============================================================================== MVPS -- Unified State Space (Proposal) A meta-companion that articulates, in precise terms, why the observatory profile, the data-plane profile, and the kernel profile produce the same coherence algebra. Leonardo Melegassi Catellix Research Version: 0.1 (proposal -- the formal claim, not the proof) Date: 2026-05-21 Status: meta-companion to MVPS_THREE_LAYER_MATHEMATICAL_EVIDENCE.txt v1.1 MVPS_DATAPLANE_PROFILE.txt v0.1 MVPS_KERNEL_PROFILE.txt v0.1 ============================================================================== Abstract. The MVPS framework has, by this point, three sibling profiles: - The observatory profile defined in MVPS_THREE_LAYER_MATHEMATICAL_EVIDENCE.txt v1.1. Vantage type: external network prober. - The data-plane profile defined in MVPS_DATAPLANE_PROFILE.txt v0.1. Vantage type: forwarding-silicon next-hop / queue / port. - The kernel profile defined in MVPS_KERNEL_PROFILE.txt v0.1. Vantage type: in-kernel subsystem observer (eBPF / tracepoint). Each profile defines a different vantage type V_*, composes a bundle B_*(t) = (V_*)^N over N >= 2 vantages, computes the same three-axis coherence (C_1, C_2, C_3), the same operational Hamiltonian H, the same Mahalanobis phase distance Phi_D, and the same operational phase label Phi_K. The constructions of the three axes differ in *implementation substrate* -- a JSON object on a controller, a register array on a Tofino, a BPF map in a Linux kernel -- but produce, by design, the same input-output signatures. The central claim of this document is that this is not a coincidence. There is an underlying *operational state space* such that the network surface, the silicon surface, and the kernel surface are each a faithful projection of the same dynamical object. The three vantage types and the three coherence values they emit are *witnesses* to the same state, observed through three different microscopes. This document does three things. (1) It states the unification claim formally, in light category theory: there is a functor C from a category of vantage types to a category of coherence spaces, and the three existing profiles are three objects in the domain category that are mapped by C to functorially equivalent objects in the codomain. (2) It enumerates the empirical predictions the claim entails -- in particular, the cross-surface predictions that can be falsified by joint observation across two or three profiles on the same host or path. (3) It catalogues the open questions whose resolution determines whether the claim is (a) a useful organising principle, (b) a precise theorem about a well-defined invariant, or (c) false in the strong form stated and true only in a weakened form to be specified. Three honest framings used throughout this document. (a) This document is the strongest claim in the entire MVPS family. It is also the most exposed to falsification: any host / silicon / kernel triple that exhibits, under controlled stress, Phi_K transitions in one surface that are *uncorrelated* with Phi_K transitions in another surface contradicts the strong form of the claim. The framework welcomes such evidence; the open-questions section (U9.x) enumerates exactly what experiments would yield it. (b) The category-theoretic vocabulary is used for precision, not for prestige. Section 3 keeps the formal machinery light and the prose self-contained for a reader without category-theory background. Readers who prefer to skip the categorical formulation may jump from Section 2 directly to Section 4 (empirical predictions) without loss of coverage. (c) The unification is, today, a *conjecture supported by construction*: it has been *designed in* by writing the three profiles to share the same axiomatic skeleton. Empirical confirmation requires the open work catalogued in MVPS_DATAPLANE_PROFILE.txt (D9.x), MVPS_KERNEL_PROFILE.txt (K9.x), and this document (U9.x). The conjecture's value is heuristic and organisational until that work lands; even so, the heuristic is already strong enough to predict cross-surface signatures and to organise the future publication and validation plan. Companion artefacts: - MVPS_THREE_LAYER_MATHEMATICAL_EVIDENCE.txt v1.1 - MVPS_DATAPLANE_PROFILE.txt v0.1 - MVPS_KERNEL_PROFILE.txt v0.1 - draft-melegassi-ippm-mvps-bundle - https://catellix.com/v11-evidence.html ============================================================================== 1. The shape of the claim, in plain language ============================================================================== Consider an operator who runs a single host on a single carrier- grade network at a single instant. There is a single "operational reality" at that instant: certain packets are in flight, certain queues are partially full, certain kernel data structures are being touched, certain BGP UPDATEs are