# Formative Propagation Grammar (FPG) ## Blueprint for Success v0.1 **Working status:** Research blueprint / hypothesis program **Date:** 2026-08-21 **Placement:** Foundational rung beneath *Formative Field Computing (FFC)* **Epistemic posture:** Proposed framework and experimental program. It is not an established theory, not evidence of a universal law, and not a claim about consciousness. --- ## Executive thesis The program begins from a simple candidate relation: \[ \boxed{ \text{structured perturbation} \times \text{formed medium} \times \text{current susceptibility} \times \text{formation history} \rightarrow \text{selectively recruited continuation} } \] The recursion that makes it interesting is: \[ \boxed{ \text{recruitment} \rightarrow \text{formation} \rightarrow \text{changed medium} \rightarrow \text{changed future susceptibility} } \] The working intuition is not that a signal necessarily carries a symbolic destination. A formed structure may transform the signal while it propagates, and only structures presently compatible with the resulting condition may respond. > **The address may be a condition rather than a location.** This gives a more precise target than "brainwaves are important," "spiderwebs are fractal," or "fields compute." The target is an experimentally testable grammar of **conditioned recruitment through a formed, history-bearing propagation medium**. A second formulation captures the strict CRR connection: > **Can a consequence of prior formation become a condition-sensitive future route without encoding the destination explicitly?** The immediate goal is not to prove that this grammar is universal. The goal is to find out whether it exists at all, where it exists, what physical state carries it, how much explanatory compression it provides, and whether it survives controlled transfer across radically different substrates. Success is therefore not "the theory wins." Success is **progressively shrinking the unknown while preserving every negative result, closure, and falsifier.** --- # 1. Why this belongs below Formative Field Computing FFC currently asks whether a structured global or regional field can interact with a history-responsive physical substrate so that local response depends on local state, formation history, material susceptibility, and the applied condition. FPG is one rung more primitive. It does not require an electromagnetic field, a magnetic material, a neuron, a spider, or even biology. It asks whether the following relation can be established in any formed propagation medium: \[ F_t + G_t + S_t + H_t \rightarrow R_t \] where: - \(F_t\) = structured perturbation at time \(t\); - \(G_t\) = formed topology / geometry / connectivity; - \(S_t\) = current physical or functional susceptibility; - \(H_t\) = consequential formation history not already captured by the declared snapshot; - \(R_t\) = recruited response / propagation outcome. If the response changes the medium or susceptibility: \[ (G_t,S_t,H_t,R_t) \rightarrow (G_{t+1},S_{t+1},H_{t+1}) \] then the same later perturbation can encounter a different admissibility landscape. FFC becomes a candidate engineered physical realization of this deeper grammar: \[ \boxed{ \text{FPG} \rightarrow \text{candidate physical implementation through FFC} } \] For now, store FPG **under FFC as a foundations program**, not merged into the FFC paper itself. Recommended structure: ```text Formative_Field_Computing/ ├── 00_FFC/ ├── 01_Foundations/ │ └── Formative_Propagation_Grammar/ │ ├── BLUEPRINT_v0.1.md │ ├── protocols/ │ ├── source_registry/ │ ├── preregistrations/ │ ├── analyses/ │ ├── simulations/ │ ├── remote_hardware/ │ ├── results/ │ └── negative_results/ └── 02_FFC_Engineering/ ``` If FPG becomes large enough to stand independently, it can later graduate into its own project without changing the conceptual lineage. --- # 2. What the spiderweb changed The useful observation is not that spiderwebs resemble neural networks or sacred geometry. The useful observation is that an orb web can simultaneously be: - a progressively constructed morphology; - a boundary-conditioned physical network; - a persistent consequence of prior construction; - a propagation medium; - a mechanical filter; - a sensor surface; - a routing structure; - a damage-bearing structure; - a repairable structure; - and part of the animal's action-perception loop. The spider does not need to carry an explicit coordinate list for the completed web. Construction can be studied as a sequence of local decisions made on a structure that the spider itself is continuously changing. A new line is not merely output. It becomes part of the state on which later construction depends. That gives the basic formation loop: \[ \boxed{ \text{construct} \rightarrow \text{change available structure} \rightarrow \text{change subsequent construction possibilities} } \] The same web also carries perturbations. A disturbance does not need to arrive at the spider as an abstract label. It has physical structure: \[ F(t)= \{ A,\omega,\phi,\Delta t, \text{envelope}, \text{harmonics}, \text{direction}, \text{sequence} \} \] and the web transforms it according to topology, tension, damping, material, junction geometry, boundary conditions, and current state. A first-order representation is: \[ F_{\text{out}} = T(G_t,P_t,S_t)\,F_{\text{in}} \] where \(P_t\) denotes physical parameters such as stiffness, damping, line density, and tension. If prior formation has altered those parameters, then: \[ F_{\text{out}} = T(G_t,P_t,S_t,H_t)\,F_{\text{in}} \] The central experimental question becomes: > **Can formed structure turn a perturbation into a route?** A stronger version is: > **Can prior formation change which later routes are recruited by an otherwise matched perturbation?** --- # 3. Provenance of the hypothesis The origin of the spiderweb hypothesis should be preserved because it matters to the exposure audit, not because a dream is evidence. The candidate instruction — *look to spiderwebs for the grammar we seek* — was generated before the literature search that revealed the relevant construction, vibration, playback, simulation, and behavioral datasets. It was resisted rather than welcomed, persisted across more than one sleep cycle, and was verbalized to Melissa before serious external adjudication. This establishes only: \[ \boxed{ \text{candidate generation preceded targeted evidence inspection} } \] It does **not** establish: - privileged access to truth; - supernatural information; - scientific evidence for the hypothesis; - or unusual reliability of dream-generated ideas. It does justify preserving future cases prospectively. ### Sleep-hypothesis ledger For any future deliberate overnight problem-solving attempt: 1. Write the problem before sleep. 2. Record what information was already known. 