Research-Stack/6-Documentation/docs/geometric-substance/GEOMETRIC_SUBSTANCE_CANONICAL.md

Geometric Substance — Canonical Reconciliation

Status: Canonical. This document supersedes the scattered and conflicting usage spread across VOCABULARY_LOCK.md, NotationNomenclatureRegistry.md, the otom/docs/* conceptual notes, and the individual Lean docstrings. Where a term has carried multiple expansions across conceptual revisions, the entry here is defined by mechanism, not by label, and the superseded labels are recorded so older documents remain traceable.

Reading rule (project-wide): labels in the legacy stack are unreliable — they were assigned across many revisions and by an automated labelling pass. Trust the mechanism (the Lean definition, the proven theorem, the formula), not the name. Every claim below is tagged with an evidence tier in §7.

---

1. The geometric substance

The object the system manipulates is a single covariant geometric datum on a fixed model space, viewed through many projections. The model space is

ℂ⁸  ≅  ℝ¹⁶   with complex structure J   (a U(8) / Kähler G-structure)

This is not an interpretation imposed from outside — it is what the code builds. Law15_Field.goldenSpiral16 acts block-diagonally on 8 complex planes, each block [[a,−b],[b,a]] with λ = a + ib = φ⁻¹·e^{iθ_g}, and it is admitted only because complex-scalar multiplication commutes with J ("passes the conformal Kähler gate"). The negative controls confirm the intent: shear16 is rejected (non-orthogonal) and complex conjugation is rejected because CᵀJC = −J (anti-holomorphic).

The geometry is Cartan geometry, not Riemannian. The dictionary:

Cartan role	Realization in the stack
connection	polarity accumulator B along braid strands
curvature / torsion	braid residual R_ij = B_ij − (B_i + B_j) (BraidDiatCodec Layer 3)
development onto the model	golden contraction s' = c + φ⁻¹·(s − c), with ‖Sᵗs − c‖ = φ⁻ᵗ‖s − c‖ (PistSimulation §8)
G-structure admissibility	the conformal Kähler gate — preserve J (Law15_Field); this is the FAMM filter

The geometry carries intrinsic torsion: the super left-invariant forms are not closed, so R_ij ≠ 0 is torsion rather than an optional add-on. This is the load-bearing reason the model is Cartan (torsionful) and not Levi-Civita (torsion-free). The torsion residual is a five-channel bracket (lower, upper, gap, κ, φ), carrying an explicit curvature channel κ and a phase channel φ.

The super extension is a gated module — off by default. The odd / graded directions come from the chirality flag (left / right / achiral, BraidDiatCodec Layer 1) and the eigensolid's (2k−1, 2k) coordinate pairing (BioSight G3), which is a ℤ/2 grading. Whether the grading is genuinely super — Grassmann-anticommuting, picking up the (−1)^{deg·deg} sign — is a design decision, not a fact already in the code. The program proceeds from the proven bosonic Kähler base; super-Cartan is an extension module that stays off until its gate passes (§7). This is the promotion-ladder discipline applied to the geometry itself: the core (DIAT bijection, φ-contraction, the encoder, the differential test) depends on none of the super structure, so it advances regardless of how the gate resolves.

---

2. The observer / observerless duality

The stack is the Observerless Research Stack: truth lives in receipt-bearing, invariant-preserving events, not in any privileged frame. Law17_Observer makes this exact — the observer is a typed projection, not an agent:

"The observer is not a separate agent but a typed projection: Π₁₆→₃ applied to the object. The measurement residual tracks what was lost in projection."

So:

• Observer = a choice of projection Π (an ObserverGate, or an orientation in SO(n) per ObserverAngle). It produces one locality-specific shape: the projected silhouette of the object (cube-along-the-diagonal → hexagon).
• Collapse residual ε_collapse = ‖M_before − M_after‖ — the mass Π discards. Aligned angles minimize it; the aligned angle reveals the object's minimal intrinsic dimension.
• Observerless = the covariant object that is true across all Π — the frame-free invariant, i.e. the equivalence class under the structure group. The "observerless-observer symbol" is Π itself stripped of any subject: an observation with no one behind it.

"Covariant geometries in locality-specific shapes" is exactly this: one invariant object, the structure group acting on it, many projected shapes.