propagating. This reality is a single dynamical object S(t). The observatory profile observes S(t) through external probes. The data-plane profile observes S(t) through forwarding silicon. The kernel profile observes S(t) through in-kernel hooks. Each profile sees a different *projection* of S(t). The observatory sees the path-level shadow; the data-plane sees the hop-level shadow; the kernel sees the subsystem-level shadow. The projections are not interchangeable: an Einstein-bound violation visible to the observatory is not directly visible to the kernel; a runqueue saturation visible to the kernel is not directly visible to the observatory. The claim is that, despite these projections being different, the *algebraic structure* they produce is the same. Each projection yields a vector in [0, 1]^3 (the three coherence axes), a scalar Hamiltonian, a Mahalanobis distance, and a four-valued phase label. The maps from raw observation to these algebraic outputs are different *constructions*, but they satisfy the same *axiomatic specification*. And -- this is the empirically testable part -- the four-valued phase labels emitted by the three projections on the same incident *agree*, modulo bounded lag. In other words: the three coherence values and the one phase label produced by each profile are *names for the same thing* observed in different ways. The framework is, at its core, a way to give the same name to the same thing through three different microscopes. This is the abstraction that Henri Poincare meant by "l'art de donner le meme nom a des choses differentes" except that, here, the things are *not* different at all once the abstraction is in place. They are the same operational reality, named the same way, with the irrelevant differences in observational substrate factored out. ============================================================================== 2. The vantage type as the carrier of substitution ============================================================================== To make the claim precise we have to be careful about what "the same algebra" means when the inputs are radically different. The mechanism that the three profiles share is *vantage substitution*: the bundle B(t) is parameterised by a vantage type V; the axes C_1, C_2, C_3 are defined as functions on (V)^N -> [0, 1]; if a second vantage type V' admits the same *abstract observation interfaces* that V does, then the same axis definitions apply verbatim to (V')^N. The abstract observation interfaces required for v1.1's algebra are exactly three, one per axis. We restate them here as *interface contracts* that any vantage type must satisfy to be admissible. The construction of the profiles to date demonstrates that the contracts are satisfiable in radically different substrates. --------------------------------------------------------------------------- 2.1 Interface I_1 -- a notion of pairwise causal compatibility --------------------------------------------------------------------------- Required of vantage type V: for any pair (V_a, V_b), there must exist a binary predicate compatible(V_a, V_b, t) in {0, 1} and a non-negative bound delta_ab^min in [0, infinity) such that compatible(V_a, V_b, t) = 1 iff the joint observations of V_a and V_b at tick t respect the minimum-separation bound delta_ab^min. The three existing profiles satisfy this interface as follows: - Observatory: delta_ab^min = 2 * d_ab / c_f (great-circle distance over fibre speed); compatible iff RTT_a + RTT_b >= delta_ab^min (Einstein bound). - Data-plane: delta_ab^min = 2 * d_ab / c_f computed in the same way for the geographic separation between the next-hop egress and the peer's nearest measurement point; compatible iff observed packet RTTs respect the bound. - Kernel: delta_ab^min = kernel-causality bound between coupled subsystems (number of context switches + memory barriers separating two coupled observation points, measured at kernel boot); compatible iff observed event-time gaps respect that bound. The three constructions are arithmetically different. The *interface* is identical: a Boolean per pair, a non-negative bound, and a violation count that drives C_1^Einstein. --------------------------------------------------------------------------- 2.2 Interface I_2 -- a notion of empirical distribution --------------------------------------------------------------------------- Required of vantage type V: for each vantage V_i at tick t, an empirical distribution p_i over some alphabet A_V (or, equivalently, a sketch from which such a distribution can be reconstructed in the asymptotic limit of unbounded resources). The three profiles satisfy this interface as follows: - Observatory: p_i is the empirical hop-distribution over hop IPs observed by external prober V_i in the bundle window. A_V = the set of distinct hop IPs seen. - Data-plane: p_i is a Count-Min sketch over a