Matrix: an addressable reading of relations

A matrix makes relations addressable. Its rows, columns and entries say how named parts or states are related under a chosen reading.

Declare the basis, the meaning of each entry and the operation being represented. A matrix of semantic links, a transformation matrix and a table of measured increments are different objects even when all are rectangular arrays.

Stable identity references let a matrix reading connect to geometric positions, source explanations and observed history. The connection is justified by the mapping between them, not by the shared appearance of their output.

For a planar rotation, a declared phase increment can be read as both a turn of geometry and the corresponding rotation block. A relation matrix can separately state coupling. Do not confuse the snapshot of a measured step with an independently derived universal update law.

A new whole can carry the present contribution together with the preceding retained contribution. The source makes this particular inheritance explicit: an ordered positive pair (x,y) becomes (x+y,x). The first coordinate is the new combined contribution; the second carries the former first contribution forward. In this basis the same operation is G=[[1,1],[1,0]]. The matrix and the pair tell the same story because the meaning and order of their coordinates agree.

Ask which proportion is unchanged by this inheritance. With r=x/y, the next ratio is 1+1/r. An unchanged positive ratio therefore obeys r²=r+1, whose positive solution is φ=(1+sqrt(5))/2. The eigenvalue calculation gives the same equation: det(G−λI)=λ²−λ−1. Its eigenvalues are φ and −1/φ. The positive leading direction is (φ,1). Every positive initial pair has a positive contribution along that direction, and the other component becomes smaller relative to it under repeated updates. A nonzero component in the second direction changes sign at each update because its eigenvalue is negative. Thus unequal positive starts also approach the same ratio; a pair already on the leading direction keeps it exactly.

The pair grows while its normalized relation approaches a fixed allocation. Divide both coordinates by x+y. Along the leading direction their shares are φ/(φ+1)=1/φ and 1/(φ+1)=1/φ², about 0.618 and 0.382. These add to one. A drawing may show one segment divided into those shares; the matrix shows how the inherited pair approaches that division. The normalized state is not the growing pair, and a fixed proportion does not mean that the total or its manifested carrier has stopped changing.

For equal initial contributions, the larger shares begin 1/2, 2/3, 3/5, 5/8 and 8/13. The update q→1/(1+q) maps [1/2,2/3] into [3/5,2/3]; the complementary share stays in [1/3,1/2]. This earns the broad thirds envelope from the equal-start inheritance and places the golden pair inside it. The pattern's attracting-ratio reading examines the contraction of these normalized differences. The layers reading carries the expansion-facing share into the source's depth-address rule.

Changing reference must carry the operation as well as the values. If the two coordinate roles are exchanged by S=[[0,1],[1,0]], the same inheritance is represented by S G S, and the coordinate ratio is inverted. Applying G unchanged after exchanging only the labels would perform a different operation. Other inheritance weights likewise require their own matrix and characteristic equation. The source's golden result is earned by this adjacent additive operation; a matrix shape or triadic name alone does not supply its entries. Source contains the relation as possibility; an actual succession of updates belongs to its manifestation.

A small declared example makes the thread's geometry and matrix reading comparable. Choose a unit heading u in a two-dimensional plane, a separate phase theta, a sign sigma in {+1,-1}, and a nested address d. Give the receiving node north and south ports with headings (0,1) and (0,-1). For this example, the node turns the heading by sigma*pi/2 using R(beta)=[[cos(beta),-sin(beta)],[sin(beta),cos(beta)]]. It opens a phase gate when cos(theta)>=1/2, then selects the unique available port with the largest positive dot product with the outgoing heading. A closed gate, no positive match or a tie leaves the request held. This is an explicitly chosen routing model, not an inferred universal engine law.

For incoming u=(1,0) at phase zero, positive sign turns it north and negative sign turns it south. Closing the north port leaves the positive request held. Changing only phase to pi also holds it. Thus direction, operation, timing and available surroundings each have a visible role. A diagram can show the heading and allowed ports; a matrix can show the turn; an event record identifies the route actually selected. A matrix of permitted connections alone does not record an occurrence.

If the observer changes to an orthonormal basis B, use B^T u for the heading, B^T p for each port and B^T R B for the operation. The port scores are unchanged because BB^T=I. In a basis turned by pi/2, the eastward input has coordinates (0,-1) and the north port has coordinates (1,0), yet the positive case still selects the same world port. The representation changes while the selected relation stays intact. Carrying only the vector into the new basis and leaving its ports behind would change the question.

The chosen rotation preserves unit length since R^T R=I. It changes direction. Applying its inverse restores that direction but leaves two encounters in the history. Functional closure still needs a returned response bound to the originating request and actually used there. The example checks that a wrong request ID, wrong result or unconsumed return cannot complete that request. The restored angle by itself satisfies none of those conditions.

