|
| 1 | +# Addendum 04a: Merkle Root for Eineindeutigkeit Resolution |
| 2 | + |
| 3 | +## Add to: `src/spo/clam_path.rs` (Phase 5, currently building) |
| 4 | + |
| 5 | +## The Problem |
| 6 | + |
| 7 | +Three queries converge on the same concept — semantic, explicit, and relational |
| 8 | +all land in same σ-bands. That's Eineindeutigkeit (unique determination). |
| 9 | +But WHERE does the concept live? |
| 10 | + |
| 11 | +Without content-addressing, the address depends on which query found it first. |
| 12 | +Three paths → three ClamPath addresses → three Redis keys → **duplication**. |
| 13 | + |
| 14 | +## The Solution: ClamPath + MerkleRoot in word[0] |
| 15 | + |
| 16 | +``` |
| 17 | +word[0] of CogRecord8K (u64, 64 bits): |
| 18 | +┌────────────────────┬────────────────────────────────────┐ |
| 19 | +│ ClamPath (24 bits) │ MerkleRoot (40 bits) │ |
| 20 | +│ HOW you got here │ WHAT lives here │ |
| 21 | +│ navigation/lineage │ identity/canonical address │ |
| 22 | +└────────────────────┴────────────────────────────────────┘ |
| 23 | +
|
| 24 | +ClamPath = structural path through B-tree (depth + split decisions) |
| 25 | +MerkleRoot = truncated blake3 of content fingerprint → canonical identity |
| 26 | +``` |
| 27 | + |
| 28 | +## Why Both |
| 29 | + |
| 30 | +``` |
| 31 | +ClamPath alone: path-dependent → same concept gets different addresses |
| 32 | +MerkleRoot alone: no lineage, no subtree range queries, no causality chains |
| 33 | +Together: navigate via ClamPath, resolve identity via MerkleRoot |
| 34 | +``` |
| 35 | + |
| 36 | +## MerkleRoot Derivation |
| 37 | + |
| 38 | +The Merkle root is derived from the three-plane binary fingerprints. |
| 39 | +This is deterministic: same concept = same fingerprint = same root. |
| 40 | + |
| 41 | +```rust |
| 42 | +use blake3; |
| 43 | + |
| 44 | +/// Truncated Merkle root for content-addressed identity. |
| 45 | +/// 40 bits = ~1 trillion collision space. Sufficient for BindSpace. |
| 46 | +#[derive(Clone, Copy, PartialEq, Eq, Hash)] |
| 47 | +pub struct MerkleRoot(pub u64); // only lower 40 bits used |
| 48 | + |
| 49 | +impl MerkleRoot { |
| 50 | + /// Derive from three-plane binary fingerprints. |
| 51 | + /// The canonical address IS the content hash. |
| 52 | + pub fn from_planes( |
| 53 | + s_binary: &[u8; 2048], // 16384-bit S-plane |
| 54 | + p_binary: &[u8; 2048], // 16384-bit P-plane |
| 55 | + o_binary: &[u8; 2048], // 16384-bit O-plane |
| 56 | + ) -> Self { |
| 57 | + // Hash each plane separately, then combine. |
| 58 | + // This is a Merkle tree with 3 leaves. |
| 59 | + let s_hash = blake3::hash(s_binary); |
| 60 | + let p_hash = blake3::hash(p_binary); |
| 61 | + let o_hash = blake3::hash(o_binary); |
| 62 | + |
| 63 | + // Combine: hash(S || P || O) — order matters (S < P < O is canonical) |
| 64 | + let mut hasher = blake3::Hasher::new(); |
| 65 | + hasher.update(s_hash.as_bytes()); |
| 66 | + hasher.update(p_hash.as_bytes()); |
| 67 | + hasher.update(o_hash.as_bytes()); |
| 68 | + let root = hasher.finalize(); |
| 69 | + |
| 70 | + // Truncate to 40 bits |
