feat(arm-discovery): D-ARM-14 Phase 1 — splat-top-k oracle + DOLCE skeleton projector#438
Merged
Merged
Conversation
…eleton projector The two aerial-side seams the Wikidata-HHTL pipeline needs, verifiable standalone (the real splat + ontology producers stay in heavy workspace crates / jc, which the zero-dep aerial crate can't build against — so they're documented as the consumer-side / offline step). - aerial::TopKDistance: a CodebookDistance backed by a sparse per-node top-k neighbour list — the shape the 10000^2 BLASGraph Gaussian-splat actually emits (keep top-k per node, certified by jc EWA-sandwich; not a dense dim^2 table). Symmetric, nearest-on-duplicate, bounds-checked. - FeatureSpec::checked_slot: shared bounds-checked item->slot mapping. - aerial::ontology: DolceCategory (Endurant/Perdurant/Quality/Abstract + basin nibble + IRI — the HHTL axis template) + OntologyProjector (FeedProjector -> rdfs:subClassOf / rdf:type DOLCE skeleton SPO). End-to-end test: a splat top-k makes each occupation near its DOLCE class; aerial discovers the occupation->class skeleton edge; the projector renders `wd:f0_0 rdfs:subClassOf dolce:Endurant`; an unused facet is never invented. Float stays OFFLINE in jc only (ewa_sandwich + sigma_codebook_probe rho=0.9973 + pflug Le); aerial's online path is integer (the CAM-PQ doctrine end-to-end). Verified: 41/41 tests (37 + 4 new) + clippy -D warnings clean on both default and --features ndarray-simd. Zero-dep preserved. Remaining (documented): real jc/blasgraph splat producing the lists; Wikidata loader; gated on D-ARM-7 Jirak floor (jc::jirak). Map: .claude/knowledge/splat-codebook-aerial-wikidata-compression.md https://claude.ai/code/session_012SorR8UbtEvYmbX8cXftj7
📝 WalkthroughWalkthroughThis PR implements Phase 1 of the D-ARM-14 feature, introducing the aerial-side semantic discovery infrastructure: a DOLCE ontology projector that renders discovered entity-class rules into skeleton triples, a sparse top-k distance oracle for codebook neighbor lookups, and supporting encoding validation with full test coverage. ChangesD-ARM-14 Phase 1 Implementation
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
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: bd0e832ae6
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
+135
to
+138
| consequent | ||
| .first() | ||
| .and_then(|it| self.class_iris.get(it.category as usize)) | ||
| .cloned() |
There was a problem hiding this comment.
When projecting an unfiltered extract_rules result, this maps the object using only the consequent category and ignores which feature produced it. extract_rules emits candidates for every non-antecedent feature, and the new top-k oracle symmetrizes occupation↔class edges, so the same discovery pass can include reverse or other-feature rules whose category happens to be 0..3; those would be serialized as DOLCE facets even though their consequent is not the class axis, producing false rdfs:subClassOf/rdf:type skeleton triples. The projector needs to carry/validate the class feature (or the caller-facing API needs to filter before projection) instead of treating every consequent category as a class index.
Useful? React with 👍 / 👎.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
🧹 Nitpick comments (2)
.claude/knowledge/splat-codebook-aerial-wikidata-compression.md (1)
97-98: 💤 Low valueConsider aligning "SHIPPED" terminology with STATUS_BOARD conventions.
Line 97 uses "SHIPPED" for Phase 1, but according to STATUS_BOARD's legend (line 17), "Shipped" specifically means "Merged to main." The work is on branch
claude/jolly-cori-clnf9-darm14(not merged), so STATUS_BOARD correctly shows "In progress."The context here makes it clear ("D-ARM-14, branch
..."), so there's no ambiguity in practice. However, for consistency across board docs, consider:- **Phase 1 DONE** (D-ARM-14, branch `claude/jolly-cori-clnf9-darm14`): ...or
- **Phase 1 LANDED** (D-ARM-14, branch `claude/jolly-cori-clnf9-darm14`): ...This matches the AGENT_LOG's "Phase 1 DONE" and avoids overloading "SHIPPED."
The scoped-conjecture revision on line 98 is good—it honestly distinguishes "seams exist" from "full pipeline built."
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In @.claude/knowledge/splat-codebook-aerial-wikidata-compression.md around lines 97 - 98, Update the Phase 1 status wording to match STATUS_BOARD conventions: replace the header text "Phase 1 SHIPPED" (the line that references D-ARM-14 and branch `claude/jolly-cori-clnf9-darm14`) with either "Phase 1 DONE" or "Phase 1 LANDED" so it aligns with the legend where "Shipped" means merged to main; ensure the surrounding sentence still notes the branch and that the full pipeline is not yet built (the existing scoped-conjecture line can remain)..claude/board/STATUS_BOARD.md (1)
638-638: 💤 Low valueConsider condensing the "In progress" description to match STATUS_BOARD conventions.