flow-key hash for the i-th next-hop. A_V is the (very large but bounded) hash space. - Kernel: p_i is a Count-Min sketch over per-subsystem keys (run-queue-latency buckets, IRQ vectors, allocation orders, etc.), collapsed onto a 6-bin universal alphabet {idle, light-load, contention, saturation, starvation, failure}. All three feed the same axis: C_2 = 1 - JSD_norm({p_i}). The Jensen-Shannon divergence is defined on any finite-alphabet distribution; the only profile-specific construction is the alphabet itself. --------------------------------------------------------------------------- 2.3 Interface I_3 -- a notion of touched-object set --------------------------------------------------------------------------- Required of vantage type V: for each vantage V_i at tick t, a finite set S_i of "touched objects" with a meaningful set-overlap operation. The three profiles satisfy this interface as follows: - Observatory: S_i is the directed edge set E_i of vantage i's traceroute (consecutive-hop pairs). - Data-plane: S_i is a Bloom filter populated by source IPs of return-path ICMP-TTL-exceeded messages observed on the i-th next-hop. - Kernel: S_i is a Bloom filter populated by hash-coded references to kernel data structures (per-CPU runqueues, zones, request queues, sockets, IRQ vectors) that vantage i interacted with during the tick. All three feed the same axis: C_3 = mean pairwise Jaccard(S_i, S_j). The Jaccard index is defined on any pair of finite sets; the only profile-specific construction is what the sets contain. --------------------------------------------------------------------------- 2.4 The substitution principle --------------------------------------------------------------------------- Any vantage type V'' that satisfies the three interfaces I_1, I_2, I_3 above -- in the sense of providing per-vantage implementations of each that respect the contracts -- is an admissible MVPS vantage type, and v1.1's axes C_1, C_2, C_3 apply to (V'')^N without algebraic change. The three existing profiles are three witnesses to the satisfiability of these interfaces in radically different substrates. The unification claim of this document is that any future vantage type satisfying the same three interfaces produces the same algebra and (modulo bounded lag) the same Phi_K decisions on shared incidents. ============================================================================== 3. Light category-theoretic formulation (optional) ============================================================================== The substitution principle of Section 2.4 has a natural categorical statement. Readers without category-theory background may skip directly to Section 4. Define a category Vantage as follows. Objects : equivalence classes of vantage types satisfying interfaces I_1, I_2, I_3 of Section 2. Morphisms : interface-preserving maps. A morphism f: V -> V' consists of three components (f_1, f_2, f_3), each respecting the corresponding interface contract. Define a category Coherence as follows. Objects : the unit cube [0, 1]^3 together with the four- element set {BAU, WATCH, ALARM, CRITICAL} and the structure-preserving operations (multiplication, negative-log, Mahalanobis evaluation against a specified positive-definite 3x3 matrix Sigma^{-1}, exponential decay with parameter k). Morphisms : maps that preserve the cube structure, the phase- label ordering, and commute with the negative-log / Mahalanobis structure. Coherence has one object up to the choice of Sigma^{-1} and k; those are profile-specific calibration parameters, not algebraic changes. The construction in v1.1, the data-plane profile, and the kernel profile together define a functor C : Vantage -> Coherence. On objects, C maps a vantage type V to the coherence object [0, 1]^3 x {BAU, WATCH, ALARM, CRITICAL} with the Sigma^{-1} and k appropriate to V. On morphisms, C maps an interface-preserving map f: V -> V' to the identity on the Coherence object (since there is, up to calibration, only one). The unification claim now reads: * The functor C is *constant on objects* up to calibration; equivalently, every admissible vantage type produces the same coherence structure. * The empirical claim layered on top of that algebraic claim is that the *phase labels emitted by C(V), C(V'), C(V'')* on the same dynamical object S(t) agree modulo bounded lag. The algebraic claim is true by construction (this is what it means to have written the three profiles against the same interface contracts). The empirical claim is open and falsifiable (Section 5 enumerates predictions; Section 6 enumerates experiments). Readers comfortable with the categorical vocabulary will note that the algebraic claim is, formally, the statement that C is the constant functor on a particular Coherence object up to calibration; the empirical claim is the stronger statement that, viewed as a *natural