| 71 | + let bytes = root.as_bytes(); |
| 72 | + let val = u64::from_le_bytes([ |
| 73 | + bytes[0], bytes[1], bytes[2], bytes[3], bytes[4], 0, 0, 0, |
| 74 | + ]); |
| 75 | + MerkleRoot(val & 0xFF_FFFF_FFFF) // mask to 40 bits |
| 76 | + } |
| 77 | + |
| 78 | + /// Derive from single composite fingerprint (backward compat). |
| 79 | + /// Used when planes aren't separated yet. |
| 80 | + pub fn from_fingerprint(fp: &[u8; 2048]) -> Self { |
| 81 | + let hash = blake3::hash(fp); |
| 82 | + let bytes = hash.as_bytes(); |
| 83 | + let val = u64::from_le_bytes([ |
| 84 | + bytes[0], bytes[1], bytes[2], bytes[3], bytes[4], 0, 0, 0, |
| 85 | + ]); |
| 86 | + MerkleRoot(val & 0xFF_FFFF_FFFF) |
| 87 | + } |
| 88 | + |
| 89 | + /// The Redis key for this concept's canonical address. |
| 90 | + pub fn redis_key(&self) -> String { |
| 91 | + format!("ada:bind:{:010x}", self.0) |
| 92 | + } |
| 93 | +} |
| 94 | +``` |
| 95 | + |
| 96 | +## Packing into word[0] |
| 97 | + |
| 98 | +```rust |
| 99 | +impl ClamPath { |
| 100 | + /// Pack ClamPath (24 bits) + MerkleRoot (40 bits) into a single u64. |
| 101 | + pub fn pack_with_merkle(&self, root: MerkleRoot) -> u64 { |
| 102 | + let clam_bits = self.to_u24() as u64; // 24 bits |
| 103 | + let merkle_bits = root.0 & 0xFF_FFFF_FFFF; // 40 bits |
| 104 | + (clam_bits << 40) | merkle_bits |
| 105 | + } |
| 106 | + |
| 107 | + /// Unpack from word[0]. |
| 108 | + pub fn unpack_with_merkle(word0: u64) -> (ClamPath, MerkleRoot) { |
| 109 | + let clam_bits = (word0 >> 40) as u32; // upper 24 bits |
| 110 | + let merkle_bits = word0 & 0xFF_FFFF_FFFF; // lower 40 bits |
| 111 | + (ClamPath::from_u24(clam_bits), MerkleRoot(merkle_bits)) |
| 112 | + } |
| 113 | +} |
| 114 | +``` |
| 115 | + |
| 116 | +## Self-Healing Redis Address |
| 117 | + |
| 118 | +This is the key insight for distributed operation: |
| 119 | + |
| 120 | +``` |
| 121 | +1. Concept arrives → three-plane fingerprints computed |
| 122 | +2. MerkleRoot derived deterministically: blake3(S || P || O) |
| 123 | +3. Redis key = ada:bind:{merkle_root_hex} |
| 124 | +4. ClamPath = structural location in B-tree |
| 125 | +
|
| 126 | +If Redis evicts the key: |
| 127 | + - Recompute MerkleRoot from fingerprints (deterministic) |
| 128 | + - Key is reconstructed: same content → same hash → same address |
| 129 | + - Tree heals itself |
| 130 | +
|
| 131 | +If two paths find the same concept: |
| 132 | + - Both compute same MerkleRoot (content-addressed) |
| 133 | + - Both write to same Redis key |
| 134 | + - Eineindeutigkeit: last-write-wins is fine because content is identical |
| 135 | + - ClamPaths may differ (different navigation routes) — that's OK, |
| 136 | + ClamPath is HOW, MerkleRoot is WHAT |
| 137 | +
|
| 138 | +If concept is modified (evidence updates the soaking register): |
| 139 | + - MerkleRoot changes (new content → new hash) |
| 140 | + - Old key naturally expires (TTL or eviction) |
| 141 | + - New key created at new address |
| 142 | + - No dangling references: anyone who had the old root will miss, |
| 143 | + recompute from current fingerprints, find the new address |