3. On waking, record the returned candidate **verbatim before searching**. 4. Record confidence and immediate objections. 5. Hash/timestamp the entry. 6. Record the exposure audit. 7. Define what external evidence would support, weaken, or falsify it. 8. Only then search. 9. Preserve misses and useless outputs as carefully as hits. This converts an established personal ideation practice into a testable hypothesis-generation process without granting it epistemic authority. --- # 4. The conceptual stack FPG should unify existing work without swallowing it. ## 4.1 RAdT — the broad admissibility framework RAdT asks whether formation history changes the space, cost, probability, or authority of subsequent formation. Its core relation remains: \[ A_t = A_G(X_t,H_t,E_t,C_t) \] with: \[ X_{t+1}\in A_t. \] FPG supplies a candidate **propagation mechanism** by which a changed formed medium could alter what is recruitable next. FPG does not prove RAdT. RAdT does not require FPG. ## 4.2 FSS — the formation vocabulary FSS contributes the useful distinctions: - formation grammar; - formative motif; - formative residue; - deprivation shadow; - return pathway; - generative density; - formative closure; - entroresilience; - representation lag. FPG should use these only where they improve measurement rather than decorate the work. ## 4.3 Ageometrics / Sequence–Snapshot Divergence The core audit is: \[ P(Y\mid S,U,H) \quad\text{versus}\quad P(Y\mid S,U). \] For FPG: - \(S\) = best declared present-state description; - \(U\) = controlled perturbation; - \(H\) = prior formation history; - \(Y\) = propagation/recruitment/behavioral outcome. If increasingly complete present state screens history off: \[ P(Y\mid S^*,U,H)=P(Y\mid S^*,U), \] then **snapshot closure** has been achieved. That is a successful result. The historical influence has been localized into present physical state. If history remains predictive after a serious effort to complete \(S^*\), the residual becomes a new object of investigation, not a license to declare mystery. ## 4.4 CRR — conditioned reconstruction CRR asks whether dormant prior state can be selectively reactivated by present conditions acting on persistent consequences of prior formation. Its governing distinction must remain: > **return address = navigation, not evidence** and the stricter challenge remains: > **Can a consequence of prior formation serve as a condition-sensitive return path without encoding the destination explicitly?** FPG is a candidate physical mechanism for generating such return paths. ## 4.5 Bounded Selective Reconstruction BSR contributes the operational discipline: > **preserve broadly / activate selectively / expand when locally justified / retain a return path** and the concept of the **Minimum Sufficient Present**: the smallest active or reconstructable state sufficient for the present task. FPG asks how a formed propagation medium might physically help select that state. ## 4.6 Formative Field Computing FFC is downstream engineering: \[ \text{structured field} + \text{history-responsive substrate} \rightarrow \text{selective admissible activation}. \] FPG asks the prior question: > What grammar must any physical medium satisfy before "fieldborne formation" is more than a metaphor? ## 4.7 BECS / NeuroSignal BECS and NeuroSignal already emphasize signal **fidelity**, not maximum signal strength: - amplitude; - timing; - duration; - routing; - reception; - interpretation; - feedback; - noise floor. FPG adds the possibility that the propagation medium and receiver susceptibility are themselves partly formed by prior interaction. ## 4.8 Baby AI / FormationCore / Cognitive Basin The machine-continuity work provides a software analogue of the same discipline: \[ \text{encounter} \rightarrow \text{record} \rightarrow \text{formation consequence} \rightarrow \text{changed later admissibility}. \] FPG should not be retrofitted as proof that Baby AI is brain-like. The useful connection is architectural: persistent consequence should be allowed to change future accessibility without rewriting the past or granting history automatic authority. ## 4.9 Evolution Evolution provides a larger-scale conceptual analogue: \[ \text{surviving formation} \rightarrow \text{reachable future variation}. \] Extinction can remove entire future lineage pathways; ecological disruption can simultaneously open others. The important principle is not teleology but: > **evolution searches outward from surviving formation, not freely across all conceivable forms.** This is a potential scale-invariance test of the grammar much later. It is not part of the first experimental claim. --- # 5. Working formalism Let a formed system at time \(t\) be: \[ \Sigma_t=(G_t,P_t,S_t,H_t) \] where: - \(G_t\): graph/topology/geometry; - \(P_t\): physical edge/node parameters; - \(S_t\): current dynamic susceptibility; - \(H_t\): prior formation history not intentionally assumed to be reducible to \(G_t,P_t,S_t\). A perturbation is: \[ F_t = F(A,\omega,\phi,\Delta t,\mathcal E,D,Q,\ldots) \] where \(\mathcal E\) is envelope, \(D\) directionality, and \(Q\) temporal sequence or modulation. A propagation operator produces a distributed response: \[ R_t = \mathcal T(\Sigma_t,F_t) \] and a recruitment operator identifies the activated subset: \[ K_t = \mathcal R(R_t,\Sigma_t). \] The resulting response may modify the system: \[ \Sigma_{t+1} = \mathcal U(\Sigma_t,F_t,R_t,K_t). \] The effective admissible set is: \[ \mathcal A_t = \mathcal A(\Sigma_t,F_t). \] The core recursive claim under test is therefore: \[ \boxed{ \Sigma_t \rightarrow \mathcal A_t \rightarrow K_t \rightarrow \Sigma_{t+1} \rightarrow \mathcal A_{t+1} } \] No universality is assumed. --- # 6. Candidate measurable quantities ## 6.1 Formative susceptibility How strongly does a location/subsystem respond to a declared perturbation under its present formed state? \[ \chi_i(F,t) \] This can be scalar, spectral, phase-dependent, or state-dependent. ## 6.2 Propagation transfer function For source \(a\) and receiver \(b\): \[ T_{a\rightarrow b}(\omega,t) = \frac{R_b(\omega,t)}{F_a(\omega,t)}. \] Useful measurements include: - amplitude transfer; - phase lag; - group delay; - attenuation; - resonance peaks; - bandwidth; - transient shape; - cross-frequency conversion where relevant. ## 6.3 Recruitment set \[ K_t(F)=\{i:\chi_i(F,t)>\theta_i\} \] The threshold need not be fixed if the native system uses graded response; ranking or probability can be used instead. ## 6.4 Formation-induced routing divergence Apply the same perturbation before and after a declared formation event: \[ D_{\mathrm{route}} = d\big(K_t(F),K_{t+1}(F)\big). \] A nonzero value is not automatically a history effect; it first shows that the present substrate changed. ## 6.5 Snapshot Closure Burden How much present-state description must be added before history no longer improves prediction? Conceptually: \[ B_{\mathrm{closure}} = \min_{S^*} I(S^*) \quad \text{s.t.