transformation* between observation functors of the same underlying state S(t), C is a natural isomorphism modulo bounded lag. The empirical claim is the one worth proving, since the algebraic claim is essentially tautological once the interfaces of Section 2 are accepted. ============================================================================== 4. Empirical predictions ============================================================================== The unification claim makes specific, testable predictions when two or three profiles are deployed on the same physical system and observe the same incident. We enumerate the five most important. P1. Phase-label agreement modulo lag. Prediction. Under a controlled stress that perturbs an operational reality observable in two profiles (e.g. a noisy- neighbour memory event observable in V_kernel and in V_dataplane via egress-queue back-pressure), the phase-label transitions Phi_K_kernel(t) and Phi_K_dataplane(t) agree up to a bounded lag tau_12 specific to the two surfaces. Falsification. If, on repeated independent runs, the two surfaces produce Phi_K transitions that are *uncorrelated* (or anti-correlated), the unification claim is false in the strong form. A weaker form -- "the two surfaces produce correlated but not aligned signals" -- may remain consistent with the data; the open work item U9.1 characterises the lag distribution required for that weakening to be defensible. P2. Per-axis attribution consistency. Prediction. When an incident is dominantly causal (e.g. a Layer-1 fibre cut that introduces a propagation-delay anomaly), the observatory and data-plane profiles both attribute the dominant Phi_D contribution to C_1, not to C_2 or C_3. When an incident is dominantly informational (e.g. a flow-distribution shift caused by a load-balancer change), both profiles attribute the dominant contribution to C_2. When the incident is topological (e.g. an ECMP rebalance), both attribute to C_3. Falsification. If, across a curated incident catalogue, the two surfaces consistently disagree on which axis dominates -- e.g. the observatory blames C_3 while the data-plane blames C_1 -- the unification claim is, again, false in the strong form. P3. Cross-surface anticipation. Prediction. Some incidents originate at one surface and propagate to others with measurable lag. A memory-pressure event originates in V_mm (kernel surface), propagates to V_blk (kernel), back-pressures V_sock (kernel), and only then degrades V_dataplane / V_network. The unification claim predicts that the *kernel* Phi_K transition precedes the *data-plane* and *observatory* Phi_K transitions on such incidents, and that the lag tau_kernel->dp is shorter than the lag tau_kernel->observatory (because the data-plane is observationally closer to the kernel than the observatory is). Falsification. If the lags do not respect the predicted ordering, the operational-locality argument that motivates the unification claim is weakened. The framework would remain useful as three independent profiles, but the unification would not constitute a genuine causal model of state propagation. P4. Calibration transfer (weak form). Prediction. A Sigma^{-1} calibrated on one workload class in the observatory profile predicts, qualitatively, the Sigma^{-1} that will be calibrated on the same workload class in the data-plane or kernel profile. Quantitatively the matrices differ; structurally (e.g. which off-diagonal entries are large) they agree. Falsification. If the Sigma^{-1} matrices have no recognisable structural overlap across surfaces, the unification claim is empirically empty: each surface might as well be calibrated independently and used as an independent monitor. P5. Joint phase-label improves precision. Prediction. The conjunction Phi_K_kernel = ALARM ^ Phi_K_dataplane = ALARM (joint ALARM across two surfaces) has higher precision against ground-truth incidents than either alone, at modest recall cost. This is the *dual- confirmation* property that any honest multi-surface framework should exhibit. Falsification. If joint ALARM does not improve precision over the better of the two single-surface alarms, the unification claim is operationally inert: there is no additional information in the joint observation. Each surface, in that case, is a complete monitor in itself. The five predictions are not independent (P3 implies a constraint that P1 must respect; P4 is a precondition for P5 to be testable in a reusable way). But each is independently falsifiable, and together they describe a *minimal experimental contract* the unification claim must satisfy to be more than ornamental. ============================================================================== 5. What the unification does NOT claim ============================================================================== The negative space is as important as the positive one. The