| 144 | +``` |
| 145 | + |
| 146 | +## Merkle Tree for Subtree Integrity (Future) |
| 147 | + |
| 148 | +When the hive goes multi-writer, extend the per-concept MerkleRoot |
| 149 | +to a full Merkle tree over the CLAM subtree: |
| 150 | + |
| 151 | +``` |
| 152 | + root_hash |
| 153 | + / \ |
| 154 | + left_hash right_hash |
| 155 | + / \ / \ |
| 156 | + leaf_0 leaf_1 leaf_2 leaf_3 |
| 157 | + (concept fingerprints) |
| 158 | +``` |
| 159 | + |
| 160 | +Each CLAM split level has a hash that summarizes its children. |
| 161 | +To verify a subtree after network partition: |
| 162 | + |
| 163 | +```rust |
| 164 | +/// Verify subtree integrity after reconnection. |
| 165 | +/// Compare roots — if they match, entire subtree is consistent. |
| 166 | +/// If they diverge, walk down to find the first differing leaf. |
| 167 | +pub fn verify_subtree( |
| 168 | + local_root: MerkleRoot, |
| 169 | + remote_root: MerkleRoot, |
| 170 | + clam_path: &ClamPath, |
| 171 | +) -> SubtreeStatus { |
| 172 | + if local_root == remote_root { |
| 173 | + SubtreeStatus::Consistent |
| 174 | + } else { |
| 175 | + // Walk the Merkle tree to find divergence point |
| 176 | + SubtreeStatus::Diverged { at_depth: /* compare level by level */ } |
| 177 | + } |
| 178 | +} |
| 179 | +``` |
| 180 | + |
| 181 | +But this is ice cake 21. For now: per-concept MerkleRoot in word[0] is sufficient. |
| 182 | + |
| 183 | +## Dependency |
| 184 | + |
| 185 | +```toml |
| 186 | +# Cargo.toml |
| 187 | +blake3 = "1" |
| 188 | +``` |
| 189 | + |
| 190 | +Blake3 is fast (~1GB/s on modern CPUs), deterministic, and the truncated |
| 191 | +40-bit output is sufficient for ~1 trillion address space. |
| 192 | + |
| 193 | +## Integration with Existing ClamPath |
| 194 | + |
| 195 | +The existing `04_btree_clam_path_lineage.md` spec defines ClamPath as u16 (16 bits) |
| 196 | +with depth in a separate u8. Repack as u24 to fit alongside MerkleRoot in word[0]: |
| 197 | + |
| 198 | +```rust |
| 199 | +impl ClamPath { |
| 200 | + /// Pack bits (u16) + depth (u8) into 24 bits. |
| 201 | + pub fn to_u24(&self) -> u32 { |
| 202 | + ((self.depth as u32) << 16) | (self.bits as u32) |
| 203 | + } |
| 204 | + |
| 205 | + /// Unpack from 24 bits. |
| 206 | + pub fn from_u24(packed: u32) -> Self { |
| 207 | + ClamPath { |
| 208 | + bits: (packed & 0xFFFF) as u16, |
| 209 | + depth: ((packed >> 16) & 0xFF) as u8, |
| 210 | + } |
| 211 | + } |
| 212 | +} |
| 213 | +``` |
| 214 | + |
| 215 | +## Summary |
| 216 | + |
| 217 | +| Component | Bits | Purpose | |
| 218 | +|-----------|------|---------| |
| 219 | +| ClamPath.bits | 16 | B-tree split decisions (navigation) | |
| 220 | +| ClamPath.depth | 8 | Valid bit count (confidence/resolution) | |
| 221 | +| MerkleRoot | 40 | Content-addressed canonical identity | |
| 222 | +| **Total** | **64** | **= word[0] of CogRecord8K** | |
| 223 | + |
| 224 | +ClamPath = HOW you got here. |
| 225 | +MerkleRoot = WHAT lives here. |
| 226 | +Redis key = WHERE it's stored. |
| 227 | +Self-healing = WHY it works distributed. |
0 commit comments