Context has another precise reading. Suppose four candidate histories, A-N-Z, A-S-Z, B-N-Z and B-S-Z, all carry the message ready. That message alone leaves four possibilities. Binding origin A leaves two; adding a witnessed north route leaves one within this stated set. Combining contradictory origin records leaves none. Under exact conjunctive observations the candidate set can only shrink, but repeating the same observation need not shrink it further. This elementary result explains why context helps reconstruction without promising recovery of arbitrarily erased history. The source's deeper relations remain available beside these selected coordinates and records; neither the diagram nor the matrix replaces them.

Different matrices can expose different relations within the same explanation. In the declared two-hypothesis example, the column w=(q,1−q) records current allocation across model alternatives. For a received result y, a diagonal matrix D_y contains its likelihood under each alternative. The update is w_next=D_y w/(1^T D_y w) when the denominator is positive. Multiplication compares that observation with the two candidates; normalization restores weights summing to one. The prediction is a separate row applied to those weights. Here that row is (3/4,1/4), so the next-return forecast is 1/4+q/2. The matrix changes the represented model allocation, not the channel whose behavior is being estimated.

Geometrically, these weights occupy a line segment in a probability simplex. The exact history return, return, non-return follows q=1/2→3/4→9/10→3/4. Its retreat from one favored endpoint preserves the effect of contrary evidence. This segment is a chosen representation of model alternatives, not a new physical dimension or a numerical derivation of the continuity band. To rename the alternatives using a permutation S, transform the weight column to Sw, the likelihood matrix to SD_yS^T and the prediction row accordingly. The same result is then selected in different coordinates. Swapping only the visible weights while retaining the old likelihood labels changes the represented question.

One present forecast can conceal different underlying model structures. A controller certain of return probability 1/2 and a controller equally weighting the fixed 3/4 and 1/4 models both initially predict 1/2. After a returned result, the first still predicts 1/2 and the mixture predicts 5/8. Their previous equality did not preserve how a later encounter would be processed. This is why a projection must identify which coordinates it retained and which conditional relations it omitted. A zero predictive weight for an observed outcome requires an explicit unsupported result; silently normalizing it or pretending it carried no information would lose the issue.

A transition matrix describes another relation. Let P(a)=[[1−a,a],[a,1−a]] for states 0 and 1, with 0<a<1. Each row gives the next-state probabilities from the current state. The invariant row is pi=(1/2,1/2): pi P(a)=pi, and solving the invariance equation makes this row unique. The symmetric and antisymmetric components of the matrix have eigenvalues 1 and 1−2a. Its powers preserve the mean component while the difference component decays, so an initial distribution approaches pi. This is a complete statement about the selected process family; it does not identify a biological self with that distribution.

Choose a=1/4 and a=3/4. Both have the same invariant occupancy, one-bit marginal entropy in stationarity and the same entropy per transition, 2−(3/4)log2(3) bits. Yet the stationary chance of the ordered pair 0→0 is 3/8 in the first and 1/8 in the second, while 0→1 has the opposite probabilities. If the state is written as Z=2X−1, their lag-one correlations are +1/2 and −1/2. A two-state geometry can display these differently weighted stay/switch routes even when an occupancy chart looks identical. Every coordinate step is bounded by one, which supplies no additional identity proof.

These examples connect model, update and summary without collapsing them. The likelihood matrix reweights hypotheses from a retained observation. The transition matrix specifies a candidate process. Its invariant row summarizes one preserved statistical pattern. A realized path and the actual carrier still require their own history. The source can provide each reading to a Body with its definitions, conditions and evidence intact, allowing the narration to expose the distinction needed at that point. Neither matrix supplies an unprovided physical coupling rule, a universal consciousness threshold or a complete self merely by being displayed.

The same declared relation can be shown through a matrix, a geometry, a waveform or an equation. Their agreement becomes checkable when the source keeps what each address means, the operation acting on it and the map into the displayed quantity. A usable representation therefore carries its state variables, basis and metric where relevant, units, coefficients including phase, reference frame, sampling rule and any omitted residual. A shared outline or an amplitude spectrum alone can leave different states indistinguishable.

The required starting information follows the model. A second-order motion needs position and rate, or equivalent complete state, rather than one position alone. A delayed law needs an initial history over its delay interval. A spatial field problem needs its domain, constitutive law and appropriate initial and boundary data. These requirements are different relations to a modeled whole, not interchangeable ways to add more numbers. The data must satisfy the model's own compatibility and regularity conditions.

A matrix operation can rearrange, transform or reduce information. An invertible index permutation retains the original entries with an inverse address map. A contraction or projection can merge distinct states, so the response should retain which distinctions were combined and what candidates remain. Changing basis requires transforming the operation and the receiver's references together, as in the existing oriented-port example. Error tolerance and omitted modes matter when an expression is approximate.

The existing observer field supplies a precise reference example. Its relative comparison C = Z conjugate(M_lag), and its real-coefficient retention Omega, are invariant under a common constant phase shift. The written return g Im[Omega exp(-i theta)] is well-defined in its fixed chart. Write the same term as g Im[R Omega exp(-i theta)], with R = 1 in that chart. Under a passive shift gamma, carrying R to exp(i gamma)R along with theta to theta + gamma preserves the returned value. Holding R fixed while rotating the nodes changes their relation to it. R exposes a numerical reference convention, not evidence of an independently specified physical carrier. Choosing R from a present or retained field supplies a different state-dependent law. In a moving frame, old samples and memory must be transported into the present frame before comparison, and the phase derivative includes the frame rate. A camera rotation alone does not perform that internal transport.