unification claim explicitly does not say: (a) That the three surfaces are *physically equivalent*. They are not. A packet RTT is not a runqueue wait is not an ECMP-selection counter. They are different physical things observed in different ways. The claim is about the *algebra of coherence over them*, not about the things themselves. (b) That Phi_K from any one surface is sufficient. Each surface sees its own projection; some operationally relevant structure lives entirely in the kernel (e.g. lock contention) and does not project meaningfully to the network surface. The unification predicts agreement when a shared dynamical event is observable in two surfaces; it does not predict that every event is observable in every surface. (c) That the three surfaces can be combined into a single *bundle*. They cannot, at present: the bundles have different vantage types, different tick rates (10 ms for data-plane, 100 ms for kernel, 1 s+ for observatory), and different storage substrates. A unified bundle would be a separate piece of engineering work (U9.6) that is not implied by the algebraic unification claim. (d) That the framework's open questions are smaller because of the unification. They are larger: every open question in v1.1, the data-plane profile, and the kernel profile remains open, and the new open questions enumerated below (U9.x) are additional. (e) That the unification claim is "the" theoretical contribution of the framework. It is *one* contribution, and an organising one; the more concrete contributions live in the per-profile documents and depend on per- profile empirical validation. ============================================================================== 6. Concrete cross-surface experiments ============================================================================== The five predictions of Section 4 translate into concrete experiments. We sketch four; a real experimental protocol is the subject of U9.3. E1. Noisy-neighbour propagation. Run a database container co-located with an analytics container on a host instrumented with both the kernel profile (MVPS-K) and the data-plane profile (MVPS-DP, in software via VPP). Drive the analytics container through the standard stress-ng + fio + iperf3 recipe of MVPS_KERNEL_PROFILE.txt Sec. 5. Record Phi_K_kernel(t) and Phi_K_dataplane(t) at 10 Hz for 60 minutes per run, 50 runs. Hypothesis to confirm: P1 (lag-bounded agreement), P3 (kernel-first transition), P5 (joint ALARM precision gain). E2. BGP-event correlation. Run the observatory profile against a transit operator's BGP feeds while simultaneously running the data-plane profile against the operator's edge ECMP groups. During a planned maintenance window, record Phi_K_observatory(t) and Phi_K_dataplane(t) for 24 hours. Hypothesis to confirm: P1, P2 (axis attribution). E3. Fault-injection orthogonality test. On the same instrumented host, inject *one-axis-pure* faults in turn: - causal-only: cgroup-level CPU throttle that creates scheduler-latency anomalies with no memory / I/O component; - informational-only: deliberate flow-distribution shift via socat that perturbs C_2 without touching C_1 or C_3; - topological-only: deliberate kernel lock-contention via a synthetic load that touches a single shared zone. Record per-axis D^2 attribution across the three surfaces. Hypothesis to confirm: P2 (consistent per-axis attribution). E4. Calibration transfer pilot. Calibrate Sigma^{-1}_observatory on a 1-week trace from a RIPE Atlas anchor; use its structural pattern (which off- diagonals are large) to initialise Sigma^{-1}_dataplane and Sigma^{-1}_kernel on the same operator's hosts; measure time-to-stable-calibration with and without the structural prior. Hypothesis to confirm: P4 (calibration transfer). ============================================================================== 7. Relation to existing unifications ============================================================================== The unification claim of this document is not new in spirit; framework-wide unifying objects appear throughout science and engineering. We position MVPS's unification briefly against four reference points. 7.1 Information geometry (Amari et al.). Information geometry unifies probability distributions as points on a manifold, with the Fisher information metric playing the role of a Riemannian metric. MVPS's coherence triple (C_1, C_2, C_3) is not, today, formulated as a manifold; the open question U9.4 asks whether the right geometric structure on the coherence cube is metric, affine, or order-theoretic. 7.2 Statistical thermodynamics. Statistical thermodynamics unifies macroscopic observables (temperature, pressure, entropy) as functions of microscopic state distributions. MVPS's operational Hamiltonian H, and the K-block observables of v1.1 Sec. 3, are *physics- inspired* in this sense, but the analogy is descriptive, not derived (v1.1 errata E3). The unification claim of this document is more modest: it is about the algebraic structure of coherence across vantage types, not about a deep thermodynamic identity. 