} \quad P(Y\mid S^*,F,H) = P(Y\mid S^*,F). \] This is a practical Ageometrics quantity. ## 6.6 Address Leakage Fraction For strict CRR-style tests, quantify how much explicit destination information enters through code, metadata, stimulus encoding, topology labels, experiment construction, or evaluator leakage. The ideal strict test approaches: \[ \mathrm{ALF}\rightarrow 0. \] A system that secretly performs `lookup(target_id)` has demonstrated retrieval, not condition-addressed recruitment. ## 6.7 Residual build specification For web construction: \[ L(\text{corpus}) = L(\text{shared grammar}) + \sum_i L(\text{residual}_i). \] The grammar matters only if residuals shrink against fair conventional baselines and complexity is not merely hidden in the interpreter, environment model, or adapter. --- # 7. The claim ladder Each rung must be independently earned. | ID | Candidate claim | Minimum supporting result | Primary falsifier / closure | |---|---|---|---| | H0 | Coarse condition labels discard useful signal information | Raw waveform features improve held-out prediction over condition labels | No reliable improvement | | H1 | Formed structure transforms perturbations into location/state-dependent signals | Same input produces predictable distributed waveform differences through measured structure | Structure adds no predictive information | | H2 | Current susceptibility changes effective routing | Matched input + matched topology but different measured state gives different recruitment | State adds no predictive information | | H3 | Formation changes future routing | A formation event changes later response to the same input | Difference vanishes when present physical changes are included | | H4 | History contains residual predictive value beyond declared present state | \(P(Y|S,F,H)\) beats \(P(Y|S,F)\) on held-out data | Snapshot closure | | H5 | Selective recruitment can occur without explicit destination addressing | Low-ALF stimulus recruits the appropriate prior pathway | Hidden target/address leakage explains effect | | H6 | A compact grammar predicts held-out construction/response | Grammar + residual beats fair baselines without hidden complexity | No compression/prediction advantage | | H7 | The same operational relation recurs across substrates | Frozen assay succeeds in multiple mechanistically distinct systems | Effect is substrate-specific | | H8 | An engineered history-responsive medium can exploit the relation usefully | Controlled simulator/remote hardware/physical implementation | No robust benefit over conventional routing | | H9 | FFC can use the grammar as an engineered field mechanism | Structured fields recruit formed susceptibility in preregistered tests | Ordinary addressing or static resonance explains all results | A positive result at H3 is **not** H4. A positive result at H4 is **not** H5. A positive result at H5 in spiders is **not** H7. H7 still would not establish a universal law. --- # 8. Blueprint for success ## Phase 0 — Freeze the program before mining outcomes Create a canonical research manifest containing: - hypothesis version; - exact variable definitions; - source registry; - prior exposure audit; - known literature outcomes already seen; - dataset licenses; - raw-data hashes; - train/validation/test split rules; - baseline models; - metrics; - falsifiers; - multiple-comparison plan; - stopping rules; - promotion criteria. The program should explicitly distinguish: **DISCOVERY DATA** — may be inspected while designing features and adapters. **ADJUDICATION DATA** — held out from prediction-making to the extent practically possible. Dataset virginity is not required. **Prediction blindness is.** --- ## Phase 1 — Recover the spider's construction grammar ### Primary source Johns Hopkins 2021 *Distinct movement patterns generate stages of spider web-building* archive, containing raw web-building video plus tracking models and derived embeddings. ### Goal Predict the next construction action from progressively richer information. Represent the growing web as: \[ G_t=(V_t,E_t) \] and each construction action as: \[ a_t = ( \text{position}, \text{local topology}, \text{line class}, \text{movement}, \text{attachment}, \text{stage}, \text{recent history} ). \] ### Model ladder \[ M_0: P(a_{t+1}\mid \text{coarse construction stage}) \] \[ M_1: P(a_{t+1}\mid G_t) \] \[ M_2: P(a_{t+1}\mid \text{local }G_t,S_t) \] \[ M_3: P(a_{t+1}\mid \text{local }G_t,S_t,H_t) \] \[ M_4: \text{strong flexible black-box predictor} \] The black-box model is important. A compact grammar earns interest only if it remains competitive with a serious predictive baseline. ### Split discipline Hold out at the highest practical level: - entire webs; - preferably entire spiders; - construction sessions; - altered geometry classes where possible. Do not randomly split adjacent frames from the same construction into train and test. ### Success questions 1. Does local state predict next construction better than coarse stage labels? 2. Does recent history improve prediction after current geometry is included? 3. Can a small grammar recover most of the predictive performance of a much larger model? 4. Does the grammar generalize to unusual or damaged geometry? 5. Which information is genuinely residual? ### Crucial interpretation If history ceases to help once enough present-state information is added, record **snapshot closure**, not failure. --- ## Phase 2 — Spider vibration as condition-addressed routing ### Primary sources 1. Johns Hopkins 2026 *Dynamic vibration-driven feedback shapes predator–prey interactions in orb-weaving spiders*: synchronized top/side video with natural-frequency, acoustic playback, and piezoelectric controls. 2. Dryad 2019 *Decoding the locational information in the orb web vibrations...