feat(graph): Backend::MailboxSoa — classid node-match + CLAM/CAKES neighborhood (Inc 0)#544
Open
AdaWorldAPI wants to merge 9 commits into
Open
feat(graph): Backend::MailboxSoa — classid node-match + CLAM/CAKES neighborhood (Inc 0)#544AdaWorldAPI wants to merge 9 commits into
AdaWorldAPI wants to merge 9 commits into
Conversation
…(Inc 0 first slice) cypher-kanban-ast-unification-v1 Inc 0, the verified-safe half: the substrate IS the graph (E-GUID-IS-THE-GRAPH), so MATCH (n:Label) is a classid prefix-route over the zero-dep MailboxSoaView contract, resolved off the class column with zero value decode. - graph_router::Backend gains the MailboxSoa variant (the named router gap). - graph/mailbox_scan.rs: match_nodes_by_class (classid route; reads only the class column) + match_node_by_local_key (local_key->row via row_for_local_key, None-fallback to positional address). - Gates: parity (matched set == reference classid filter); F2 zero-value-decode proven structurally by a GuardedSoa whose energy()/meta_raw() panic on access; key-index point lookup. 4/4 green, no new clippy warnings. Edge-traversal ((a)-[r]->(b)) deliberately deferred, grounded not faked: CausalEdge64 (the edges_raw column) is an SPO triple, NOT a row->row adjacency pointer, and the View exposes no EdgeBlock adjacency accessor. That is the edge-representation boundary the 5+3 council said to pin first (verdict 4b); it lands as the next slice once the classid-resolved edge rep + adjacency accessor are added. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CcpLeEC3XK8Eye53GKBVvi
|
Warning Review limit reached
More reviews will be available in 55 minutes and 34 seconds. Learn how PR review limits work. Your organization has used up its prepaid credits, and credit purchases are no longer available. Enable the review add-on in the billing tab to keep reviews running — you're only billed for reviews past your plan's rate limits ($0.25/file). ⌛ How to resolve this issue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based credits. 🚦 How do rate limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan refill rate. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, the refill rate gradually slows as usage increases. The highest same-day bursts are limited more strictly. Please see our Fair Usage Limits Policy for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (1)
📝 WalkthroughWalkthroughAdds ChangesMailboxSoA Node-Match Routing and Zero-Decode Neighborhood Facets
Sequence Diagram(s)sequenceDiagram
participant CypherRouter
participant match_nodes_by_class
participant match_node_by_local_key
participant MailboxSoaView
rect rgba(100, 149, 237, 0.5)
Note over CypherRouter,MailboxSoaView: MATCH (n:Label) — class route
CypherRouter->>match_nodes_by_class: view, class_id
match_nodes_by_class->>MailboxSoaView: class_id() column only
MailboxSoaView-->>match_nodes_by_class: entity_type rows
match_nodes_by_class-->>CypherRouter: Vec<NodeMatch {row, Backend::MailboxSoa}>
end
rect rgba(144, 238, 144, 0.5)
Note over CypherRouter,MailboxSoaView: MATCH (n {key}) — local-key route
CypherRouter->>match_node_by_local_key: view, local_key
match_node_by_local_key->>MailboxSoaView: row_for_local_key(local_key)
MailboxSoaView-->>match_node_by_local_key: Option<usize>
match_node_by_local_key-->>CypherRouter: Option<NodeMatch>
end
rect rgba(255, 218, 185, 0.5)
Note over CypherRouter,MailboxSoaView: Neighborhood facets — zero-decode geometry
CypherRouter->>match_nodes_by_class: clam_contained(view, query)
match_nodes_by_class->>MailboxSoaView: hhtl_path_at(row) per candidate
MailboxSoaView-->>match_nodes_by_class: Option<NiblePath>
match_nodes_by_class-->>CypherRouter: Vec<NodeMatch> in CLAM subtree
end
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e6e2b3e1c4
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Comment on lines
+61
to
+64
| let classes = view.class_id(); | ||
| classes | ||
| .iter() | ||
| .enumerate() |
There was a problem hiding this comment.
This iterates the entire class_id() slice instead of the view's logical row count. That is fine for the test GuardedSoa, but the real in-memory MailboxSoA<N> reports n_rows() == populated while entity_type() returns the full backing array capacity, initialized to zeros (cognitive-shader-driver/src/mailbox_soa.rs documents this phantom-row guard around n_rows). In contexts using that view, MATCH can return unpopulated padding rows (for example an empty mailbox queried for class 0, or stale padding after the logical size shrinks), corrupting node-match results; please clamp the scan with view.n_rows()/take(view.n_rows()).
Useful? React with 👍 / 👎.
