Update WORK_LOG.md - document past 5 minutes: integer spiral packing, f(n) layout, Cartan floor, torus winding, fingerprint resolution, 43/43 tests

WORK_LOG.md

@@ -61,3 +61,52 @@ Standalone formula for the hardest Layer 3 conjecture: a Cartan connection of ty
 ---
 
 ## Pre-2026-06-26
+
+
+---
+
+## 2026-07-02
+
+### Integer Spiral Packing Fix
+**Status:** RESOLVED
+- Upstream rewrote phi_corkscrew_index: packed signed coefficients without offset encoding
+- Bug: Not injective ((-1,1) and (1,0) both give 1 in base 2)
+- Fixed in upstream
+- Added canonical corpus-wide version to spectral_codebook.py
+- Base: 328,469 / Offset: 164,234 recorded in JSON header
+- Unpack inverts exactly
+- Round-trip verified over all 250 matrices
+
+### f(n) Layout (Decorative) - NEW
+**Status:** IMPLEMENTED
+- Each entry gets radius_sq (= spiral index) and angle_frac
+- Computed in decimal (indices reach ~10^44 where float64 keeps zero fractional bits)
+- Verified against VERIFICATION_LOG V007: f(20121) reproduces (-137.80079576, -33.64432624)
+- Min angular separation: ≈ 4.6×10^-6 of a turn
+
+### Cartan Δ-Floor Enhancement
+**Status:** IMPLEMENTED
+- Upstream only had 1-D ρ-snapping
+- Added Fisher version on Δ_7 with Δ = 17/1792 exact (CartanConnection.lean:70)
+- 13 fingerprint pairs fall below Δ
+- Merging 196 fingerprints into 187 Δ-resolution codewords vs 9 gap-rule clusters
+- Both rules emitted side by side
+- **Caveat:** |c|/Σ|c| projection destroys sign/scale
+- This corpus has two distinct fingerprints at Fisher distance exactly 0
+- Fisher layer is similarity, not identity
+
+### Torus Winding Improvement
+**Status:** IMPLEMENTED
+- Upstream's _safe variant only detected saturation
+- TorusWinding now carries a_exact/b_exact (plain ℤ)
+- torus_to_spiral_index prefers them
+- Round trip lossless
+- Regression-tested at n = 10^9
+- **Note:** Lean mirror in BraidEigensolid.lean needs same widening (noted in docstring, Lean untouched)
+
+### Fingerprint Discrepancy Resolution
+**Status:** RESOLVED
+- 192-vs-196 fingerprint discrepancy was stale docstring
+- Both methods agree coefficient-for-coefficient on all 250 matrices (196 unique)
+- 43/43 tests pass (19 new)
+- Docs extended with exact-vs-decorative labels per layer