*: finite-element models examining how longitudinal/transverse wave amplitudes at the spider's contact points carry source-location information. ### Goal Ask whether the actual waveform and formed web carry predictive information that is lost when the experiment is reduced to a categorical condition. ### Extract perturbation features At minimum: - amplitude; - dominant frequencies; - full spectral density; - phase relationships; - onset slope; - duration; - envelope; - impulse structure; - harmonics; - directionality where available; - source location where available. ### Extract response features - reaction latency; - first movement direction; - leg/contact-point response where measurable; - path toward source; - attack/capture transition; - abort/no-response; - temporal sequence of movement; - distributed web response if sensor data exist. ### Model ladder \[ R\sim \text{original condition label} \] \[ R\sim F(t) \] \[ R\sim F(t)+S_t \] \[ R\sim F(t)+S_t+H_t. \] A useful first result is simply: \[ \text{waveform model} > \text{condition-label model}. \] That would show that a legacy experiment compressed away information useful to our question. --- ## Phase 3 — Build the minimal mechanical web simulator Do not begin with a full biological model. Start with the smallest falsifiable propagation medium: - nodes; - tensioned edges; - edge length; - mass density; - stiffness; - damping; - boundary constraints; - junction coupling. Use a simple mass-spring/string network first. Move to finite-element analysis only when the simpler model fails for a known reason. For each source perturbation \(F\): 1. record response at every node/contact point; 2. derive amplitude and phase transfer; 3. identify recruited nodes; 4. alter one structural element; 5. replay the identical perturbation; 6. quantify routing divergence. Key test: \[ F_{\text{same}} + G_A \rightarrow R_A \] versus \[ F_{\text{same}} + G_B \rightarrow R_B. \] Then make the medium adaptive: \[ w_{ij}(t+1)= f(w_{ij}(t),R_{ij}(t),H_{ij}(t)). \] Now test whether prior traffic/formation changes later routing without an explicit target address. ### Simulation options - custom Python/JAX/PyTorch physical network; - differentiable graph dynamics; - local finite-element tools; - SimScale Community for selected structural simulation checks. Simulation is not evidence about spiders unless the model is validated against spider data. It is evidence about the mechanism implemented by the simulator. --- ## Phase 4 — Legacy experimental reanalysis across creatures The key opportunity is that earlier investigators often asked: > Does condition A differ from condition B? We ask: > What structure in the perturbation, present state, and history predicts the response? Use a common ontology across species: \[ (F,G,S,H,R) \] while keeping native variables and native baselines. ### Tier A — weakly electric fish The Dryad archive for *Peripheral sensory coding through oscillatory synchrony in weakly electric fish* includes receptor recordings, behavioral playback data, and electric-organ-discharge waveforms. Questions: - Which waveform features predict receptor or behavioral response? - Do similar waveforms produce different responses under different organism states? - Do preceding pulse sequences change later susceptibility? - Can the response be predicted from condition label alone, or does detailed waveform structure matter? ### Tier B — Drosophila electrosensation The 2025 Dryad archive for *Sensation of electric fields in Drosophila melanogaster* includes behavioral data, calcium imaging, measured stimulus voltages, and simulations. Questions: - Does exact stimulus trajectory outperform nominal voltage/category? - Can orientation changes be treated as perturbation-route changes? - Does recent stimulation alter later response beyond measured present state? - Can neural and behavioral response be linked through the same input representation? ### Tier C — mouse cortical stimulation DANDI:000458 contains simultaneous EEG, extracellular electrophysiology, and cortical electrical stimulation in head-fixed mice. NWB is especially useful because it is designed to preserve stimulus data with neurophysiology and behavior. Questions: - How does identical or matched stimulation propagate across recorded sites? - How do state changes alter recruitment? - Which features predict latency, phase, spectral transfer, and recruitment order? - Does recent stimulation/history improve prediction after current neural state is modeled? ### Expansion pool Later source discovery should include: - mechanosensation; - acoustic playbacks; - electrosensation; - magnetic-field responses; - optogenetic pulse trains; - peripheral nerve stimulation; - sensory entrainment; - vibration ecology; - locomotor perturbation; - sleep/anesthesia transitions; - regeneration/repair experiments. Do not add a dataset because it sounds interesting. Add it only if it exposes at least two of \(F,G,S,H,R\) with enough resolution to test a declared hypothesis. --- ## Phase 5 — Reconstruct pathways, not only response magnitude For multichannel neural or mechanical recordings: \[ F(t) \rightarrow R_1(t),R_2(t),\ldots,R_n(t). \] Measure: - latency; - phase lag; - spectral transfer; - coherence; - attenuation/amplification; - transient propagation; - recruitment order; - transient coalition structure; - directionality where identifiable. Infer an **effective propagation graph** for each trial/window. Then compare: \[ G_{\text{structural}} \] with: \[ G_{\text{effective}}(F,S,H,t). \] This directly tests the distinction: \[ \boxed{ \text{physical connectivity} \neq \text{currently admissible propagation} } \] The effective pathway may change while gross anatomy does not. --- ## Phase 6 — Detailed neuronal morphology simulation Use Arbor or a comparable cable-equation simulator with real or validated neuronal morphologies. The purpose is not to claim a neuron is a spiderweb. It is to ask whether the same operational relation occurs in a different physical model. Test: \[ \text{same input waveform} + \text{different dendritic path} \rightarrow \text{different transformed condition} \] and: \[ \text{same morphology} + \text{different membrane/channel state} \rightarrow \text{different recruitment}. \] Where realistic mechanisms are available, include: - membrane capacitance/resistance; - branch diameter; - ion-channel distribution; - active dendritic conductances; - synapse location; - recent activity; - back-propagating spikes. The important quantity is not "does the neuron spike?" alone. Preserve the waveform transformation along the route. --- ## Phase 7 — Population/network simulation Use NEST or an equivalent simulator to move from morphology to recruitment across large spiking networks. Implement explicit controls for: - static connectivity; - state-dependent excitability; - synaptic plasticity; - structural plasticity where available; - phase/timing; - multiple stimulus patterns; - recent-history manipulation. Run strict CRR-style hidden-address tests. A useful network must not know the desired target through: - population IDs embedded in the stimulus; - a global lookup table; - evaluator-generated route hints; - training labels available at test time; - direct calls to the target population. The stimulus should create a condition. The formed network should determine what responds. --- ## Phase 8 — Remote neuromorphic adjudication Before seeking a biological laboratory, move frozen assays onto remotely accessible physical hardware. EBRAINS currently provides browser/Jupyter access to BrainScaleS and SpiNNaker for testing and accepted research projects, with PyNN-described networks and user-supplied inputs. Use this as a cross-substrate test: \[ \text{same conceptual protocol} \rightarrow \text{software simulator} \rightarrow \text{digital neuromorphic hardware} \rightarrow \text{analog/mixed-signal neuromorphic hardware}. \] Do **not** require numerically identical outputs. Freeze the operational success criterion instead, such as: > Two formation histories receive the same later stimulus. Does prior formation alter selective recruitment under the same declared present-state controls? If an effect exists only in one simulator, that is information. If it survives multiple substrates, "simulator quirk" becomes a weaker explanation. --- ## Phase 9 — Tabletop synthetic web A formal laboratory is still unnecessary for the first physical mechanics test. Construct a deliberately simple synthetic network: - tensioned thread/filament; - measured geometry; - piezo actuator; - contact piezos or accelerometers; - waveform generator; - multichannel acquisition; - controlled edge tension; - removable/reconfigurable links. Protocol: 1. build \(G_A\); 2. inject frozen \(F_1,F_2,\ldots\); 3. record all channels; 4. remove/change one link or tension; 5. construct \(G_B\); 6. replay identical waveforms; 7. compare transfer and recruitment; 8. perform repair with a non-identical geometry; 9. test whether function can recover without geometry being restored exactly. This probes a key engineering idea: > **Repair the grammar's ability to continue, not necessarily the original geometry.** Later, introduce history-bearing elements such as tunable, hysteretic, memristive, magnetic, mechanical, or other adaptive junctions — but only after the passive system is understood. --- ## Phase 10 — Biological laboratory only when it can adjudicate A biological experiment becomes worth the cost when we can arrive with: - frozen code; - source-controlled simulator; - legacy-data results; - remote-hardware results where possible; - a preregistered perturbation; - predicted waveform transformation; - predicted recruited region/behavior; - explicit falsifiers; - no tuning allowed after reveal. The laboratory's job should be: \[ \boxed{ \text{adjudicate a prediction, not help us invent one} } \] A strong spider experiment would be something like: > After a declared radial/tension manipulation, frozen waveform \(F_3\), but not matched waveform \(F_1\), should alter the relative response at specified contact points and produce a predicted change in orientation latency. The exact prediction must come from prior modeling/data, not be written after observing the spider. --- ## Phase 11 — Engineering return to FFC Only after the lower-level grammar survives should FFC inherit it. The engineering question then becomes: > Can a purpose-built substrate have a susceptibility landscape that is persistently altered by formation history, so that a later structured field selectively recruits useful state without explicit global addressing? Possible future substrate families include: - memristive networks; - spintronic / nanomagnetic arrays; - magnonic systems; - photonic resonator networks; - coupled oscillator arrays; - adaptive mechanical metamaterials; - other history-responsive materials. These are candidates, not current claims. The FFC leap is no longer: > Can a field make selected matter respond? Ordinary resonance already demonstrates condition-selective response. The hard leap is: \[ \boxed{ \text{Can formation write a sufficiently rich susceptibility landscape for later structured perturbations to perform useful recruitment?} } \] --- # 9. Analysis grammar shared across datasets To prevent each substrate from becoming an unrelated bespoke project, use the same escalating model family wherever possible. ## Model 0 — Original label baseline \[ R \sim C \] where \(C\) is the experimenter's condition label. ## Model 1 — Perturbation model \[ R \sim F \] Does raw/derived waveform structure contain information the original label discarded? ## Model 2 — Present-state model \[ R \sim F+S+G \] Does current state/susceptibility improve prediction? ## Model 3 — History model \[ R \sim F+S+G+H \] Does prior formation improve held-out prediction? ## Model 4 — Snapshot-enrichment sequence Progressively expand \(S\) until history closes or remains unresolved. ## Model 5 — Native-domain strong baseline Use the strongest serious model natural to the source field. Do not ask FPG to beat a strawman. ## Model 6 — Black-box ceiling Use a flexible predictor to estimate how much predictable signal is available in the data at all. The compact grammar is interesting if it approaches the black-box ceiling while using substantially less information or complexity. --- # 10. The crucial causal tests Prediction alone is not enough. ## 10.1 History ablation Hold present measured state as closely matched as possible while varying formation history. ## 10.2 Structural ablation Change one edge, junction, tension, or connection while replaying identical input. ## 10.3 Waveform ablation Hold source/location constant and vary one waveform dimension at a time. ## 10.4 Receiver-state ablation Hold input and gross structure constant while changing susceptibility. ## 10.5 Hidden-address audit Remove or randomize labels that could leak the intended destination. ## 10.6 Repair assay Damage the network, then allow a different geometry to restore function. Test functional equivalence rather than geometric identity. ## 10.7 Cross-substrate replay Freeze an operational protocol and repeat it in another implementation. ## 10.8 Held-out perturbation prediction Withhold specific waveform/geometry/history combinations until after the model and prediction are frozen. --- # 11. What "evidence" means here A result is evidence for a hypothesis when it is more expected under that hypothesis than under the declared alternatives. Therefore: > "Not evidence for universality" does **not** mean "not evidence." A spider result can be evidence for a CRR-like or FPG-like mechanism **in that spider/web system** while remaining insufficient to claim universality. Use the standing epistemic ledger: ### OBSERVED Directly measured, recorded, imaged, or detected. ### DERIVED Mathematical consequence of observations and declared assumptions. ### STRONGLY INFERRED Multiple independent observations support it, but the relevant historical event or mechanism was not directly witnessed. ### MODEL-DEPENDENT Follows within a specified simulator/model family. ### CONSISTENT WITH Compatible with the hypothesis but not discriminating. ### SPECULATIVE Plausible candidate without decisive evidence. ### UNKNOWN Evidence does not adjudicate. Rule: > **Do not promote inference into knowledge merely because the inference is conventional or appealing.