Biohybrid Neural Interface Laboratory / Stage 02
The questions that determine
whether a living bridge works
← Open the laboratory · Run controlled experiments · Stage 02 methods & validation
A living interface must be excitable, conduct activity, establish functional junctions and produce an interpretable measurement. These 34 questions connect the biological evidence to the actual calculations, the controls needed to interpret them and the measurements still missing.
Test causality before adding biological complexity. A receiving electrode can record graft activity even when no host neuron fires. The new workbench compares assigned junctions with an otherwise identical synapses-off model and a zero-stimulus control, separates known source contributions, and preserves every tested point in controlled response studies.
The answers describe an idealized numerical laboratory. They do not turn the canonical Hodgkin–Huxley benchmark into a calibrated mammalian graft or a prediction of an implant’s safety or performance.
Each answer has an evidence label. Expand a question to see its practical consequence, verification requirement, remaining uncertainty and primary sources. The labels apply to the stated reasoning, not to the complete simulated architecture.
Evidence & model identity
What the laboratory represents, and which biological claims it can support.
01What is the laboratory actually calculating?
It solves membrane voltages and active sodium/potassium gates on branched neuronal cables, passes terminal events through assigned conductance synapses, and computes electrode voltages from total membrane currents. The anatomy shares the electrical geometry. A successful numerical pathway therefore has a calculated mechanism; its existence does not establish that a particular living graft would behave the same way.
- What this changes
- Retain the verified cable engine and make its measured state, source currents and interventions available to the next experiment layer.
- How to verify it
- Require stable zero-input behavior, calculated action potentials, propagation, conditional synaptic activation and current balance before interpreting any response map.
- What remains unknown
- Channel densities, synthetic geometry and synapse assignments remain surrogate choices rather than measurements of a specific graft.
Sources: Hodgkin AL (1952): A quantitative description of membrane current and its application to conduction and excitation in nerve · Biohybrid Laboratory: Methods & validation
02Which parts of the living-electrode concept have been demonstrated experimentally?
Micro-TENN studies support engineered neuronal populations linked by long axonal tracts within hydrogel microcolumns. Adewole and colleagues demonstrated controlled construct fabrication, optical stimulation/readout and implantation proof of concept. These observations motivate the geometry and relay question. They do not supply this simulator’s electrical threshold, synaptic conductance or chronic transfer function.
- What this changes
- Attach evidence to the particular modeled component, and keep the complete electronic-to-graft-to-host pathway labeled as an explored architecture.
- How to verify it
- Every quantitative experimental claim needs its preparation, measurement modality and source; synthetic defaults must remain distinguishable from those observations.
- What remains unknown
- A matched dataset connecting stimulus, graft spike, host postsynaptic current and extracellular readout is still needed for biological calibration.
Sources: Struzyna LA (2015): Rebuilding Brain Circuitry with Living Micro-Tissue Engineered Neural Networks · Adewole DO (2021): Development of optically controlled “living electrodes” with long-projecting axon tracts for a synaptic brain-machine interface
03What does Science Corporation’s work establish, and what remains an architecture proposal?
The named Brown et al. 2024 preprint concerns an optogenetically stimulated cortical biohybrid implant and behavioral reporting in mice. Science’s separate architecture article discusses neurons associated with electronics and prospective scaling. Optical input and behavioral reporting do not constitute a measured high-bandwidth electrical recording system, and a microwell count is not an electrical channel count.
- What this changes
- Keep the named preprint and company architecture as separate evidence records; do not import prospective neuron counts, bandwidth or longevity as achieved specifications.
- How to verify it
- Source labels must identify the reviewed preprint version and the different status of company-described future capabilities.
- What remains unknown
- Complete integrated-device geometry, simultaneous electrical input/output performance and long-term reproducibility are not calibrated by this laboratory.
Sources: Brown J (2024, bioRxiv v1): Optogenetic stimulation of a cortical biohybrid implant guides goal directed behavior · Hodak M and the Science Corporation team (2024): Biohybrid neural interfaces: an old idea enabling a completely new space of possibilities
04Why retain the canonical Hodgkin–Huxley model at 6.3 °C?
