Seam: Read-Time Field-Propagation Layer
Definition
Three names from three design documents denote one artifact: a read-time, post-atomic, deterministic, model-agnostic step that takes the drawn position(s)’ pressure (delta-P plus state) and propagates it across the fixed 78-position relational graph, outputting a whole-spread pressure field that is injected into synthesis. The roadmap’s “propagation mechanism” (the named how behind its relational layer), the Council’s “Level-2 Composition Engine,” and 03d’s “Coupling Matrix” are the same operation: propagate delta-P along couplings with attenuation, after the atoms are built, over the 78-position relational network, with identical input and output. As 03d puts it, this is “the same computation the roadmap split across Gap B and Gap D — they are one field,” and that field “IS the relational layer the roadmap calls highest-leverage.” The documents cross-assert the identity by name; the seam records that convergence rather than asserting a new one.
Why It’s Load-Bearing
Each document independently reached the highest-leverage piece of the reading engine and named it differently, which risks three teams building three layers. The seam fixes that there is one layer to build, derived once: local coupling rules over a fixed graph replace the combinatorial enumeration of cross-card interactions (which “explodes into millions”), so a full 78-position field is computable from any draw at near-zero cost on every model tier. The roadmap’s Gap B (patterns among drawn cards) and Gap D (status of all 78 while interpreting one) collapse into a single computation. Remove the seam and the relational layer fragments back into “what to compute” (roadmap) divorced from “how pressure reaches another position” (Coupling Matrix), with the Council’s layer-placement ruling floating unattached.
Guard against false merge (disposition caveat): This read-time relational/composition layer is NOT the homonymous UX/product “composition layer” discussed in the generation and pipeline handoffs (05_Generation_Spec, HANDOFF docs). They share the word “composition” but operate at different layers — one is the field-math step inside reading computation, the other is product/UX assembly. Do not merge them.
Confidence Tier
DERIVED — the identity is forced by the documents’ own cross-assertions, not by external analogy: all four discriminators (same operation, same layer/timing, same coordinate graph, same I/O) align, and the convergence was independently re-logged in the harvest files before this node existed. Confirmed by adversarial verification 2026-06-30; awaiting Council ratification. One sub-grain claim is held at candidate tier: whether the Coupling Matrix is identical to or nested within the Composition Engine is an open question, not settled by this seam. The coupling coefficient magnitudes (0.6 / 0.4 / 0.25 / 0.2 / 0.1) are likewise flagged-for-calibration candidates, not derived values; only the coupling structure is firm.
Cross-References
- Principle_The_Field — the relational field this layer computes and feeds into synthesis (strengthen edge: this seam supplies the read-time mechanism by which the Field is populated from a draw)
Canon Narratives
- source: 03d_Coupling_Matrix_Derivation — the Coupling Matrix derivation; two-layer model (state-gated relief + always-on structural adjacency), worked Compassion/Too-Much propagation, stability constraint (lines 16, 100)
- source: 00_Reading_Quality_Roadmap — continuity note unifying Gap B + Gap D as one field via the Coupling Matrix (line 124)
- source: 03b_Council_Responses — Council Q6 arbitration: relational = Level-2 Composition Engine, field propagation at read-time in the composition layer, “matches 03d” (lines 17, 93)
Open Questions
- Is the Coupling Matrix identical to the Composition Engine, or nested within it (one sub-component of a broader read-time composition step)? Held at candidate tier pending Council.
- Coupling coefficient magnitudes and propagation order (2 vs 3 hops) await one joint calibration pass with rho and beta (03d section 6).