** --- # 12. Anti-self-deception gates This program is unusually vulnerable to beautiful metaphors. Build the defenses in now. ## Gate A — No spiderweb mysticism Do not treat geometric resemblance as mechanism. ## Gate B — No universality promotion Cross-substrate recurrence must be demonstrated, not assumed. ## Gate C — No hidden complexity A "small grammar" cannot hide its real complexity in an adapter, global map, learned embedding, environment oracle, or lookup structure. ## Gate D — No hidden address Conditioned recruitment must not smuggle target identity into the stimulus or controller. ## Gate E — No history double-counting If history is already encoded in present measured state, do not count it again as independent history. ## Gate F — No post hoc waveform hunting Freeze feature families or use nested discovery/adjudication splits. ## Gate G — No organism leakage Avoid train/test splits that allow identity/session-specific signatures to masquerade as general grammar. ## Gate H — No baseline games Use serious native-domain baselines and a flexible predictive ceiling. ## Gate I — No consciousness promotion Nothing in FPG establishes phenomenal consciousness, sentience, or personhood. ## Gate J — No mechanism substitution A result in a mechanical web is not evidence that neurons use the same physical mechanism. It establishes an operational relation in that substrate. ## Gate K — Preserve negative results A null, closure, failed transfer, or broken generalization is a result and remains in the permanent ledger. --- # 13. What would count as a successful program? There are multiple legitimate success endpoints. ## Success Level 1 — Grammar extraction A compact local/history-aware grammar predicts held-out spider construction materially better than geometry/stage baselines. ## Success Level 2 — Information recovery Raw perturbation structure predicts responses better than the original experimenter's condition labels. ## Success Level 3 — Susceptibility The same perturbation produces different predictable outcomes under different measured present states. ## Success Level 4 — Formative consequence A controlled formation event changes later routing/recruitment. ## Success Level 5A — Snapshot closure The history effect disappears after sufficient present-state measurement. This successfully localizes the formative residue. ## Success Level 5B — Residual sequence divergence History remains predictive after a serious, preregistered snapshot-enrichment process. This creates a stronger open problem. ## Success Level 6 — Strict conditioned recruitment The appropriate prior pathway/state is selectively recruited with low address leakage and no explicit destination lookup. ## Success Level 7 — Cross-substrate recurrence The same operational relation survives at least two mechanistically distinct substrates under frozen criteria. ## Success Level 8 — Physical hardware The relation survives transfer to remotely accessed or tabletop physical hardware. ## Success Level 9 — Novel prediction The program predicts an unrevealed biological or physical outcome prospectively and beats declared baselines. ## Success Level 10 — FFC engineering An engineered history-responsive substrate uses structured perturbations/fields to perform useful selective recruitment. The program does **not** need to reach Level 10 to be scientifically successful. --- # 14. Failure conditions that would improve the theory The following are valuable outcomes: - current state fully screens off history; - waveform detail adds nothing over condition labels; - a compact grammar fails against a native baseline; - spider construction requires global state rather than local grammar; - condition-addressed retrieval collapses into explicit addressing; - cross-species recurrence fails; - neuromorphic hardware does not preserve simulator behavior; - passive mechanics explain the apparent history effect; - repair restores only geometry, not function; - the effect is unique to one biological niche; - FFC adds no engineering value over conventional routing/resonance. Each failure removes a degree of freedom and makes the surviving theory smaller. --- # 15. Networking implications The conventional abstraction is often: \[ \text{nodes compute; links transport}. \] FPG asks whether the interconnect itself can participate in computation: \[ \boxed{ \text{interconnect} = \text{state} + \text{filter} + \text{router} + \text{memory substrate} } \] Candidate architecture: \[ e_{ij} = ( \text{gain}, \text{delay}, \text{phase}, \text{bandwidth}, \text{trust}, \text{recent use}, \text{formation state} ). \] A signal then does not need a complete destination address. Its structure interacts with the formed network to determine which routes remain effective. This suggests: - condition-addressed multicast; - history-dependent routing; - adaptive propagation; - self-healing networks; - function-preserving rerouting without geometric restoration; - low-overhead retrieval from distributed state; - potentially new security surfaces and new security controls. These are downstream engineering hypotheses. --- # 16. Chip-architecture implications A conventional chip strongly separates: \[ \text{memory} \leftrightarrow \text{interconnect} \leftrightarrow \text{compute}. \] FPG suggests a candidate substrate where those roles partially merge. A junction may carry persistent state. An edge may transform a waveform. A route may be recruited because of current susceptibility. A response may alter later susceptibility. The core device relation is: \[ r_i(t) = G\left( s_i, H_i, \{w_{ij}\}, F(t) \right). \] Then: \[ (s_i,w_{ij})_t \rightarrow (s_i,w_{ij})_{t+1}. \] This is not yet a chip design. It is a requirement for any chip architecture claiming to implement the grammar. The most important hardware question is: > **Can the substrate compute part of the address while the perturbation travels?** --- # 17. Neural implication Do not reduce neurons to wires. A morphologically detailed neuron contains branching cable structure, capacitance, resistance, active conductances, synapses, and state-dependent nonlinearities. At population scale, oscillations and traveling activity can modify phase and excitability. The candidate relation is: \[ \boxed{ \text{perturbation} \rightarrow \text{propagation through formed morphology} \rightarrow \text{waveform transformation} \rightarrow \text{state-dependent recruitment} } \] The research question is not: > Are brainwaves the answer? It is: > **Do oscillatory/traveling dynamics participate in determining which otherwise available neural pathways become functionally admissible?