…rep boundary (§4b) Operator correction: adjacency lives in two places, classid/key-resolved, never query-guessed: 1. HHTL cascade in the GUID key = CLAM hierarchical neighborhood (NiblePath is_ancestor_of/prefix; graph/neighborhood/clam.rs) — free, zero value decode. 2. 16-byte EdgeBlock = explicit typed edges per EdgeCodecFlavor: CoarseOnly (16x8) = 12 in-family + 4 external; Pq32x4 (32x4) = turbovec residue edges. 3. edges_raw = CausalEdge64 = SPO causal arcs (separate facet). The class picks the rep (classid -> ClassView -> EdgeCodecFlavor). Unblocks the deferred edge half of #544: next slice exposes the HHTL/key + EdgeBlock per row on MailboxSoaView, then CLAM prefix-route + EdgeBlock slot-deref, both zero-value-decode. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CcpLeEC3XK8Eye53GKBVvi
…lix Signed360 is the exact orthogonal point, "where" is a decode ladder Operator correction: the helix is not more adjacency, it is the EXACT orthogonal LOCATION. Adjacency (HHTL/CLAM near, EdgeBlock connected) is relational; helix Signed360 (ValueTenant::HelixResidue, signed full-sphere golden-spiral, 6B) is the absolute exact coordinate. "Where" is a decode-cost ladder: 1. HHTL/CLAM containment - key prefix, zero value decode (which cluster). 2. Helix PLACE - deterministic from the address, zero value decode. 3. Helix RESIDUE - Signed360 6B in the value slab, one value-tenant decode (exact). Router consequence: proximity query = key (free); exact-position query = read the HelixResidue tenant (a value decode, costed as such). Grounded in canonical_node ValueTenant::HelixResidue + ValueSchema::Compressed. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CcpLeEC3XK8Eye53GKBVvi
…LFD, not a containment check; the full geometry-of-a-node surface Operator: ndarray also has chaoda. Grounded in ndarray/src/hpc/clam.rs (CAKES Alg1/4/6, panCAKES Alg2, CHAODA Phase 4 anomaly_scores from LFD) + perturbation- sim chaoda (CHAODA-lite, names ndarray ClamTree as production). The CLAM facet is the manifold engine: containment + CAKES ranked-NN (attraction) + CHAODA anomaly (repulsion) + panCAKES compression, one tree, LFD the shared measure. Synthesized geometry-of-a-node: off one GUID the substrate answers which-cluster (CLAM, free), nearest-similar (CAKES), how-anomalous (CHAODA), exact-location (helix Signed360, value decode), connected-to (EdgeBlock), caused-by (CausalEdge64) - a complete geometric+relational surface, each at its own decode cost; the router dispatches a query to the cheapest facet that answers it. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CcpLeEC3XK8Eye53GKBVvi
…ixes (Inc 0, the manifold facet) panCAKES == radix trie == HHTL (E-PANCAKES-IS-RADIX-IS-HHTL): the CLAM cluster tree IS the radix trie of the classid·HEEL·HIP·TWIG nibble paths already in the keys, so the structural neighborhood is pure prefix arithmetic, zero value decode. - NiblePath::common_prefix_depth (contract) — the radix-trie nearest-neighbor measure; longest-common-prefix = CAKES attraction. +1 unit test. - MailboxSoaView::hhtl_path_at (contract) — per-row HHTL NiblePath, deferred- binding default None (canon NodeRow already carries key(16); the override exposes what's there). - graph::mailbox_scan::clam_contained (is_ancestor_of = the radix subtree = CLAM cluster) + cakes_nearest (common_prefix_depth ranking, k-NN). Both key-only, zero value decode. - Tests: containment = radix subtree (rows 0,1,2 under 1·2; leaf narrows to 0); CAKES ranks by shared depth [(0,3),(1,2),(2,2)]; deferred-None yields nothing (coarser-facet fallback); F2 zero-value-decode extended to CLAM/CAKES (the GuardedSoa value columns still panic-guarded). 7/7 mailbox_scan + 21/21 hhtl green, clippy clean. This is the first dispatch-table facet beyond the classid node-match: proximity/ neighborhood resolves on the key (CLAM/CAKES), free, per E-CLAM-IS-THE-MANIFOLD- ENGINE. Edge-deref (EdgeBlock) + helix exact-location (value decode) are the next tiers. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CcpLeEC3XK8Eye53GKBVvi
AdaWorldAPI
changed the title
…ly/4-external slot decode (Inc 0)
The third dispatch-table facet: explicit typed edges (a)-[r]->... under the
classid-resolved EdgeCodecFlavor (E-ADJACENCY-IS-KEY-AND-EDGECODEC). EdgeBlock is
bytes 16..32 (the edge region), NOT the value slab, so still zero value decode.
- MailboxSoaView::edge_block_at(row) -> Option<EdgeBlock> (contract, deferred
default None; the canon NodeRow carries edges(16), the override exposes it).
- graph::mailbox_scan::{EdgeNeighbors, edge_slots_coarse}: under CoarseOnly,
decode the 12 in-family + 4 external slots to their populated (non-zero) refs,
family vs external. Pq32x4 (turbovec residue) / CoarseResidue are refused -
they are NOT adjacency, never coerced to slots (boundary 4b: classid-resolved,
not query-guessed).
- Slot-byte -> neighbor-row resolution is deliberately deferred (the basin-local-
index convention + zero-collision is the next encoding decision, analogous to
local_key->row); this facet lands the structure (which slots are edges, family
vs external, under which flavor), never fakes the row resolution.
