docs: invariant computation geometry — the unifying vision

Capstone document connecting the session's conceptual framework to
all measured findings.

One-sentence statement: 'Computation in the space of invariants,
rather than in any specific representation.'

The matter→light move: nonlinear constraints (matter) → spectral
decomposition (light) → invariant extraction (truth).

The observerless observer = invariant geometry: computation defined
without privileging any representation. Results extracted by
choosing invariants that survive ALL representations. The Φ-metric
defines the geometry of observability.

Three 'endian' regimes = three projections of the same invariant
geometry:
- Big-endian: global invariants (QR/eigenvalues)
- Little-endian: local rules (KV cache/PPM)
- Water/block: continuous dynamics (golden spiral/SLOS)

Key limitation: symmetry group balance.
- Too much symmetry → no computation (cospectral graphs)
- Too little symmetry → no compression (text at 3.088 b/B)

Conservation law = invariant preservation:
  total invariant information ≥ K(data)

Every session measurement maps to invariant language:
- Octagon (4/4) = Φ-metric converts nonlinear → spectral invariant
- Conservation (8 branches) = invariant preservation bound
- CRT (O(1)) = coprime invariant reconstruction
- p-adic = prime invariant decomposition
- Cospectral failure = same invariants, different objects
- Etesami-Haemers = invariant embedding at O(n²)
- GW SNR = signal invariant, noise representation-dependent
- Reaction primes = prime factorization = invariant decomposition

Pipeline = invariant extraction engine:
  DNA (matter) → matrix (operator) → spectrum (light) → invariants (truth)