The updated D-ARM-14 status is accurate and consistent with the AGENT_LOG, but it's unusually verbose compared to other "In progress" entries in this file. Most follow a terse pattern (e.g., line 540:
**In progress** | branch \...``), while this entry includes ~180 words of technical detail.Suggestion: move the detailed Phase 1 description to the AGENT_LOG entry only (already done) and keep this cell brief:
**In progress (Phase 1)** | branch `claude/jolly-cori-clnf9-darm14`; TopKDistance + DOLCE ontology projector seams landed; remaining: splat list production, Wikidata loader; gates on D-ARM-7The current version is not wrong—just denser than the established pattern. The AGENT_LOG is the canonical place for detailed narrative.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In @.claude/board/STATUS_BOARD.md at line 638, Condense the D-ARM-14 "In progress" cell in STATUS_BOARD.md to match the terse pattern used elsewhere: replace the long Phase 1 narrative with a short line like "**In progress (Phase 1)** | branch `claude/jolly-cori-clnf9-darm14`; TopKDistance + DOLCE ontology projector seams landed; remaining: splat list production, Wikidata loader; gated on D-ARM-7". Move the detailed Phase 1 text (the parts mentioning aerial::TopKDistance, the sparse splat top-k, aerial::ontology::{DolceCategory, OntologyProjector}, wikidata-hhtl-load.md, CAM-dedup, etc.) into the AGENT_LOG entry so the STATUS_BOARD cell stays concise and consistent with other entries.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Nitpick comments:
In @.claude/board/STATUS_BOARD.md:
- Line 638: Condense the D-ARM-14 "In progress" cell in STATUS_BOARD.md to match
the terse pattern used elsewhere: replace the long Phase 1 narrative with a
short line like "**In progress (Phase 1)** | branch
`claude/jolly-cori-clnf9-darm14`; TopKDistance + DOLCE ontology projector seams
landed; remaining: splat list production, Wikidata loader; gated on D-ARM-7".
Move the detailed Phase 1 text (the parts mentioning aerial::TopKDistance, the
sparse splat top-k, aerial::ontology::{DolceCategory, OntologyProjector},
wikidata-hhtl-load.md, CAM-dedup, etc.) into the AGENT_LOG entry so the
STATUS_BOARD cell stays concise and consistent with other entries.
In @.claude/knowledge/splat-codebook-aerial-wikidata-compression.md:
- Around line 97-98: Update the Phase 1 status wording to match STATUS_BOARD
conventions: replace the header text "Phase 1 SHIPPED" (the line that references
D-ARM-14 and branch `claude/jolly-cori-clnf9-darm14`) with either "Phase 1 DONE"
or "Phase 1 LANDED" so it aligns with the legend where "Shipped" means merged to
main; ensure the surrounding sentence still notes the branch and that the full
pipeline is not yet built (the existing scoped-conjecture line can remain).
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: 24489581-507e-418d-ac26-1131b663d997
📒 Files selected for processing (8)
.claude/board/AGENT_LOG.md.claude/board/STATUS_BOARD.md.claude/knowledge/splat-codebook-aerial-wikidata-compression.mdcrates/lance-graph-arm-discovery/src/aerial/codebook.rscrates/lance-graph-arm-discovery/src/aerial/mod.rscrates/lance-graph-arm-discovery/src/aerial/ontology.rscrates/lance-graph-arm-discovery/src/encode.rscrates/lance-graph-arm-discovery/src/lib.rs
AdaWorldAPI
pushed a commit
that referenced
this pull request
May 30, 2026
… + auto-resolve
User instruction: "create the integration plan as a list of possibilities,
then use the council and brutally honest review to recalibrate, then continue
autoattended autonomous decision making and auto resolve."
Composed `post-438-integration-options-v1.md` with 8 options + 6 combinations.