It is a reproducible active-cable benchmark with an independently testable formulation. It provides real membrane dynamics and conduction, but its squid-axon provenance remains visible. Warming the gates with a temperature multiplier or drawing cortical-looking dendrites does not turn it into a validated cortical or stem-cell-derived neuron model.
- What this changes
- Preserve this model as a named benchmark while designing a later, separately identified mammalian calibration track.
- How to verify it
- Export the model version and temperature. Compare waveform, conduction and threshold evidence only within the declared parameterization.
- What remains unknown
- Relevant cell-type kinetics, maturation state and measured current-clamp responses must be supplied together before making preparation-specific claims.
Sources: Hodgkin AL (1952): A quantitative description of membrane current and its application to conduction and excitation in nerve · Biohybrid Laboratory: Methods & validation
05How realistic is the neuronal anatomy?
The geometry contains soma, branched dendrites and segmented axons in physical units, but the branching is synthetic. It is an electrically meaningful test structure rather than a specimen reconstruction. Every rendered electrical segment corresponds to a model segment; display width exaggeration must remain disclosed and must not alter cable area or resistance.
- What this changes
- Use the same geometry for field coupling, compartment inspection and visualization, with physical scale bars and explicit display-only exaggeration.
- How to verify it
- Check compartment identity, branch connectivity, membrane area and scale under mesh refinement and both 3D/2D views.
- What remains unknown
- Real graft arborization, axon caliber distributions and target locations need morphological measurements or a documented reconstruction dataset.
Sources: Biohybrid Laboratory: Methods & validation · Struzyna LA (2015): Rebuilding Brain Circuitry with Living Micro-Tissue Engineered Neural Networks
Causal transmission
Distinguish a functioning modeled relay from direct stimulation and distant electrical sources.
06What evidence is sufficient to call a response dependent on the modeled synaptic relay?
Observed spike order alone is insufficient. Run the same geometry and stimulus with the graft-to-host synaptic conductances removed, and compare each receiving cell. If a host response disappears under that intervention while the graft still conducts, the response depends on the modeled junction under those conditions. This is a model counterfactual, not proof of a biological synapse.
- What this changes
- Provide an automatic matched intact/synapse-off comparison rather than asking users to infer causality from waveform shape.
- How to verify it
- Hold every non-synaptic setting fixed; verify identical graft traces, host outcomes, local synaptic-event order and reproducible comparison results.
- What remains unknown
- If both conditions spike, changed timing or counts can still show a synaptic contribution, but spike presence alone cannot identify an exclusive path.
Sources: Biohybrid Laboratory: Methods & validation · Histed, Bonin & Reid (2009): Direct activation of sparse, distributed populations of cortical neurons by electrical microstimulation
07Could the electrode excite host tissue directly?
Yes. The imposed extracellular field acts on every modeled cable, not only the engineered soma. Histed and colleagues found that electrical microstimulation can directly recruit distributed cortical cells, consistent with activation of nearby axonal elements. In this model, a host spike that remains with junction conductance set to zero is compatible with direct field activation and cannot be credited exclusively to the bridge.
- What this changes
- Label control-persistent host activity explicitly and retain the stimulus field on the host in both compared runs.
- How to verify it
- Include a configuration that produces a host response without synaptic events, plus the default configuration in which the matched control remains below threshold.
- What remains unknown
- A modeled direct response does not identify the excitation site or spatial recruitment pattern of a real cortical preparation.
Sources: Histed, Bonin & Reid (2009): Direct activation of sparse, distributed populations of cortical neurons by electrical microstimulation · Rattay F (1999): The basic mechanism for the electrical stimulation of the nervous system
08Why can the receiving electrode show a waveform when no host neuron fires?
An electrode samples the superposed extracellular field. Currents from the graft can reach a contact near the host, and a subthreshold host membrane can also carry current. A nonzero neural waveform therefore does not imply local host spikes or successful host-graft transmission. Stimulus artifact provides another distinct contribution.