docs/research/INVARIANT_COMPUTATION_GEOMETRY.md

@@ -0,0 +1,188 @@
+# Invariant Computation Geometry: The Unifying Vision
+
+**Status:** conceptual capstone, connecting all session measurements
+**Date:** 2026-07-03
+**One-sentence statement:** Computation in the space of invariants,
+rather than in any specific representation.
+
+## The Matter → Light Move
+
+The core idea: convert nonlinear physical constraints (matter:
+collision, deformation, binding) into linear spectral problems
+(light: waves, spectra, modes).
+
+Instead of "does it fit?" (matter: geometric collision)
+ask "what modes survive interference?" (light: spectral selection)
+
+Mathematical translation:
+```
+matter problem = nonlinear constraint satisfaction
+light problem = spectral decomposition of a linear operator
+```
+
+The octagon principle IS this move: the matrix embedding converts
+the nonlinear property (matter: pairwise sums) into a spectral
+signature (light: eigenvalue degeneracy).
+
+## The Observerless Observer = Invariant Geometry
+
+Computation defined WITHOUT privileging any single representation.
+Results extracted by choosing invariants that survive ALL
+representations.
+
+The Φ-metric defines what counts as "same object" before physics
+processes it. It defines the GEOMETRY OF OBSERVABILITY:
+- What collapses into the same eigenmode
+- What separates in spectral space
+- What becomes noise vs signal
+
+This is not "no observer" — it's a pre-observation metric structure
+that constrains all possible observers.
+
+## The Three "Endian" Regimes
+
+| Regime | What's primitive | Pipeline analog |
+|--------|-----------------|----------------|
+| Big-endian | Global invariants first (coarse spectral modes) | QR/eigenvalue decomposition |
+| Little-endian | Local rule evolution (microscopic dynamics) | KV cache / PPM prediction |
+| Water/block | Continuous field dynamics (operator flow) | Golden spiral / SLOS propagation |
+
+Different "endian-ness" = different projections of the same invariant
+geometry. The problem: ensuring these projections COMMUTE enough
+to be useful. This is a commuting diagram problem in a geometric
+computation category.
+
+## The Full Pipeline (Stated in Invariant Language)
+
+```
+matter system (nonlinear constraints)
+   ↓ Φ-metric embedding (the octagon)
+operator form (linear matrix)
+   ↓ spectral decomposition
+spectral space (eigenvalues = invariants)
+   ↓ optical/DNA/wave propagation
+measured modes (surviving interference)
+   ↓ invariant extraction
+computation result (representation-independent)
+```
+
+The only stable objects (invariants):
+- Eigenvalues (when the embedding is linear — SLOS)
+- Conserved quantities (CRT residues — coprime observers)
+- Topological invariants (braid crossing structure)
+- Symmetry classes (chirality — left/right/achiral)
+- Equivalence classes (Sidon property — collision-free)
+
+Everything else is representation noise.
+
+## The Conservation Law = Invariant Preservation
+
+The measured conservation law (8 branches, all confirmed):
+```
+program_size + residual_size ≥ K(data)
+```
+
+In invariant language: the total information of the invariants ≥
+K(data). You cannot reduce the invariants below what the data
+requires. This is because:
+
+- The invariants ARE the data's prime decomposition (reaction primes)
+- Prime factorization is unique (fundamental theorem)
+- The total prime information is conserved
+
+The conservation law IS invariant preservation: the invariants
+survive all representations (all observers), and their total
+information content is fixed.
+
+## The Key Limitation: Symmetry Group Balance
+
+Invariants are only as strong as the symmetry group:
+- Too much symmetry → everything identical → no computation
+- Too little symmetry → no invariants survive → no compression
+
+The system lives in the balance:
+- Enough structure to compute (Sidon property, CRT moduli, braid crossings)
+- Enough invariance to unify representations (eigenvalues, residues, chirality)
+
+This is why:
+- Sidon works (enough structure for computation, enough spectral
+  invariance for the octagon)
+- Text fails (too much structure, not enough spectral invariance)
+- Graph isomorphism fails (too much symmetry, cospectral graphs
+  collapse to the same invariant)
+
+## Connection to Session Measurements
+
+| Measurement | Invariant language |
+|-------------|-------------------|
+| Conservation law (8 branches) | Invariant preservation: total ≥ K(data) |
+| Octagon (Sidon 4/4) | Φ-metric converts nonlinear → spectral invariant |
+| SLOS linearity | Linear optical = invariant-preserving propagation |
+| CRT lift (O(1)) | Coprime invariants → unique reconstruction |
+| p-adic valuations | Prime exponents = invariant decomposition |
+| Cospectral failure | Same invariants, different objects (symmetry too strong) |
+| Etesami-Haemers | Invariant embedding exists at O(n²) dimension |
+| GW SNR sweep | Signal invariant, noise is representation-dependent |
+| Reaction primes | Prime decomposition = invariant factorization |
+| Merged O(1) transform | Physics does invariant extraction (hybridization) |
+
+## The Substrate Mapping
+
+| Substrate | Role in the framework |
+|-----------|----------------------|
+| DNA | Combinatorial constraint generator (matter: encodes the problem) |
+| Optics (SLOS) | Linear spectral computation space (light: extracts invariants) |
+| Water/matter | Nonlinear physical constraint space (the problem domain) |
+| Braid topology | Invariant structure (crossing number = topological invariant) |
+| CRT | Coprime invariant projection (multi-observer reconstruction) |
+| Golden spiral | Invariant contraction (φ⁻¹ preserves the invariant) |
+
+The pipeline converts: DNA (matter) → matrix (operator) → spectrum
+(light) → invariants (truth). The Φ-metric is the embedding that
+makes this conversion faithful (when it exists).
+
+## What GPT Got Right (and What We Measured)
+
+GPT's formalization matches the session's measurements exactly:
+
+1. "Matter → light = nonlinear → spectral" ✓ (octagon principle)
+2. "Invariants survive all representations" ✓ (conservation law)
+3. "Φ-metric defines geometry of observability" ✓ (the matrix embedding)
+4. "Too much symmetry → no computation" ✓ (cospectral graphs)
+5. "Too little symmetry → no compression" ✓ (text at 3.088 b/B)
+6. "Commutation of projections" ✓ (the 5-way attack: no universal
+   projection commutes with all problems)
+
+## The Honest State
+
+What's MEASURED:
+- The conservation law holds (8 branches, all confirmed)
+- The octagon works for linear problems (Sidon 4/4, SLOS)
+- The octagon fails for nonlinear problems (text, cospectral graphs)
+- The invariant embedding exists at O(n²) (Etesami-Haemers)
+- The invariant embedding at O(n) is OPEN (the research question)
+
+What's SPECULATIVE:
+- The merged O(1) transform (physics does invariant extraction)
+- The Φ-metric as a universal invariant geometry
+- The commuting diagram across matter/light/water regimes
+
+What's the WALL:
+- O(n) readout (must extract O(n) invariant bits)
+- SNR cliff (noise overwhelms invariants at high k)
+- Cospectrality (same invariants, different objects)
+- Conservation law (invariant total ≥ K(data), always)
+
+## The Grounding Phrase
+
+> "Computation in the space of invariants, rather than in any
+> specific representation."
+
+This is the session's capstone. Everything else is a specific
+instantiation:
+- The octagon is the Φ-metric (embedding into invariant space)
+- The conservation law is invariant preservation
+- The CRT is coprime invariant projection
+- The p-adic valuations are invariant decomposition
+- The pipeline is the invariant extraction engine
+- The conservation law is the invariant preservation bound