- Tests: populated decode ([2,5] family + [1] external), all-zero = no edges,
no-block = None, non-Coarse flavors refused. F2 zero-value-decode extended.
9/9 mailbox_scan, clippy clean (sort_by_key + Reverse).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CcpLeEC3XK8Eye53GKBVvi
…anded facets CI fmt --check failed on three blocks (let-chain filter, EdgeNeighbors collect, cakes assert). Applied cargo fmt. Also refreshed the stale module doc header (it still described only the node-match half and called CLAM/EdgeBlock "deferred") to document the three landed key-resident facets + the genuinely deferred tiers (slot->row convention, helix/CHAODA/SPO costed tier). No logic change; 9/9 mailbox_scan green, fmt clean. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CcpLeEC3XK8Eye53GKBVvi
…switch 16K-from-an-angle compare is the costed value sweep, composing with the free key facets as a two-stage cascade "Switch tenant + compare across the 16K mailbox from an angle" decomposes as: batch Hamming sweep (hamming_top_k over a contiguous identity plane) = the right use of popcount on the homogeneous 16K fingerprint; "angle" = which plane (content/topic/angle) + query; "tenant switch" = column selector. Load-bearing: it composes with #544's free key facets as a two-stage HHTL cascade - CLAM/CAKES prefix prune (free, zero decode) then angle-Hamming rank over the pruned set (costed). Key prune + content rank = two halves of one query. Cost-class boundary: this facet deliberately decodes the value plane, NOT in the F2 zero-decode class, lands on its own branch with its own cost gate. Grounded in MailboxSoA content/topic/angle_row + ndarray hamming_top_k + cycle snapshot. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CcpLeEC3XK8Eye53GKBVvi
…p; IVF coarse quantizer IS the HHTL/CLAM prefix Operator: "sweep or just 90° fingerprint vector cam index" → CAM index. The value-side rank is a CAM-PQ ADC (distance-table lookups + IVF probe), never a linear Hamming sweep (that's only the no-index fallback). The load-bearing correction: CAM-PQ = IVF-PQ, and its IVF coarse quantizer IS the HHTL/CLAM prefix (turbovec: palette256 = coarse quantizer) while the PQ residual IS the turbovec Pq32x4 / value-slab codes. So #544's cakes_nearest prefix prune is literally the IVF coarse-quantization stage, not a prefilter on a scan. 90° = content/topic/angle orthogonal axes each get their own distance table; tenant switch = which orthogonal table to ADC against. Retitled the entry (was "…-SWEEP-IS-PRUNE-THEN-RANK"); fixed an accidental duplicate E-PANCAKES header. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CcpLeEC3XK8Eye53GKBVvi
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Inc 0 —
Backend::MailboxSoa: the substrate-is-the-graph dispatch table, two facets landedThe substrate IS the graph (
E-GUID-IS-THE-GRAPH); a query routes to the cheapest facet that answers it, off the GUID key, zero value decode. This PR lands the two key-only facets.Facet 1 — classid node-match (
MATCH (n:Label))match_nodes_by_class(view, class_id)— classid prefix-route; reads only the class column.match_node_by_local_key(view, local_key)— point lookup viaMailboxSoaView::row_for_local_key(None-fallback to positional).Facet 2 — CLAM/CAKES neighborhood (proximity) — panCAKES ≡ radix trie ≡ HHTL
The CLAM cluster tree isn't a separate structure — it is the radix trie of the
classid·HEEL·HIP·TWIGnibble paths already in the keys (E-PANCAKES-IS-RADIX-IS-HHTL). So neighborhood = pure prefix arithmetic:NiblePath::common_prefix_depth(contract) — the radix-trie NN measure (CAKES attraction = longest-common-prefix).MailboxSoaView::hhtl_path_at(contract) — per-row HHTL path, deferred-binding defaultNone(canonNodeRowalready carrieskey(16)).clam_contained(is_ancestor_of= the radix subtree = CLAM cluster) +cakes_nearest(common_prefix_depthk-NN). Zero value decode.Gates (tests, all green, clippy clean)
1·2; leaf query narrows to0);[(0,3),(1,2),(2,2)];NoneHHTL ⇒ scan yields nothing (coarser-facet fallback, never a wrong row);GuardedSoawhoseenergy()/meta_raw()panic — all facets complete without touching the 480 B value slab.mailbox_scan, 21/21hhtl, 674 contract green.Dispatch-table tiers still ahead (grounded, not faked)
(a)-[r:TYPE]->(b)): 12-family/4-external or 32×4 turbovec, perclassid → EdgeCodecFlavor(E-ADJACENCY-IS-KEY-AND-EDGECODEC).Signed360,HelixResiduetenant): the one tier that IS a value decode, costed as such (E-HELIX-IS-EXACT-LOCATION).E-CLAM-IS-THE-MANIFOLD-ENGINE).Additive, layout-preserving — no
NodeRow/stride/ENVELOPE_LAYOUT_VERSIONchange, no new object model (F5).🤖 Generated with Claude Code
Summary by CodeRabbit
Documentation
Chores