- What this changes
- Show graft-source, host-source, total neural and applied-field components alongside the host spike raster.
- How to verify it
- The disconnected example must retain its nonzero graft-derived recording while reporting zero host recruitment; source components must sum back to the original signal sample by sample.
- What remains unknown
- Source decomposition is available because the simulator knows every source; a real electrode does not automatically provide those separated traces.
Sources: Lindén H (2014): LFPy: a tool for biophysical simulation of extracellular potentials generated by detailed model neurons · Holt GR (1999): Electrical interactions via the extracellular potential near cell bodies · Biohybrid Laboratory: Methods & validation
09Which controls are needed, and what exactly changes in each?
The intact run retains the chosen settings. A synapse-off control changes only the assigned graft-to-host conductance to zero. A no-stimulus control changes only the stimulating current to zero. Geometry, recording position, time step, duration and membrane properties remain matched. A preset called disconnected is useful for teaching, but a causal comparison must derive its control from the actual completed run.
- What this changes
- Store normalized settings for every member of the control set and display the exact intervention.
- How to verify it
- Assert the allowed setting differences, preserve the same sampling grid, and reject mixed-version or mismatched runs instead of presenting them as a matched experiment.
- What remains unknown
- These interventions isolate mechanisms inside the defined model; biological blockers can have additional effects that are not represented here.
10How should the signed spike-timing differences be interpreted?
The current summaries subtract first upward 0 mV crossings: engineered 1 tip minus engineered 1 soma, and earliest host soma minus engineered 1 soma. These are signed timing differences. A negative value means that the tip or host crossed first; distal or mixed initiation can produce this without negative physical travel time. A positive difference also does not establish matched propagation. Physical conduction time requires identifying the same wavefront at specified monitors and in a specified direction.
- What this changes
- Retain legacy metric keys for data compatibility, but label their values as first-crossing differences and name the monitors. Do not clamp, take an absolute value or describe a negative difference as a physical conduction delay.
- How to verify it
- Check the proximal benchmark and a distal-stimulation mixed-activation case against the stored spike times. Missing crossings yield unavailable differences; actual negative differences retain their signs.
- What remains unknown
- These first-crossing summaries do not pair wavefronts across cells or pulses. Directional conduction time and per-pulse synaptic relay latency require additional event matching and causal evidence.
Sources: Biohybrid Laboratory: Methods & validation · Repository: numerical validation and retained discrepancies
11Does a host that fails to spike imply that its synapse is absent?
No. A terminal event can produce a postsynaptic conductance and membrane deflection that remains below spike threshold. The laboratory should distinguish absent connections, no presynaptic terminal event, present but weak postsynaptic input, and host spiking. Those outcomes describe different failure mechanisms and call for different parameter changes.
- What this changes
- Use terminal events, local synaptic events and host voltage together when explaining unsuccessful transmission.
- How to verify it
- Test zero coupling, weak nonzero coupling and sufficient coupling while keeping the stimulus and geometry fixed; the weak case must not be forced to fire.
- What remains unknown
- Actual synaptic failure rates, inhibitory inputs and release probabilities are not inferred from a single deterministic conductance.
Sources: Destexhe A (1994): An efficient method for computing synaptic conductances based on a kinetic model of receptor binding · Biohybrid Laboratory: Methods & validation
Stimulation & recruitment
Interpret controlled parameter changes without inventing a universal threshold or safety limit.
12Is the fraction of recruited hosts a probability of successful integration?
No. It is the fraction of the explicitly modeled host cells that crossed the chosen spike detector in that run. The electrical model is deterministic and the small synthetic population is not a random sample of a clinical graft. Calling 3 of 3 cells a 100% biological success probability would exceed the evidence.
- What this changes
- Report both the integer count and the denominator, and show individual host outcomes in controlled sweeps.
- How to verify it
- Verify recruitment counts from unique host soma identities, not terminal events or repeated spikes; preserve a separate total spike count.
- What remains unknown
- A probabilistic biological estimate would require a defined sampling population, trial variability, an uncertainty model and experimental calibration.