** A still stronger CRR-compatible question is: > Can a small present perturbation selectively reactivate a much larger prior formation because the formed network itself supplies the route? If so: \[ \text{small perturbation} + \text{large formed history} \rightarrow \text{selective reactivation}. \] This is a hypothesis. Existing neuroscience provides relevant mechanisms and data, not validation of the full claim. --- # 18. The waveform alphabet Do not prematurely reduce the signal to frequency. A useful perturbation representation may include: \[ F = ( \omega, A, \phi, \Delta t, \text{envelope}, \text{direction}, \text{polarization}, \text{harmonics}, \text{chirp}, \text{cross-frequency relation}, \text{temporal sequence} ). \] For mechanical webs, not all coordinates apply. For neuronal/electrical systems, others will be added. For FFC, polarization, spatial gradient, propagation direction, standing/traveling topology, and interference may become central. The grammar should preserve a common abstract interface while allowing substrate-native signal dimensions. --- # 19. Research data architecture Every source should enter through a provenance-bearing adapter. Recommended canonical object: ```yaml dataset_id: source_url: doi_or_persistent_id: license: downloaded_at: raw_hashes: original_question: original_conditions: available_raw_stimulus: available_raw_response: available_structure: available_state: available_history: known_outcomes_seen_before_preregistration: adapter_version: feature_extractor_version: split_manifest: held_out_assets: ``` Every derived artifact should retain: - source dataset; - source file hash; - code commit; - environment; - feature version; - split; - random seed; - model hash; - prediction timestamp; - result timestamp. This is a natural use case for the R2R evidence/provenance engine. --- # 20. Legacy-data preregistration protocol Historical datasets cannot be made literally new, but they can still adjudicate new predictions. For each dataset: 1. Publish prior exposure audit. 2. Separate literature knowledge from raw-outcome knowledge. 3. Define \(F,G,S,H,R\) before accessing held-out outcome segments. 4. Freeze feature families. 5. Freeze native baseline. 6. Freeze scoring metric. 7. Freeze tolerance/effect threshold. 8. Freeze held-out split. 9. Freeze falsifiers. 10. Hash preregistration. 11. Run once. 12. Preserve result whether positive, null, or contradictory. The key standard is: > **Evidence capable of adjudicating the registered prediction must remain unavailable to the prediction-making process.** --- # 21. First experiment to execute ### FPG-EXP-001 — Spider Construction Grammar / History Increment **Question** Does recent formation history improve prediction of the next web-building action after current observable structure is included? **Source** Johns Hopkins 2021 web-building archive. **Primary comparison** \[ M_S: P(a_{t+1}\mid S_t) \] versus \[ M_H: P(a_{t+1}\mid S_t,H_t). \] **State \(S_t\)** Begin with: - spider position/orientation; - current local web graph; - nearest junction distances; - local line angles; - construction stage; - local attachment opportunities. **History \(H_t\)** Begin with: - last \(k\) movement/action classes; - immediately previous attachment; - travel direction; - time since last construction event; - recent path through the web. **Held-out unit** Prefer whole spider or whole construction session. **Success** Predeclared improvement in held-out log loss / action prediction, with effect stable across spiders/sessions. **Closure path** Progressively enrich \(S_t\). If history advantage disappears, record the minimal state required for closure. **Strong extension** Predict continuation after interruption/damage or unusual geometry. **Do not claim** - universal grammar; - neural equivalence; - CRR; - FFC validation. This experiment exists to establish whether the construction-history question deserves the next rung. --- # 22. Second experiment to execute ### FPG-EXP-002 — Web Perturbation Information Recovery **Question** Does the measured/derived waveform predict spider response better than the original categorical experimental condition? **Source** Johns Hopkins 2026 vibration/playback archive. **Comparison** \[ R \sim C \] versus \[ R \sim F. \] Then: \[ R \sim F+S \] and, only if justified: \[ R \sim F+S+H. \] **First target** Reaction latency and first movement/orientation. **Success** Waveform features add stable held-out predictive information beyond condition identity. **Failure** Condition identity is sufficient within measurement noise. Either result constrains the program. --- # 23. Third experiment to execute ### FPG-EXP-003 — Synthetic Route Divergence **Question** Does a controlled structural change alter the distributed response to an identical perturbation in a minimal mechanical network? This should first be simulated, then reproduced on a tabletop network. **Prediction** \[ D_{\mathrm{route}}(F,G_A,G_B) > 0 \] for preregistered structural perturbations. This is expected under ordinary mechanics; the purpose is calibration. The stronger follow-up is whether a history-dependent adaptive element can create selective later recruitment **without explicit addressing**. --- # 24. Cross-project ripple map If FPG survives, it can sharpen multiple existing programs. ### RAdT Provides a concrete mechanism for history-conditioned admissibility. ### FSS Provides a measurable substrate for formative residue and return pathways. ### Ageometrics Provides matched-state/history assays in physical and biological systems. ### CRR Provides a candidate physical explanation for condition-sensitive return paths. ### BSR Suggests how a small condition might recruit the Minimum Sufficient Present rather than globally activate memory. ### FFC Provides the prerequisite propagation grammar before attempting fieldborne hardware. ### BECS / NeuroSignal Adds history-dependent susceptibility to signal-fidelity analysis. ### Baby AI / FormationCore Suggests future routing/activation mechanisms where persistent consequence changes accessibility without changing evidentiary authority. ### Cognitive Basin Offers a possible physical/algorithmic interpretation of selective activation from a large persistent possibility space. ### Networking Suggests condition-addressed, history-responsive routing. ### Chip architecture Suggests compute/memory/interconnect co-location in formed propagation media. ### Evolution Offers a later test of whether the grammar "formation changes what remains possible next" has useful explanatory power at a radically different scale. No cross-project promotion should occur automatically. Each connection must earn its own evidence. --- # 25. Public research posture If exposed through the Public Formation Stream, publish: - the hypothesis before adjudication; - the preregistration; - nulls; - snapshot closures; - failed transfers; - corrections; - successful predictions; - source hashes; - code; - exact claim status. Do not post: > "We discovered the universal grammar of nature." Post: > "We registered a test of whether history improves next-action prediction in spider-web construction after present geometry is controlled. Here is the frozen protocol." The work should emit its own formation history. --- # 26. Working maxims These are heuristics, not scientific claims. > **The address may be a condition rather than a location.