) for their hippocampal growth cones and 284 ± 58 pN for early DRG cultures. Their actin flow measurements on laminin-coated glass were 9.5 ± 0.3 and 3.4 ± 0.3 µm/min, respectively. This lab starts from an exploratory unloaded flow of 1.10 µm/min; its effective spring setting is not their hydrogel modulus. These differences and unlike conditions prevent a quantitative validation claim. Whole-cell motor traction must not be compared with a single growth cone's traction.

The strongest remaining limitations are uncalibrated kinetics, conservative soma envelopes that can overexclude narrow gaps, a fixed cortex topology, reduced radial growth cones, and exploratory discrete shaft mechanics and generic synapse kinetics. The separate electrical assay is a reduced graph test; calibrated functional connectivity, full organoid development and myelination are not established.

A minimum empirical calibration plan

1. Choose a preparation. Specify species, neuronal class, cell line, differentiation stage, matrix, coating, density and imaging interval. Do not pool unlike preparations into one default.
2. Track development over time. Annotate somata, axons, dendrites, branch births and losses, extension/retraction velocities, turning angles and contact outcomes. Retain failed and stalled trajectories.
3. Constrain mechanisms. Obtain membrane-area or tension measurements where feasible, actin-flow or traction measurements, and microtubule-state measurements. Morphology alone cannot uniquely identify all mechanical coefficients.
4. Fit and test separately. Fit a limited parameter set to one subset of experiments, lock the model, and evaluate held-out cells, times and perturbations. Report residuals and uncertainty rather than only representative images.
5. Validate the environment and the connection claim. Repeat the process for each context. Add paired electrophysiology or appropriately controlled functional imaging before assigning functional synapses to geometric appositions.

A useful first acceptance target is simultaneous agreement with extension-speed distributions, branch lifetimes, Sholl profiles and encounter outcomes. Matching only total length is underconstrained: multiple wrong combinations of growth and retraction can produce the same length.

09 / Primary-source register

These sources support specific mechanisms and interpretation boundaries. None independently validates this entire simulator.