Sources: Biohybrid Laboratory: Methods & validation · Dhobale AV (2018): Assessing functional connectivity across 3D tissue engineered axonal tracts using calcium fluorescence imaging
13What do current polarity and the two pulse phases mean?
The sign specifies current at the stimulating contact relative to its explicit return. The implemented waveform has equal-duration opposite phases, with a configurable interphase gap. The spatial field can depolarize one segment while hyperpolarizing another. A negative first phase is not a universal guarantee of excitation, and phase order can change the response.
- What this changes
- Preserve signed current and expose polarity as an experimental variable rather than a decorative switch.
- How to verify it
- Invert the imposed field with current sign, integrate each delivered phase including fractional solver intervals, and recompute membrane dynamics for both polarities.
- What remains unknown
- The current source is ideal; electrode polarization and compliance limits are not modeled.
Sources: Rattay F (1999): The basic mechanism for the electrical stimulation of the nervous system · Biohybrid Laboratory: Methods & validation
14Can a parameter sweep report a single activation threshold?
A finite sweep can identify tested values that did and did not activate a specified cell or relay outcome. It should report those response brackets, the grid spacing and any reversals. A single interpolated threshold is justified only under additional monotonicity assumptions. Experiments in retinal ganglion cells have demonstrated an upper stimulation threshold, so a globally monotone response cannot be assumed for all neural systems.
- What this changes
- Present the complete sampled response curve and retain transitions in both directions instead of hiding them behind one threshold number.
- How to verify it
- Test bracket extraction on monotone, all-negative, all-positive, discontinuous and nonmonotone outcomes; no extrapolated threshold may appear outside the tested interval.
- What remains unknown
- The cited retinal observation motivates a safeguard; its numerical thresholds are not transferred to this synthetic bridge.
Sources: Boinagrov et al. (2012): Upper threshold of extracellular neural stimulation · Biohybrid Laboratory: Methods & validation
15Should a strength–duration curve be forced to follow the Lapicque formula?
No. Calculate each duration/current condition using the active compartments first. Extracellular strength–duration behavior depends on field geometry and channel dynamics; Boinagrov and colleagues showed departures from simple intracellular descriptions. A fitted rheobase or chronaxie would be a secondary descriptive fit with residuals and a stated range, not a replacement for the solver.
- What this changes
- Start with bounded one-parameter response maps and explicit pulse duration. Add threshold-versus-duration fitting only after verified bracketing across several durations.
- How to verify it
- Use sufficient temporal resolution for each phase, report non-activation and nonmonotonicity, and compare brackets under time-step refinement.
- What remains unknown
- No fitted time constant from the surrogate should be labeled as a measured graft or human-neuron chronaxie.
Sources: Boinagrov, Loudin & Palanker (2010): Strength–duration relationship for extracellular neural stimulation: numerical and analytical models · Hodgkin AL (1952): A quantitative description of membrane current and its application to conduction and excitation in nerve
16What stimulation dose should be reported, and does charge balance establish safety?
Report current, phase duration, phase charge and nominal geometric charge density. For a rectangular phase, Q[nC] = |I[µA]| × duration[ms]. With a disk radius r[µm], density[µC/cm²] = Q[nC] × 10⁵/(πr²). Equal opposite phase charge gives zero ideal net charge, but this model has no electrode electrochemistry, tissue injury, heating or chronic exposure law, so it cannot certify a safe stimulus.
- What this changes
- Keep dose arithmetic visible with units and label it as ideal-source geometry, not a device safety recommendation.
- How to verify it
- Check unit conversions analytically, phase balance numerically and the radius-squared scaling; do not add an unsupported universal pass/fail safety threshold.
- What remains unknown
- Material, effective area, electrode polarization, repetition history and preparation-specific injury measurements are required for safety assessment; empirical models such as Shannon’s have their own domain of evidence.
Sources: Shannon (1992): A model of safe levels for electrical stimulation · Biohybrid Laboratory: Methods & validation
17How can a user tell which parameter caused a change?