Spawned 4 council reviewers in parallel (Opus, single main-thread turn) with
identical brief but diverse lenses:
R1 (architectural-fit) — B+C conditional on N1 class_id paired pass + u16
R2 (prior-art / drift) — B+H; REJECTS C as re-litigating user-owned forks;
flags AriGraph hot↔cold bridge as the missing
integration target the options doc never named
R3 (integration-coordination) — B+D; REJECTS C — in-flight collision with #439
on lance-graph-contract (31 commits, unstable,
KanbanMove const _ ≤16B size assertion)
R4 (brutal-critic / bias-hunter)— B+G; REJECTS C as ego-shipping; tier-set spec
disagreement = SPEC FREEZE, not Claude-session;
cites session pattern of 3 prior hallucinations
Auto-resolve per options-doc §5:
- B unanimous (4/4) → EXECUTED (2 stale-citation fixes this commit)
- C: 3/4 reject → NOT EXECUTED
- OD-1/2/3 + canonical-spec-self-disagreement → escalated to ISSUES.md as
explicit SPEC-OWNER decisions (R2+R4 unanimous framing)
- R2's AriGraph-bridge finding → captured as EPIPHANY (new, missed by all
prior session work): the discovery_origin byte's natural home is a column
on the AriGraph hot↔cold bridge, not the mailbox-SoA byte
- G (chess bring-up): 2/4 endorse → queued to STATUS_BOARD as D-CHESS-BRINGUP-1
(NOT executed on this branch per R1 "needs its own branch + freeze-decision
authority"); unblocked by #436's Rust Aerial+ (user-flagged 2026-05-30)
- D (help #439): 1/4 → below threshold, not executed
- H (cargo clean, 3.3G): flagged for user, not auto-executed (workflow not
architecture)
Fixed stale citations in own docs:
- discovery-origin-provenance-reconciliation §8: specs are on main now
(via main commits d1635db, 93ac046, a16d0f4, 45276eb), not on a
separate cognitive-risc-core branch as v1 of the doc claimed pre-rebase
- odoo-blueprint-survival-dossier §7 "Concrete next moves": Wave 1 went
D-ARM-13/14 via #436/#438, not D-ARM-1/2 as originally planned
Documentation only. No code or in-flight contract modified. Contract carrier
work (D-ARM-1/D-ARM-2/D-MBX-A6-P3) explicitly held until the spec owner
reconciles the canonical tier-set disagreement and the user picks
proposer-id width (OD-1).
https://claude.ai/code/session_017gZ6sPRXYPj5n7uJ7NBtRv
AdaWorldAPI
pushed a commit
that referenced
this pull request
May 30, 2026
… + auto-resolve
User instruction: "create the integration plan as a list of possibilities,
then use the council and brutally honest review to recalibrate, then continue
autoattended autonomous decision making and auto resolve."
Composed `post-438-integration-options-v1.md` with 8 options + 6 combinations.
Spawned 4 council reviewers in parallel (Opus, single main-thread turn) with
identical brief but diverse lenses:
R1 (architectural-fit) — B+C conditional on N1 class_id paired pass + u16
R2 (prior-art / drift) — B+H; REJECTS C as re-litigating user-owned forks;
flags AriGraph hot↔cold bridge as the missing
integration target the options doc never named
R3 (integration-coordination) — B+D; REJECTS C — in-flight collision with #439
on lance-graph-contract (31 commits, unstable,
KanbanMove const _ ≤16B size assertion)
R4 (brutal-critic / bias-hunter)— B+G; REJECTS C as ego-shipping; tier-set spec
disagreement = SPEC FREEZE, not Claude-session;
cites session pattern of 3 prior hallucinations
Auto-resolve per options-doc §5:
- B unanimous (4/4) → EXECUTED (2 stale-citation fixes this commit)
- C: 3/4 reject → NOT EXECUTED
- OD-1/2/3 + canonical-spec-self-disagreement → escalated to ISSUES.md as
explicit SPEC-OWNER decisions (R2+R4 unanimous framing)
- R2's AriGraph-bridge finding → captured as EPIPHANY (new, missed by all
prior session work): the discovery_origin byte's natural home is a column
on the AriGraph hot↔cold bridge, not the mailbox-SoA byte
- G (chess bring-up): 2/4 endorse → queued to STATUS_BOARD as D-CHESS-BRINGUP-1
(NOT executed on this branch per R1 "needs its own branch + freeze-decision
authority"); unblocked by #436's Rust Aerial+ (user-flagged 2026-05-30)
- D (help #439): 1/4 → below threshold, not executed
- H (cargo clean, 3.3G): flagged for user, not auto-executed (workflow not
architecture)
Fixed stale citations in own docs:
- discovery-origin-provenance-reconciliation §8: specs are on main now
(via main commits d1635db, 93ac046, a16d0f4, 45276eb), not on a
separate cognitive-risc-core branch as v1 of the doc claimed pre-rebase
- odoo-blueprint-survival-dossier §7 "Concrete next moves": Wave 1 went
D-ARM-13/14 via #436/#438, not D-ARM-1/2 as originally planned
Documentation only. No code or in-flight contract modified. Contract carrier
work (D-ARM-1/D-ARM-2/D-MBX-A6-P3) explicitly held until the spec owner
reconciles the canonical tier-set disagreement and the user picks
proposer-id width (OD-1).