---

3. The Sidon mirror (the inexact reflection)

For a Sidon set S, the mirror-translation c − S is again Sidon and — the key fact — has the same difference set:

D(c − S) = { (c − s_j) − (c − s_i) } = { s_i − s_j } = −D(S) = D(S)

because the difference set is reflection-symmetric. Therefore:

• At the observerless level (autocorrelation / difference structure / Fisher invariant), S and its mirror are exactly identical.
• At the observed level (any projected shape), they differ — and the difference is exactly the collapse residual ε_collapse(θ), which vanishes only when the observer-angle θ aligns with the reflection axis.

This is the precise content of "a mirror translation, a not exact one": exact in the covariant object, inexact in every projection, mediated by Π. The chirality flag (left / right / achiral) is the discrete ledger of it — achiral marks the angle where the mirror is exact, left/right mark the inexact pair seen off-axis. Chirality is not merely an address bit: the GWL coupling w_ij = cos(Δθ)·cos(Δφ)·(1 − 2|Δχ|)·exp(−|Δp|²/2σ²) carries the chirality difference |Δχ| as a first-class coupling term.

In the Cartan frame this is the standard fact, not a special case: a symmetry of the model G/H need not be a symmetry of a given realization — it holds only up to the structure group, exact covariantly and inexact in a fixed frame, with the residual measuring frame misalignment.

The golden angle. ObserverAngle says the aligned angle reveals minimal dimension and exact symmetry. The golden angle θ_g from PhiNUVMAP is the maximally mis-aligned orientation — the most-irrational angle, aligning with no rational symmetry axis. It is the least-privileged viewpoint, the angle that refuses to pick a frame: the closest realization of an observerless observer as an actual angle. This is why the golden contraction is the natural generic probe.

Incoherence as the shared resource. The same principle appears in three categories, and is the spine of the whole program:

category	instance	the resource
additive	Sidon set (distinct pairwise sums)	flat autocorrelation
geometric	high-dimensional sphere (near-orthogonality, Ma–Shen–Xie 2025)	clique suppression
dynamical	golden angle (Weyl equidistribution)	resonance / collision avoidance

The golden ratio appears because θ_g is provably the maximal-incoherence rotation (continued-fraction theory), not by analogy.

---

4. Vocabulary lock

Defined by mechanism. "Superseded labels" are recorded only for traceability to older revisions; do not use them.

Term	Canonical role (mechanism)	Locus	Superseded labels
DIAT	Integer address by perfect-square shell: k = ⌊√n⌋, a = n − k², b = (k+1)² − n. Provably bijective (encode_decode_roundtrip).	BraidDiatCodec.lean	"Dynamic Integer-Address Transform", "Dual-Interval Algebraic Transform"
PIST	The imperfect-square witness / audit surface; conserves mass = t·(2k+1−t) under lawful transitions.	PIST/*, ARCHITECTURE.md	"Perfectly Imperfect Square Theory" (keep as flavor; the role is the witness surface)
NUVMAP	Non-uniform projection onto a spectral / address coordinate surface (more resolution on important regions). Not a proof engine.	NUVMAP_NAMING_AND_DEFINITION.md	"Virtual Memory Address Projection", "Variable Mapping", "spectral container"
PhiNUVMAP	NUVMAP lifted to ℂ⁸ (16D) golden-ratio fractal coordinates with J; development = φ-contraction.	PistSimulation.lean §8, Law15_Field.lean	—
eigensolid	The pairwise-averaging fixed-point map C(p)_{2k−1} = C(p)_{2k} = (p_{2k−1}+p_{2k})/2; convergence is a compressor requirement. Distinct from NUVMAP — it is the merge on the surface, not the surface.	BioSight G3, BraidTreeDIATPIST.lean	(often conflated with "NUVMAP merge")
braid residual	R_ij = B_ij − (B_i + B_j) — discrete curvature/torsion, 5-channel (lower, upper, gap, κ, φ).	BraidDiatCodec.lean Layer 3	—
TreeDIAT	Tree-embedding score for routing/pruning plus a homeomorphic-embedding certificate (Kruskal WQO). Score routes; only the embedding proof certifies.	TreeDIATKruskal.lean	—
chirality	Mirror-handedness ledger (left / right / achiral); first-class coupling term `(1 − 2	Δχ	)`.
observer / Π	A typed projection (no agent); ε_collapse = ‖M_before − M_after‖ is the projection loss.	Law17_Observer.lean	—
observerless	The covariant object true across all Π — the frame-free invariant.	stack-wide (ARCHITECTURE.md)	—
FAMM	Admissibility filter = G-structure preservation (the Kähler J gate).	2-Search-Space/FAMM, Law15_Field.lean	"Frustration Aligned Memory Management"