Vary one declared parameter while holding the remaining normalized settings fixed. Include the baseline value where practical and record every condition, including unsuccessful or invalid conditions. The resulting relationship is conditional on that baseline. A one-factor sweep cannot estimate all interactions between conductivity, position, morphology and synaptic strength.
- What this changes
- Provide reproducible amplitude, coupling, recording-distance and conduction-related sweeps with visible units and base settings.
- How to verify it
- Verify the generated settings, number and order of cases, unchanged non-swept parameters, complete result rows and deterministic reruns.
- What remains unknown
- Broader uncertainty analysis needs justified parameter ranges and a designed multivariable study; arbitrary sliders are not probability distributions.
18How do axon length and diameter affect conduction in this model?
Length changes the engineered tract path; the shared axon-diameter setting changes graft trunks and host axons, with graft terminal branches at 0.8 times that diameter. These changes enter membrane area and axial resistance. The retained velocity field divides path length by a positive first tip-minus-soma difference and is unavailable otherwise. That ratio is a conditional apparent velocity; physical conduction velocity requires matched propagation.
- What this changes
- State the shared diameter scope when interpreting recruitment or recording changes. Keep channel parameters and temperature fixed for controlled comparisons, and distinguish the timing summary from a matched traveling event.
- How to verify it
- Check physical path length and affected axon diameters. Require positive matched travel time only for a conduction interpretation; preserve signed first-crossing differences in mixed-activation cases and test temporal/spatial refinement.
- What remains unknown
- This sweep cannot attribute every response change solely to graft-trunk caliber. The unmyelinated HH surrogate has no nodes of Ranvier, myelin remodeling or calibrated mammalian diameter–velocity law.
Sources: Hodgkin AL (1952): A quantitative description of membrane current and its application to conduction and excitation in nerve · Biohybrid Laboratory: Methods & validation
Extracellular measurements
Understand what electrodes measure and what the ideal conductor leaves unresolved.
19Can graft and host electrode contributions be separated without rerunning the cells?
Yes, within the stated linear volume conductor. At each saved time, multiply each compartment’s total membrane current by its contact transfer coefficient and sum by source identity. The graft and host contributions then add to the neural waveform. This partitions sources within one run; it is different from subtracting intact and disconnected runs, whose host dynamics can differ nonlinearly.
- What this changes
- Add source-resolved traces and a reconstruction residual while retaining the original total neural trace.
- How to verify it
- Check graft + host = neural for every contact/sample, and neural + applied field = ideal total. Preserve existing voltage and event results.
- What remains unknown
- The partition is model ground truth, not an inverse method that can uniquely recover source populations from experimental electrode data.
Sources: Lindén H (2014): LFPy: a tool for biophysical simulation of extracellular potentials generated by detailed model neurons · Holt GR (1999): Electrical interactions via the extracellular potential near cell bodies
20What does the finite-contact correction include?
The ideal recording value is a spatial average of extracellular potential over a flat disk, estimated with deterministic quadrature. The stimulation disk similarly distributes the imposed current over sampled sites. This includes a stated contact size rather than an ideal point, but it does not solve electrode polarization, insulating shanks or the conductivity boundary around a metal object.
- What this changes
- Retain contact dimensions and overlap/clearance checks, and describe the method as disk averaging rather than finite-element electrode physics.
- How to verify it
- Compare disk quadrature against analytic on-axis cases and distant point limits; reject contacts that intersect membranes even between quadrature sites.
- What remains unknown
- Near-contact predictions require more detailed material boundaries and converged contact integration before experimental interpretation.
Sources: Lindén H (2014): LFPy: a tool for biophysical simulation of extracellular potentials generated by detailed model neurons · Biohybrid Laboratory: Methods & validation
21Should doubling electrode distance or conductivity always halve the recorded waveform?
Only a controlled isolated point source obeys an exact inverse-distance law. An extended cell includes spatially distributed sources and sinks whose cancellation can change with location. In the homogeneous model, fixed source currents do scale inversely with conductivity; however, rerunning field stimulation at a different conductivity can also change those source currents and recruitment.