7.3 Control theory and state-space modelling. Classical control theory unifies disparate physical systems by representing them as state-space models with inputs, states, outputs, and feedback. MVPS's bundle B(t), coherence vector x(t), Mahalanobis distance D^2(t), and phase label Phi_K(t) compose, informally, a state-space model of the operational reality. Open question U9.5 asks whether this informality can be tightened to a formal control-theoretic model that admits standard stability and reachability analysis. 7.4 Categorical systems theory. Recent work in applied category theory (Spivak, Fong, et al.) provides explicit machinery for "wiring diagrams" and "operadic composition" that may furnish the unification claim of this document with a heavier formalism than the light functor of Section 3. Whether the additional weight buys anything operationally is open question U9.7. In none of these four reference frames does MVPS's unification claim become trivially true. In all four, it has a defensible corner. The framework's wagering is that the corner large enough to be useful and small enough to be honest is the *operational* corner, where the coherence axes have engineering meaning and the phase labels drive concrete actions. ============================================================================== 8. Implications for the publication and validation pipeline ============================================================================== The unification claim, if adopted as the organising frame of the MVPS framework, reshapes the publication and validation pipeline: 8.1 Single I-D, three companion profiles. The draft-melegassi-ippm-mvps-bundle I-D remains the normative reference for the bundle data structure and the observatory algebra. The three profiles (observatory in v1.1, data-plane in v0.1, kernel in v0.1) become non- normative companions, each addressing a distinct vantage substrate. The unification document is a fourth non-normative companion that articulates the relationship between the three. 8.2 Empirical work is now multi-surface by default. Validation work on any one profile should, whenever feasible, include the joint observation of at least one other profile, to feed the cross-surface predictions of Section 4. This raises the cost per validation experiment but lowers the cost per *evidential bit* about the framework as a whole. 8.3 Master's and doctoral entry points (thesis pipeline). Each prediction P1-P5 is, on its own, a defensible thesis chapter or stand-alone paper. Each experiment E1-E4 is a reproducible lab project. Together they constitute a multi- year programme for the framework's empirical maturation. The thesis-kit catalogue (PHD-CONTINUATIONS.txt) is updated by U9.8 to include this programme as a structured set of entry points. 8.4 Industrial deployment story. The unification claim does not require all three profiles to be deployed simultaneously in production for the framework to deliver value. The observatory profile alone is deployable today; the data-plane and kernel profiles are deployable independently as engineering work matures on each. The unification claim is what justifies the long- term commercial value: the same algebra, the same dashboard contract, the same alert taxonomy, scale across surfaces as the operator's appetite for instrumentation matures. ============================================================================== 9. Open questions and validation roadmap (U9.x) ============================================================================== Eight open work items derive directly from the unification claim. Each is independently testable; each contributes a distinct evidential bit toward (or against) the claim. U9.1 Lag distribution characterisation. Status : conceptual. Scope : measure the empirical lag distribution between Phi_K transitions on coupled surface pairs (kernel <-> data-plane, kernel <-> observatory, data-plane <-> observatory) under controlled stress. Determine whether the bound tau_max required by Prediction P1 (Section 4) is finite, stable across workload classes, and operationally useful. Risk : low; the experiment exists once the per-profile reference implementations land (D9.1 + K9.1). U9.2 Per-axis attribution catalogue. Status : conceptual. Scope : assemble a curated catalogue of incidents (synthetic and real) classified by the axis-of-origin (causal / informational / topological) and verify that the three profiles agree on attribution. Quantify the agreement rate; identify and explain disagreement cases. Risk : moderate; the catalogue must be sized large enough to support statistical claims (target: 1000 labelled incidents). U9.3 Cross-surface