---

5. Mechanism → port role

For folding the legacy mechanisms into BioSight / SilverSight (the port spec):

1. R_ij braid residual → curvature/torsion operator. Replaces BioSight's degree-±1 proxy and SilverSight's hardcoded nuvmap_spectral_driver.py diagonal; computed from the real graph, carrying the 5-channel bracket.
2. phiContract + goldenSpiral16 → development + admissibility. Replaces BioSight's ad-hoc G1 contraction λ = 1/√(1+B) with the φ⁻¹ golden contraction (proven φ⁻ᵗ law); FAMM admissibility = the Kähler gate.
3. TreeEmbeds + certificate → separation metric with rigor gate. Parse-tree (τ-block) separation; scalar score routes, the embedding proof certifies. No claim is promoted without an embedding witness.
4. DIAT encode_decode_roundtrip → invertibility receipt. The proven bijection is the receipt_invertible requirement, made formal.

---

6. The Laplacian resolution

Long-standing open question: does spectral binning eigendecompose the plain graph Laplacian or the sheaf Laplacian? With the model space fixed as Kähler (ℂ⁸, J), neither: the natural operator is the J-compatible complex / Dolbeault Laplacian, the one that respects the structure the development map preserves. The geometry selects the operator.

---

7. Super-Cartan as a gated extension — the gravity-consistency gate

Super-Cartan (§1) is off by default. It activates only when the gravity-consistency gate passes:

Gate. R_ij ≠ 0 (residual present) and the curvature class of the connection is identified (teleparallel if the curvature part is flat, Einstein-Cartan if not).

Why the gate exists — dissolving "the model says gravity doesn't exist." That worry holds in exactly one case: the geometry collapsing to its flat model with R_ij ≡ 0. It does not follow from "zero curvature," because the geometry is torsionful. Teleparallel gravity (TEGR / Weitzenböck) is empirically GR-equivalent with curvature identically zero and gravity carried entirely by torsion. So R_ij ≠ 0 means the geometry gravitates — teleparallel if the curvature part is flat, Einstein-Cartan otherwise. Gravity-denying is the single flat-and-torsionless point, nothing more.

The gate is the differential test. Gravity-denying = flat-and-torsionless = zero residual = no separation from a flat random baseline = nothing. The number that shows the encoding beats random is therefore the same number that shows the geometry is not flat-trivial — one obligation, read two ways. Its formal home is BaselineComparison.lean, whose verdict should read "agrees-with teleparallel-or-Einstein-Cartan," not "disagrees."

Initial recovery (provisional — to be stress-tested)

Computed by gate_residual_recovery.py on a prototype residual R_ij = B_ij − (B_i + B_j) with B_i = cos θ_i and B_ij = cos(θ_i+θ_j) + ε·χ_ij·sin(θ_i−θ_j), on a Mian-Chowla Sidon set (n = 16, max 252, phase modulus N = 505 chosen so pairwise sums never wrap). This is the residual mechanism, not the Lean BraidField codec — the magnitudes are model-dependent; the separation is not.

curvature / torsion split
  achiral (ε=0):   ‖R‖ = 18.05   torsion(sym) = 18.05   curvature(asym) = 5e-16  → TELEPARALLEL
  chiral  (ε=0.5): ‖R‖ = 18.80   torsion(sym) = 18.05   curvature(asym) = 5.27   → EINSTEIN-CARTAN
  (gravity-denying needs ‖R‖ = 0; here ‖R‖ = 18.05)

Sidon separation vs 5000 random 16-subsets of [1,252]
  structured pairwise-sum collisions : 0
  random mean ± std                  : 13.65 ± 3.38
  z-score                            : −4.04   (structured sits ~4σ below random)
  random subsets that are Sidon      : 0.00%

Reading the numbers honestly.

