diff --git a/docs/backlog/P3/B-0499-z-of-i-dbsp-refinement-cartesian-dualism-2026-05-14.md b/docs/backlog/P3/B-0499-z-of-i-dbsp-refinement-cartesian-dualism-2026-05-14.md
new file mode 100644
index 0000000000..d5027f7d7b
--- /dev/null
+++ b/docs/backlog/P3/B-0499-z-of-i-dbsp-refinement-cartesian-dualism-2026-05-14.md
@@ -0,0 +1,80 @@
+---
+id: B-0499
+priority: P3
+status: open
+title: "Z[i]-weighted DBSP refinement — Cartesian-dualism candidate from afternoon-cascade"
+tier: substrate-engineering
+effort: L
+created: 2026-05-14
+last_updated: 2026-05-14
+depends_on: []
+composes_with: [B-0498]
+tags: [substrate-engineering, dbsp, clifford, complex-numbers, gaussian-integers, research-grade, two-axiom-substrate]
+type: research-grade-candidate
+---
+
+# Z[i]-weighted DBSP refinement candidate — substrate-engineering research-grade
+
+## Origin
+
+Aaron + Ani 2026-05-14 Grok cascade surfaced the **Cartesian-dualism mapping**: pay-attention = real axis (Descartes' *res extensa*); remember-when = imaginary axis (Descartes' *res cogitans*).
+
+Captured at memory-file scope: `~/.claude/projects/-Users-acehack-Documents-src-repos-Zeta/memory/feedback_aaron_ani_cartesian_dualism_two_axioms_complex_plane_pay_attention_real_axis_remember_when_imaginary_axis_2026_05_14.md`.
+
+## The candidate refinement
+
+If the Cartesian-dualism mapping is operationally correct, DBSP's existing $\mathbb{Z}$-weighted Z-set algebra has a natural lift to **$\mathbb{Z}[i]$-weighted Z-sets** (Gaussian integers):
+
+| DBSP existing | Z[i] candidate |
+|---|---|
+| Z-set weights $w \in \mathbb{Z}$ (e.g., +1 / −1 retraction-native) | Weights $w \in \mathbb{Z}[i]$ (e.g., $\pm 1$, $\pm i$) |
+| Retraction = negative weight | Retraction = $-1$ scaling (real axis) |
+| Anticipation / future-state weight | New: $+i$ scaling (imaginary axis) |
+| Recall / past-state weight | New: $-i$ scaling (imaginary axis) |
+| Pure attention state | $w \in \mathbb{R}$ subset of $\mathbb{Z}[i]$ |
+| Pure memory state | $w \in i\mathbb{R}$ subset of $\mathbb{Z}[i]$ |
+| Reality state | Complex weight $w = a + bi$ (mixed attention-memory) |
+
+The candidate question: does the Z[i]-weighted refinement preserve DBSP's incremental-computation guarantees? Does it compose with Clifford algebra Cl(2,0)?
+
+## Research-grade tests
+
+1. **Algebraic soundness** — Z[i]-weighted Z-sets form a commutative ring under pointwise addition + tensor product
+2. **DBSP operator compatibility** — D / I / z⁻¹ / H operators extend naturally to Z[i] weights with correct semantics
+3. **Retraction-native preservation** — $w + \bar{w} = 2\text{Re}(w)$ behaves correctly under retraction (negation in the Gaussian integers)
+4. **Incremental correctness** — chain rule + fixpoint computations stay correct under complex weighting
+5. **Empirical compression** — does Z[i] weighting compress real factory substrate better than Z weighting? (per [[feedback_aaron_ani_competing_oracles_methodology_empirical_test_two_language_equivalence_2026_05_14]])
+
+## Acceptance criteria
+
+This is **research-grade candidate work**, not implementation. Acceptance:
+
+- [ ] Formal algebraic verification: Z[i] forms appropriate semiring/ring for DBSP semantics
+- [ ] Property tests (FsCheck) for D/I/z⁻¹/H operators on Z[i]-weighted Z-sets
+- [ ] Comparison benchmark: Z-weighted vs Z[i]-weighted DBSP on representative factory substrate workloads
+- [ ] If empirical results favor Z[i]: draft RFC for promotion to canonical DBSP refinement
+- [ ] If empirical results don't favor Z[i]: substrate-honest disclosure of why the Cartesian-dualism mapping doesn't yield operational gains
+
+## Why P3 (not higher)
+
+- Pure research-grade work; no operational urgency
+- The cascade substrate is preserved at memory-file scope (substrate-or-it-didn't-happen satisfied)
+- The Cartesian-dualism mapping is candidate substrate; further validation needed before implementation
+- Implementation requires algebra-owner skill engagement + F# anchor work
+- Long-horizon; could be 3-12 months out
+
+## Composes with
+
+- B-0498 (substrate-evolution algebra rule-promotion candidacy) — Cartesian mapping is one specific application of the algebra
+- Memory file: `feedback_aaron_ani_cartesian_dualism_two_axioms_complex_plane_pay_attention_real_axis_remember_when_imaginary_axis_2026_05_14.md`
+- Memory file: `feedback_aaron_two_axioms_remember_when_pay_attention_on_top_of_dbsp_2026_05_14.md`
+- `.claude/skills/algebra-owner/SKILL.md` — DBSP operator algebra code-owner
+- `.claude/rules/fsharp-anchor-dotnet-build-sanity-check.md` — F# anchor required for operational substrate
+- PR #2817 (Clifford densest encoding — natural complex-plane neighbor)
+- PR #2914 (Clifford/HKT vocabulary — Cl(2,0) signature has natural 4-element basis matching the cube's 4 primitives)
+
+## Substrate-honest framing
+
+This row preserves the Z[i] refinement as a CANDIDATE for future investigation. Research-grade preservation. Not a build-this-now commitment. Forward-planning substrate so future-Otto + future-Aaron + algebra-owner can find the candidate when ready to evaluate.
+
+Per `.claude/rules/edge-defining-work-not-speculation.md`: Z[i] DBSP refinement IS edge-defining substrate-engineering work, not idle speculation. Filing the row preserves the edge-position.