- What this changes
- Separate a recording-position experiment with unchanged cell dynamics from a full conductivity perturbation of stimulation and recording.
- How to verify it
- Test inverse-distance behavior on an analytic point source, inverse conductivity on fixed transfer coefficients, and unchanged membrane arrays when only passive recording position changes.
- What remains unknown
- Scaffold boundaries, anisotropy and frequency-dependent tissue properties are outside the homogeneous-conductor approximation.
Sources: Holt GR (1999): Electrical interactions via the extracellular potential near cell bodies · Biohybrid Laboratory: Methods & validation
22What are the voltage reference and stimulus-artifact assumptions?
The contact potentials use a distant reference in an infinite conductor, while stimulation includes an explicit source and return. The applied-field trace is the ideal volume-conductor artifact. A real amplifier measures a difference between finite contacts and may saturate, filter or blank samples during stimulation; those effects are not recreated by simply hiding the artifact curve.
- What this changes
- Label neural, applied-field and ideal-total views separately, including the chosen reference convention.
- How to verify it
- Require an equal return current, reproducible field sign, exact component reconstruction and zero applied artifact outside the specified pulse support.
- What remains unknown
- Amplifier range, input impedance, differential reference placement, filtering and recovery need a separately specified acquisition model.
Sources: Rattay F (1999): The basic mechanism for the electrical stimulation of the nervous system · Biohybrid Laboratory: Methods & validation
23Does a larger ideal extracellular waveform mean that a device can record more information?
Not by itself. Peak-to-peak voltage describes one waveform under an ideal reference. Detectability also depends on noise, bandwidth, artifact recovery and the number of independent signals. Information transfer additionally requires a defined input distribution, trial variability and decoding task. None can be inferred from the number of visible neurons or contacts.
- What this changes
- Report waveform amplitude as amplitude; avoid attaching an SNR, bit rate or channel-count claim without the corresponding acquisition and statistical model.
- How to verify it
- Keep units and measurement windows explicit and ensure that source-resolved amplitudes are not treated as additive peak-to-peak values.
- What remains unknown
- Measured electronics noise, biological background, electrode impedance and task data are needed before a recording-performance or information claim.
Sources: Biohybrid Laboratory: Methods & validation · Hodak M and the Science Corporation team (2024): Biohybrid neural interfaces: an old idea enabling a completely new space of possibilities
Numerical & software evidence
Test the calculation, reproducibility, exports and user interface separately.
24How do we know the time step is adequate?
An implicit voltage solve can remain stable while producing inaccurate spikes. Compare at least the production time step and two finer steps on a common saved-time grid. Check membrane-waveform differences, matched event times and counts. Near a threshold, a tiny discretization change can alter recruitment, so a passing suprathreshold waveform check is not enough to validate every sweep boundary.
- What this changes
- Retain the existing temporal study and refine selected response transitions in the next-stage validation set.
- How to verify it
- Use predeclared error criteria and preserve disagreements. Report differences between discrete solutions rather than calling them exact-solution errors.
- What remains unknown
- Observed reduction with refinement is evidence over the tested regime, not a proof of global accuracy or formal convergence order.
Sources: Hines M (1984): Efficient computation of branched nerve equations · Rush S (1978): A practical algorithm for solving dynamic membrane equations · Repository: numerical validation and retained discrepancies
25How do we distinguish a finer mesh from a changed experiment?
Refinement must preserve the physical branch geometry, stimulus contacts and synapse landing locations. Moving a synapse to whichever compartment happens to be nearest can change its anatomical target and produce a false convergence result. The foundational release retained a failed spatial comparison caused by exactly this kind of target change, then fixed the geometry rather than relaxing its tolerance.
- What this changes
- Treat synapse location as a physical anatomical quantity and only change its electrical discretization under mesh refinement.
- How to verify it
- Compare target coordinates, cable path, recruitment, conduction delay and contact waveforms at 25, 12.5 and 6.25 µm where applicable.
- What remains unknown
- A converged synthetic cable is still a synthetic cable; spatial numerical convergence cannot validate the biological morphology.
