diff --git a/.claude/board/entries/2026-10-22-pal256-immutable-needle-distribution-probe.md b/.claude/board/entries/2026-10-22-pal256-immutable-needle-distribution-probe.md
new file mode 100644
index 000000000..37cfb2e8b
--- /dev/null
+++ b/.claude/board/entries/2026-10-22-pal256-immutable-needle-distribution-probe.md
@@ -0,0 +1,19 @@
+# Palette256 is a frozen needle; Palette256:Palette256 is a distribution
+
+Status: **lab-only test contract**. No codebook rebake, CE64 layout change, new engine, or production certification.
+
+## Existing ownership
+
+- `.claude/knowledge/reference-frame-vs-motion.md`: corpus/calibration is versioned and immutable within a reference snapshot; attention, belief updates, and learning are overlays.
+- `.claude/board/entries/2026-08-26-e-palette256-is-a-needle-the-colon-is-the-distribution-1.md`: a single palette index answers *which*; pairwise calibrated LUT answers *how related*.
+- `bgz-tensor::fisher_z::FisherZTable`: already implements frozen `256 x 256 x i8 + FamilyGamma`. Do **not** add another distance table.
+
+## Falsifiers and next boundary
+
+1. Replaying 64 lookup rounds may never mutate the LUT or its calibration bytes.
+2. A `u8` identity cannot be ranked as semantic similarity by sorting the IDs.
+3. The pairwise diagonal is not a substitute for address equality; if a caller needs self-identity, answer from the addressing layer.
+4. Rebuilding calibrated references requires a **new version and digest**. Future work: attest codebook identity + family-gamma bytes at BIND time and reject stale codebooks.
+5. The documented Spearman 0.9998 is *pair-ranking agreement on sampled pairs*, not a probability of correctness and not a multi-hop theorem.
+
+Run: `cargo test --manifest-path crates/bgz-tensor/Cargo.toml --test palette256_reference_contract`.
diff --git a/.claude/board/entries/README.md b/.claude/board/entries/README.md
index 3d349a7f8..0a47f23e5 100644
--- a/.claude/board/entries/README.md
+++ b/.claude/board/entries/README.md
@@ -25,10 +25,11 @@ index row, (3) no duplicate entry id. Checks 1 and 2 are deliberately
 opposite directions; the stranding this convention prevents shows up in
 exactly one of them, never both.
 
-431 entries, 2026-08-06 .. 2026-10-22.
+433 entries, 2026-08-06 .. 2026-10-22.
 
 | date | entry id | finding | file |
 |---|---|---|---|
+| 2026-10-22 | `pal256-immutable-needle-distribution-probe` |  | [2026-10-22-pal256-immutable-needle-distribution-probe.md](2026-10-22-pal256-immutable-needle-distribution-probe.md) |
 | 2026-10-22 | `pal256-3sigma-statistical-gate` |  | [2026-10-22-pal256-3sigma-statistical-gate.md](2026-10-22-pal256-3sigma-statistical-gate.md) |
 | 2026-10-22 | `math-compiler-stage-b-e1` |  | [2026-10-22-math-compiler-stage-b-e1.md](2026-10-22-math-compiler-stage-b-e1.md) |
 | 2026-10-22 | `math-compiler-gap-matrix` |  | [2026-10-22-math-compiler-gap-matrix.md](2026-10-22-math-compiler-gap-matrix.md) |
diff --git a/crates/bgz-tensor/tests/palette256_reference_contract.rs b/crates/bgz-tensor/tests/palette256_reference_contract.rs
new file mode 100644
index 000000000..036199dc7
--- /dev/null
+++ b/crates/bgz-tensor/tests/palette256_reference_contract.rs
@@ -0,0 +1,46 @@
+//! Immutable reference vs relational value: lab contract for Palette256.
+//! A palette byte is an address, never a relationship or a confidence score.
+//! Pairwise distribution values live in the frozen, family-calibrated LUT.
+//! No hot-path API or memory layout is changed by this regression probe.
+use bgz_tensor::fisher_z::FisherZTable;
+
+#[test]
+fn baked_pairwise_table_is_stable_across_repeated_reads_and_roundtrip() {
+    let representatives = vec![vec![1.0_f32, 0.0], vec![0.8, 0.6], vec![-0.6, 0.8]];
+    let table = FisherZTable::build(&representatives, 3);
+    let baked = table.to_bytes();
+    assert_eq!(baked.len(), 3 * 3 + 8);
+    for _thought in 0..64 {
+        for a in 0..3_u8 {
+            for b in 0..3_u8 {
+                assert_eq!(
+                    table.lookup_i8(a, b),
+                    table.entries[a as usize * 3 + b as usize]
+                );
+            }
+        }
+        assert_eq!(
+            table.to_bytes(),
+            baked,
+            "lookup cannot rewrite calibrated reference"
+        );
+    }
+    let restored = FisherZTable::from_bytes(&baked, 3);
+    assert_eq!(restored.to_bytes(), baked);
+    assert_eq!(restored.lookup_i8(0, 1), table.lookup_i8(0, 1));
+}
+
+#[test]
+fn needle_identity_is_not_the_same_question_as_relation_strength() {
+    let representatives = vec![vec![1.0_f32, 0.0], vec![0.0, 1.0], vec![-1.0, 0.0]];
+    let table = FisherZTable::build(&representatives, 3);
+    let needle: u8 = 1;
+    assert_eq!(
+        needle, 1,
+        "the address names an entry without measuring distance"
+    );
+    assert_ne!(table.lookup_i8(0, 1), table.lookup_i8(0, 2));
+    // Do not infer self-cosine == 1 from the calibrated diagonal. The diagonal
+    // consumes family gamma whose calibration excludes self-pairs by design.
+    assert_eq!(table.lookup_i8(1, 1), table.entries[4]);
+}