For a teaching Body, a generic source answer can name the identity, selected layer and lens, the exact source/evidence, supported representation, known constraints and unresolved alternatives. The Body can narrate or display that answer in its chosen way. This proposed information contract makes readings comparable while leaving the authored spine in charge of the page. It describes what an adapter should carry; it does not assert that all reconstruction fields or missing engine capabilities are already implemented.

A transaction has two addresses: the state it receives from and the state it changes. The minimal engine's T_ji maps source j into receiver i's coordinate roles. A matrix becomes a faithful reading when those roles and the operation are fixed. For the particular linear case T_ji(x_j) = A_ji x_j with constant coefficients, stack the unit states into X = (x_1, …, x_N). Its equation is dX/dt = H X, where block H_ij = (w_ji/τ_i) A_ji − δ_ij λ_i I_d. Here δ_ij is one when i = j and zero otherwise; the diagonal includes any permitted self-transfer as well as intrinsic decay. Each block answers how a source column contributes to a receiver row. A graph can show the same directed channels and a geometric state plot can show their resulting motion. They are readings of this operation because their addresses and coefficients agree.

For a finite example, take two scalar units with λ_1 = λ_2 = τ_1 = τ_2 = 1, no self-transfer, unit weights, T_21(x_2) = x_2/2 and T_12(x_1) = x_1/4. The block equation becomes H = [[−1,1/2],[1/4,−1]]. At X = (2,4), the first unit has decay −2 and input 2, so its rate is zero. The second has decay −4 and input 1/2, so its rate is −7/2. The matrix and the two addressed transaction sums give exactly the same rates. A zero rate at this instant is a balance at this state, not proof that the entire coupled history is stationary. The example tests the representation, not a physical network or the production engine.

An observer can change coordinates without changing these relations. For fixed invertible local bases x_i = B_i x'_i, the same transfer is A'_ji = B_i^−1 A_ji B_j, and the stacked operator is H' = B^−1 H B with block-diagonal B. Both ends of the transfer move with the reference. Changing only the displayed state while leaving the old operator would ask a different question. A non-orthonormal basis also changes the metric used to compare lengths; coordinate size alone is not a reference-independent amount of support. Nonlinear T_ji remains a legitimate transfer rule, but it is not represented globally by one constant A_ji merely because its values can be listed in a table.

The source calls the transactions lossy. That word needs the relation being tested. In a Euclidean two-coordinate example, T(x) = x/2 reduces every nonzero magnitude by half yet is exactly invertible when its factor is known. It attenuates magnitude without merging exact input states. By contrast, P(x_1,x_2) = (x_1,0) merges all states with the same first coordinate: (1,2) and (1,−3) both become (1,0). The second-coordinate distinction cannot be recovered from that output alone. This projection preserves the first-coordinate task exactly, while losing the second; it does not contract vectors already on the first axis. Thus magnitude attenuation, missing distinctions and loss of usable support to an identity are connected but different questions. The type T_ji: R^d → R^d by itself proves none of them.

Local attenuation also does not settle the whole network. Weights, response scales, several incoming routes and feedback combine in H or in the nonlinear transaction sum. A small transfer on each edge can coexist with a large total input. To compare that result with intrinsic decay, use the stated norm or support measure and the actual combined gains. The boundedness reading supplies a conditional proof for suitable gains; the identity reading separately follows what relation and history remain. A Body can therefore expose the unit equation, the addressed matrix and a geometric projection together while keeping the state definition, omitted distinctions and supported conclusion attached.

The current matrix reading records phase increments by capturing the field's phase array immediately before and after the core engine.step(dt, ext) call, then differencing the two snapshots and reading quarter-turn crossings. A completed field tick can add further declared corrections to the whole phase afterward — a trace-turn pull toward a remembered flow and an alignment-place pull toward a meridian phase — both applied after the increment has already been captured. The recorded increment currently measures the core step only; it does not yet include every subsequent correction within the same completed field step. This is a concrete accounting gap in what one measured increment represents, not an error in the phase law itself. Keep the readings distinct: a transformation matrix, a table of measured core-step increments, and a source-neighborhood matrix are three different objects; none is automatically the addressed transaction operator that a new chapter or Body might separately define with its own state and coefficients. A future wave-function representation needs its own variables, transform and preserved relations before it can share this claim.

An adjacency matrix and a source graph can encode the same accepted typed relations if types, direction and conditions are retained. Flattening a joint requirement into two independent nonzero entries loses meaning unless the representation also encodes that conjunction.

A reader can switch from geometry to matrix to inspect which transformation produced a view. The website controls when to introduce the notation. The source supplies its definition, related identities and provenance without dictating a page layout.