Sources: Repository: numerical validation and retained discrepancies · Biohybrid Laboratory: Methods & validation
26What does agreement with native NEURON validate?
The stored reference comparison tests another implementation of the same HH cable, conductance synapses, geometry and finite-step conventions. Close agreement supports the implementation and unit bookkeeping. It does not eliminate both solvers’ discretization error, prove an untested parameter regime, or show agreement with living neurons.
- What this changes
- Preserve the original benchmark and source identities when adding analysis-only measurements; rerun reference verification if the numerical equations or update order change.
- How to verify it
- Check waveform error, event count/timing, extracellular signals, reference version and source hashes; retain the raw comparison traces.
- What remains unknown
- Biological validity needs experimental voltage and current data beyond software-to-software agreement.
Sources: Repository: native NEURON comparison, raw traces and source identities · Biohybrid Laboratory: Methods & validation
27What must an exported experiment contain to be reproducible?
It needs normalized settings, model and analysis versions, physical units, the source/current convention, sampling and solver resolutions, and every sweep/control condition. A summary metric without its settings is not reproducible. Electrical replay should reproduce the same arrays in the same runtime, while growth additionally requires its random seed and model parameters.
- What this changes
- Export machine-readable experiment metadata and tabular results separately from a screenshot, including source components and control identities.
- How to verify it
- Round-trip settings, verify every exported sample/column and event count, reject malformed input, and compare deterministic reruns. Verify browser file saving separately from serialization.
- What remains unknown
- Bitwise identity across different JavaScript runtimes is not promised; numerical tolerances and version identities are necessary for cross-platform comparisons.
Sources: Biohybrid Laboratory: Methods & validation · Repository: exploratory growth model and its limits
28How can long calculations stay responsive without showing stale or mixed results?
Run numerical work in a worker and treat each request as a separate generation. Cancel or supersede earlier work when settings change, and bind completed measurements to the exact settings that produced them. Partial sweep results must be labeled partial; a cancelled worker must never overwrite a newer experiment.
- What this changes
- Use progress, cancellation, bounded case counts and explicit completed-run provenance in the experiment workbench.
- How to verify it
- Test cancellation during a sweep, edits before completion, rapid reruns, stale messages and malformed worker errors. Keep rendering and user interaction responsive.
- What remains unknown
- Runtime depends on device, population size and resolution; measured development-machine timings are not a universal performance guarantee.
29What counts as verification of the scientific visualization?
Inspect both themes, responsive layouts, replay, selection and parameter changes in the browser. Confirm that the displayed frame and compartment values come from returned numerical state. Source review can check renderer logic, but it cannot substitute for testing WebGL rendering and controls on a GPU-enabled browser. A tested 2D fallback should remain available when WebGL is absent.
- What this changes
- Publish browser observations and remaining gaps separately from numerical results and build success.
- How to verify it
- Check anatomy/trace synchronization, units, time scrubbing, focus, keyboard use, clipping where supported, and mobile overflow; label unsupported verification explicitly.
- What remains unknown
- The first release’s cloud browser could not enable WebGL. That gap remains open until direct 3D testing succeeds in a supported environment.
30When is the next stage actually complete and published?
Completion requires implemented behavior, targeted numerical regression, preserved site-wide tests, a successful production build and inspectable source history. Publication additionally requires the intended commit to be merged, Cloudflare deployment to succeed and the public routes to show the new behavior. Queued checks, a local preview or generated code alone do not meet that standard.
- What this changes
- Keep a release report with branch, commit, PR, exact tests, quantitative discrepancies, build and deployment identities, and live verification.
- How to verify it
- Verify the deployed content against the tested source, exercise an intact/control comparison and a sweep, and confirm directory/navigation entries.
- What remains unknown
- Any unavailable device, GPU or browser-download check must remain visible instead of being implied by a successful build.
Sources: Biohybrid Laboratory: Methods & validation · Repository: numerical validation and retained discrepancies
Biology & the next calibration
Keep growth, integration and chronic performance tied to the evidence actually available.