https://claude.ai/code/session_017gZ6sPRXYPj5n7uJ7NBtRv
AdaWorldAPI
pushed a commit
that referenced
this pull request
May 31, 2026
…FieldMask::inherit (Wikidata-HHTL slice 1) The downstream 16^n Abstammung bucket router #438 (D-ARM-14 P1) names but did not build: subClassOf nibble path, bit-shift O(1) addressing (root/child/basin/parent/is_ancestor_of). DOLCE-agnostic by construction (basin: u8 = dolce_id 0..3 resolved THROUGH the ontology cache per OD-DOLCE b31464d — never an embedded enum; matches both #441 dolce_id and #438 basin() orderings). + FieldMask::inherit (mask-inherits-as-delta; multi-parent = orthogonal facet bit in the same mask, NOT a 2nd path). Reuses #441 FieldMask; zero-dep preserved. The N4 second-domain falsifier for the D-CLS machinery: the Wikidata D-CLS triple (class_id, shape_hash, presence_bitmask) = (ClassId, StructuralSignature, FieldMask) — all #441. Convergent with D-ARM-14, firewall preserved (aerial = zero-dep proposer; contract/ontology = hub). 4 teeth-tests (bit-shift round-trip, depth cap + out-of-range nibble ignored, prefix reachability, multi-parent-is-a-facet-bit); 501 contract lib green; clippy -D warnings + fmt clean. Board: FINDING D-CLS<->D-ARM-14 reconciliation (EPIPHANIES) + D-WIKI-HHTL-1 row. https://claude.ai/code/session_01R9AWgFa65uPnLyS2my2d2R
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.
Summary
Phase 1 of D-ARM-14 — wire the certified jc splat codebook into
lance-graph-arm-discovery(Aerial+) so it can discover the OWL/DOLCE SPO skeleton that drives thewikidata-hhtl-load.mddeterministic compression.This PR lands the two aerial-side seams the pipeline needs, verifiable standalone. The heavy producers — the blasgraph Gaussian-splat top-k and the
lance-graph-ontologyDOLCE hydrators — live in workspace crates the zero-dep aerial crate can't build against, andjc::ewa_sandwichis a proof, not a table-builder. So they stay the documented consumer-side / offline step; this PR builds exactly what plugs into them.What's in it
aerial::TopKDistanceCodebookDistancebacked by a sparse per-node top-k neighbour list — the shape the 10000² BLASGraph Gaussian-splat actually emits (keep top-k per node, certified by jc's EWA-sandwich Σ-push-forward; not a densedim²table). Symmetric, nearest-wins-on-duplicate, bounds-checked. This is the production oracle shape;MatrixDistanceremains the dense test/reference impl.FeatureSpec::checked_slotItem → slotmapping (fails fast instead of aliasing another feature's block).aerial::ontology::DolceCategory0x0..=0x3) + IRI — the HHTL axis template perogit-owl-dolce-ontology-compartments.md.aerial::ontology::OntologyProjectorFeedProjectorrendering discovered rules as thewikidata-hhtl-load.mdskeleton —rdfs:subClassOf(P279) /rdf:type(P31) SPO, DOLCE facets as the class axis.Verification
Load-bearing test (
ontology::tests::discovers_and_projects_dolce_skeleton): a splat top-k makes each occupation near its DOLCE class → aerial discovers theoccupation → classskeleton edge → the projector renderswd:f0_0 rdfs:subClassOf dolce:Endurant, and an unused facet is never invented. Crate stays zero-dep / standalone.Float discipline (preserved)
Float lives only OFFLINE in jc — the EWA-sandwich Σ-push-forward,
sigma_codebook_probe(ρ=0.9973),pflug(Lε-faithful HHTL). jc emits a frozen integer artifact; aerial's online path is all integers (the splat-top-k lists areu32, evidence countsu32). Build the codebook offline (float OK), address it online with integer codes — the CAM-PQ doctrine end-to-end. No new aerial dependency: the certified lists are passed in through theCodebookDistanceseam.Remaining (Phase 2+, documented)
jc::jirak) before any rule is promoted to a live skeleton.Full wiring map:
.claude/knowledge/splat-codebook-aerial-wikidata-compression.md(finding E-ARM-JC-RESOLVES-BOTH-SEAMS).Board hygiene (same commit)
STATUS_BOARD(D-ARM-14 → In progress, Phase 1) ·AGENT_LOGrun · the wiring knowledge doc's status section.https://claude.ai/code/session_012SorR8UbtEvYmbX8cXftj7
Generated by Claude Code
Summary by CodeRabbit
New Features
Documentation