• The ~4σ separation (z = −4.04; 0% of random subsets clean) is the load-bearing, genuinely empirical result: the structured input is strongly separated from random on the defining Sidon property. This is the "something rather than nothing."
• The achiral curvature ≈ 0 is a structural identity, not an emergent surprise — a symmetric joint gives a symmetric residual whose antisymmetric (curvature) part is exactly zero. What it demonstrates is that the curvature/torsion split is real and chirality-controlled: chirality is precisely what turns teleparallel (achiral) into Einstein-Cartan (chiral), consistent with the chirality ledger of §3.
• The torsion magnitude (18.05) only certifies ‖R‖ ≠ 0 (not flat-trivial); its value is an artifact of the cos prototype.

Verdict (provisional). ‖R‖ ≠ 0, the curvature class is identified (teleparallel achiral / Einstein-Cartan chiral), and the structure separates from random at ~4σ. So the model affirms gravity in a named, GR-equivalent formulation; it does not point to gravity not existing. The gate opens — provisionally, resting on the §8 "sound reading" that R_ij is specifically teleparallel torsion and that gravity-denying is the "nothing" case. Neither is proven; this is an initial recovery at the RESIDUAL_TESTED rung.

Stress-tests before promotion.

1. Replace the cos prototype with the real BraidField.braidCross residual.
2. Scale n; confirm the separation holds or strengthens.
3. Use the Hachimoji ℤ/360 modulus (needs a mod-360 Sidon / Singer difference set; wrap-collisions change the residual).
4. Q16.16 fixed-point fidelity — does the split survive the real arithmetic.
5. Harder null: the geometric S⁷ / sphere baseline (Ma–Shen–Xie), not just random distinct subsets.
6. Chirality source: derive χ_ij from the real braid word / the GWL |Δχ| term, rather than the fixed all-left pattern used here.
7. Encode the verdict in BaselineComparison.lean.

---

8. Evidence tiers

Using the stack's own promotion ladder (RAW_IDEA → SANITIZED_METAPHOR → TOY_MODEL → TYPED_MODEL → RESIDUAL_TESTED → COST_ACCOUNTED → PROOF_CANDIDATE → CORE_MODULE):

Proven (Lean theorems — PROOF_CANDIDATE / CORE_MODULE). DIAT encode_decode_roundtrip bijection; PhiNUVMAP φ-contraction law; TreeEmbeds node/leaf monotonicity; Law17 collapse-residual; Q0_2 codec round-trip lemmas (native_decide). These are not in question.

Sound reading (TYPED_MODEL — mathematically defensible synthesis, not yet proven in-repo). The Cartan dictionary of §1; the Sidon-mirror result of §3 (the difference-set identity is a theorem; the observerless-exact / observer-inexact framing is the reading of it); the golden-angle = least-privileged-frame identification; the Dolbeault-Laplacian resolution of §6; the gate readings of §7 — that R_ij is specifically teleparallel torsion, and that gravity-denying is the flat-and-torsionless "nothing" case.

Stack-flagged speculative (TOY_MODEL / RAW_IDEA — the stack's own bars). ObserverAngle compression (Toybox, "not for production until 6.5σ", speculative-materials/ObserverAngleCompression.md); higher super-Cartan cohomology (cocycles / definite forms / brane molecule — nLab itself says general super-Cartan "remains to be explored"); the super/odd anticommutation decision of §1. The differential "something rather than nothing" number now has an initial recovery (§7, RESIDUAL_TESTED): ~4σ structured-vs-random separation on a prototype residual, pending the seven stress-tests before promotion.

---

9. Open threads

• The Sidon-mirror notch test. The gate's separation number (§7) is computed; the remaining geometric form is the angle sweep — ε_collapse(θ) predicted to have a sharp zero at the reflection axis and a positive floor elsewhere, flat for a random sequence. That notch is the §3 reading made quantitative.
• The super/odd decision. Impose Grassmann anticommutation on the odd block, or keep it plain graded. Gated behind §7.
• The differential number, hardened. Carry the §7 initial recovery through its seven stress-tests; run phi.encode_phi over Corpus250 vs a matched S⁷ null against the G2 bound, wrapped as an ErdosHarness-style receipt.
• Lock adoption. Propagate this vocabulary into the repo docs, retiring the superseded labels in §4 (see ARCHIVE_SUPERSEDED.md).