31When does a modeled growing neurite become a functional synapse?
A geometric encounter is only an encounter. The exploratory module adds a separately assumed formation probability and maturation interval before marking a candidate. Even that candidate is not an experimentally verified synapse. Transferring candidate count into electrical settings requires an explicit action, and postsynaptic firing still depends on the subsequent electrical solve.
- What this changes
- Keep encounter, candidate maturation and electrical activation as distinct states and keep days separate from millisecond electrophysiology.
- How to verify it
- Use seeded replay, verify no future events leak into the growth time scrubber, and show exactly what is transferred: target count, not grown trajectories or measured weights.
- What remains unknown
- Culture elongation measurements do not determine in-vivo guidance, synapse formation probability, survival or circuit specificity.
Sources: Repository: exploratory growth model and its limits · Adewole DO (2021): Development of optically controlled “living electrodes” with long-projecting axon tracts for a synaptic brain-machine interface
32Can the interface-degradation experiment predict long-term implantation performance?
It can show how specified conductivity, separation and coupling changes affect this model. Those parameters are not a clock for months after implantation. Biran and colleagues measured tissue responses around chronic silicon implants, but that evidence does not calibrate a living graft’s survival, immune response or mechanical evolution.
- What this changes
- Present degradation as a perturbation study with explicit changed parameters and avoid chronic survival curves without longitudinal data.
- How to verify it
- Separate effects on stimulation, neural recruitment and recording amplitude; do not interpret a smaller waveform as proof of neuron loss.
- What remains unknown
- Viability, vascularization, immunity, glia, micromotion, material degradation and electrode impedance require independent measured models.
Sources: Biran R (2005): Neuronal cell loss accompanies the brain tissue response to chronically implanted silicon microelectrode arrays · Biohybrid Laboratory: Methods & validation
33Does more connectivity imply learning, selectivity or useful network computation?
No. The current links are fixed excitatory conductances. Adding them can alter recruitment and timing, but there is no validated plasticity rule, inhibitory balance, task objective or learned representation. A larger population would also introduce convergence and performance costs, so it should follow demonstrated value from the small inspectable network.
- What this changes
- First quantify per-host recruitment and stimulus dependence; add plasticity or more cell classes only with a declared model and independent checks.
- How to verify it
- Keep connection counts and weights explicit, check deterministic recruitment, and avoid naming connectivity changes learning or restoration.
- What remains unknown
- Measured cell types, inhibitory/excitatory connectivity, short-term dynamics and task-level behavior are needed to support stronger network claims.
Sources: Destexhe A (1994): An efficient method for computing synaptic conductances based on a kinetic model of receptor binding · Dhobale AV (2018): Assessing functional connectivity across 3D tissue engineered axonal tracts using calcium fluorescence imaging · Biohybrid Laboratory: Methods & validation
34What evidence would justify the next biological model beyond this benchmark?
Choose a named preparation and obtain morphology, temperature, passive properties, current-clamp responses, spike shape and conduction measurements, plus paired or otherwise causal host-graft transmission measurements. Fit a defined subset, reserve independent protocols for validation, and retain the HH benchmark beside the new model. Complexity earns predictive value only when those new predictions survive tests outside the fit.
- What this changes
- Make causal controls and reproducible response maps the immediate stage, then use them as the test harness for a separately versioned cortical or stem-cell-derived model.
- How to verify it
- Predeclare fitted quantities, held-out protocols, uncertainty ranges and acceptance criteria; report model failures and parameter non-identifiability.
- What remains unknown
- No current dataset in this laboratory establishes graft-specific thresholds, chronic stability, safe stimulation or human efficacy.
Sources: Adewole DO (2021): Development of optically controlled “living electrodes” with long-projecting axon tracts for a synaptic brain-machine interface · Dhobale AV (2018): Assessing functional connectivity across 3D tissue engineered axonal tracts using calcium fluorescence imaging · Biohybrid Laboratory: Methods & validation
Run a matched causal audit, compare it with the calculated signal sources, and read the quantitative verification and retained discrepancies.