experimental protocol. Status : not started. Scope : a written, peer-reviewable protocol describing the four experiments E1-E4 of Section 6 in detail sufficient for independent replication. Includes hardware specification, software versions, ground-truth labelling procedure, and statistical analysis plan. Risk : low; this is straightforward methodological work. U9.4 Geometry of the coherence cube. Status : open theoretical question. Scope : determine whether the coherence cube [0, 1]^3 carries a natural metric (Fisher information?), affine (information-geometric divergence?), or order-theoretic (lattice on phase labels?) structure under which the substitution principle of Section 2.4 becomes a structure- preserving map. The answer determines whether the unification claim is geometric, statistical, or combinatorial in flavour. Risk : high; this is the hardest item on the list and the one most likely to require a collaborator from applied mathematics. U9.5 Control-theoretic state-space embedding. Status : open theoretical question. Scope : write the MVPS trajectory ( x(t), Phi_K(t) ) as a formal state-space model x(t+1) = f( x(t), u(t) ) + w(t) with explicit input variable u(t) (operator actions, environmental drivers) and noise w(t). Determine reachability, observability, and stability properties; map MVPS's hysteresis (Sec. 4.2 of MVPS_DATAPLANE_PROFILE) to standard control-loop hysteresis theory. Risk : moderate. The form of the equations is open; fitting them to observed data requires care. U9.6 Unified bundle data structure. Status : open engineering question. Scope : design (and reference-implement) a single bundle data structure that carries observatory, data- plane, and kernel observations at compatible tick rates, with shared tick alignment and shared Sigma^{-1} calibration. This is the engineering crystallisation of the unification. Risk : moderate. The tick-rate mismatch (1 s observatory, 100 ms kernel, 10 ms data-plane) requires either down-sampling (information loss) or hierarchical representation (engineering complexity). U9.7 Categorical-systems-theory formulation. Status : conceptual. Scope : recast the unification claim in the language of applied category theory (operadic composition, wiring diagrams, polynomial functors). Determine whether the additional formalism yields statements that the light functor of Section 3 does not. Risk : low-to-moderate; the formalism exists and is well-developed. The value of applying it here is itself the open question. U9.8 Thesis-pipeline mapping. Status : work-in-progress, internal. Scope : update PHD-CONTINUATIONS.txt to mark Predictions P1-P5 and Experiments E1-E4 as structured master's and doctoral entry points. The unification document becomes the index of those entry points; the per-profile documents become the substrate they operate on. Risk : low. ============================================================================== 10. References ============================================================================== Normative. [MVPS-MATH] Melegassi, L. "MVPS -- Three-Layer Mathematical Structure". Catellix Research, v1.1, 2026-05-20. [MVPS-DP] Melegassi, L. "MVPS -- Data-Plane Profile (Proposal)". Catellix Research, v0.1, 2026-05-21. [MVPS-KERNEL] Melegassi, L. "MVPS -- Kernel Profile (Proposal)". Catellix Research, v0.1, 2026-05-21. [MVPS-BUNDLE] Melegassi, L. "The MVPS Bundle". draft-melegassi-ippm-mvps-bundle, work in progress. Informative. [POINCARE] Poincare, H. "Science et Methode". Flammarion, Paris, 1908. ("L'art de donner le meme nom a des choses differentes...") [AMARI] Amari, S. "Information Geometry and Its Applications". Springer, 2016. [SPIVAK] Spivak, D. I. and Fong, B. "Seven Sketches in Compositionality: An Invitation to Applied Category Theory". Cambridge University Press, 2019. [STROGATZ] Strogatz, S. H. "Nonlinear Dynamics and Chaos", 2nd ed. Westview Press, 2014. [ASTROM] Astrom, K. J. and Murray, R. M. "Feedback Systems: An Introduction for Scientists and Engineers", 2nd ed. Princeton University Press, 2021. [SCHEFFER] Scheffer, M. et al. "Early-warning signals for critical transitions". Nature, 461:53-59, 2009. ============================================================================== Document history ============================================================================== v0.1 2026-05-21 Initial draft. Meta-companion to v1.1, the data-plane profile v0.1, and the kernel profile v0.1. Articulates the substitution principle as a constant functor C: Vantage -> Coherence and enumerates the five empirical predictions (P1-P5) and four experimental sketches (E1-E4) that falsifiability of the unification claim requires. Authors: L. Melegassi (Catellix Research). ============================================================================== End of document ==============================================================================