** > **Formation can become its own future address — but an address is navigation, not evidence.** > **Create the condition. Let admissible structure respond. Preserve what changes the future.** > **Do not chase the completed morphology. Watch formation.** > **Do not ask only what moved. Ask what changed the available continuations.** > **The interconnect may be part of the computation.** > **Repair the ability to continue, not necessarily the old geometry.** > **Make the laboratory adjudicate, not invent, the hypothesis.** > **When chasing a gremlin, follow the broken shit.** --- # 27. Immediate execution order 1. Freeze this v0.1 blueprint. 2. Create the FPG folder beneath FFC. 3. Create a source registry with hashes, licenses, sizes, and known-exposure notes. 4. Ingest the Johns Hopkins 2021 spider construction dataset metadata and code first. 5. Write FPG-EXP-001 preregistration before exploratory outcome mining. 6. Build the canonical \(F,G,S,H,R\) schema and dataset adapter interface. 7. Replicate one published analysis from the source dataset to validate ingestion. 8. Build the \(M_0\)–\(M_4\) model ladder. 9. Run held-out spider/session evaluation. 10. Perform snapshot-enrichment until closure or residual divergence. 11. Freeze result. 12. Move to JHU 2026 vibration/playback data. 13. Run waveform-vs-label comparison. 14. Build the minimal mechanical web simulator. 15. Validate it against published/legacy web-vibration measurements. 16. Expand to weakly electric fish and Drosophila. 17. Add DANDI:000458 cortical-stimulation reanalysis. 18. Reconstruct effective propagation pathways, not only scalar response. 19. Move the frozen assay to Arbor/NEST. 20. Move compatible network assays to EBRAINS BrainScaleS/SpiNNaker. 21. Build the synthetic tabletop web only after the simulator has a frozen prediction. 22. Approach a biological lab only after we can hand it a specific falsifiable result to adjudicate. 23. Promote surviving mechanism into FFC engineering. 24. Preserve every failure and closure. --- # 28. Definition of overall success The strongest defensible success target for this program is: \[ \boxed{ \begin{array}{l} \text{A compact, preregistered grammar predicts how structured perturbations}\\ \text{are transformed into selective recruitment by formed media;}\\ \text{formation changes future susceptibility in controlled tests;}\\ \text{history is either localized into measurable present state or shown to add}\\ \text{residual predictive value under matched-state assays;}\\ \text{strict recruitment survives an address-leakage audit;}\\ \text{and at least part of the relation transfers across distinct substrates.} \end{array} } \] Engineering success would then be: \[ \boxed{ \text{use that grammar deliberately in a history-responsive physical substrate} } \] which is where FPG hands the problem back upward to Formative Field Computing. --- # 29. Verified starting resources Verified 2026-08-21. ### Spider construction **Corver et al. (2021), Johns Hopkins Research Data Repository** *Distinct movement patterns generate stages of spider web-building* Raw videos, LEAP/DeepLabCut models, and embeddings. DOI: https://doi.org/10.7281/T1/BMATHH Repository: https://archive.data.jhu.edu/dataset.xhtml?persistentId=doi:10.7281/T1/BMATHH ### Spider vibration / behavior **Hung, Corver & Gordus (2026), Johns Hopkins Research Data Repository** *Dynamic vibration-driven feedback shapes predator–prey interactions in orb-weaving spiders* Raw synchronized video across natural-frequency, acoustic playback, and piezoelectric control conditions. DOI: https://doi.org/10.7281/T1TX91CG Repository: https://archive.data.jhu.edu/dataset.xhtml?persistentId=doi:10.7281/T1TX91CG ### Spider vibration / source localization **Mortimer et al. (2019), Dryad** *Decoding the locational information in the orb web vibrations of Araneus diadematus and Zygiella x-notata* Finite-element web vibration data and locational-information analysis. DOI: https://doi.org/10.5061/dryad.7rm572m ### Weakly electric fish **Baker, Huck & Carlson (2016), Dryad** *Peripheral sensory coding through oscillatory synchrony in weakly electric fish* Receptor recordings, behavioral playback, and EOD waveforms. DOI: https://doi.org/10.5061/dryad.ck54h ### Drosophila electrosensation **Tadres et al. (2025), Dryad** *Sensation of electric fields in Drosophila melanogaster* Behavior, calcium imaging, stimulus voltage measurements, and simulations. DOI: https://doi.org/10.5061/dryad.5mkkwh7d8 ### Mouse cortical stimulation **DANDI:000458** *Simultaneous electroencephalography, extracellular electrophysiology, and cortical electrical stimulation in head-fixed mice* NWB neurophysiology dataset. https://dandiarchive.org/dandiset/000458 **NWB / DANDI standard** NWB is designed to preserve electrophysiology, optical physiology, tracking, and stimulus data. https://docs.dandiarchive.org/getting-started/data-standards/nwb/ ### Detailed neuron simulation **Arbor** Morphologically detailed cable-cell simulation; neuronal morphology represented as branched electrical cables with capacitance/resistance and related dynamics. https://docs.arbor-sim.org/ ### Network simulation **NEST Simulator** Large neural-network simulation with neuron, synapse, stimulation, recording, and plasticity models. https://nest-simulator.readthedocs.io/ ### Remote neuromorphic hardware **EBRAINS Neuromorphic Computing Platform** BrainScaleS and SpiNNaker access for testing/accepted research, with browser/Jupyter and PyNN workflows. https://wiki.ebrains.eu/bin/view/Collabs/neuromorphic/Getting%20access/ ### Cloud structural simulation **SimScale Community** Free testing/learning tier with selected analysis types and current simulation quota. https://www.simscale.com/product/pricing/ --- # 30. Final boundary statement The program begins with a provocative unification: \[ \boxed{ \textbf{structure determines susceptibility; condition encounters susceptibility; matching produces recruitment; recruitment changes structure.} } \] Everything after that must be earned. Spiderwebs are not proof of neurons. Neurons are not proof of FFC. Simulation is not proof of biology. Cross-substrate recurrence is not proof of universality. A return path is not evidence. A dream is not evidence. But each can supply a disciplined test of the next rung. The practical strategy is therefore: \[ \boxed{ \textbf{ legacy data \rightarrow grammar \rightarrow simulation \rightarrow cross-species reanalysis \rightarrow neural models \rightarrow remote hardware \rightarrow tabletop physics \rightarrow prospective biology \rightarrow FFC engineering } } \] And the program should stop, branch, or close wherever the evidence tells it to.