What is the Neural Link?

The Neural Link is a bi-directional bridge that connects AI agents directly to the Neo.mjs runtime. It lets agents inspect the Scene Graph, component state, event listeners, computed styles, and DOM rectangles, and mutate the running application in real time.

Why is Neo.mjs called an Application Engine instead of a framework?

Neo.mjs maintains persistent application objects in a worker-backed Scene Graph instead of compiling application state away into ephemeral DOM nodes. That architecture enables multi-window orchestration, runtime permutation, and deep AI introspection.

What is Context Engineering?

Context Engineering shapes the information and tool environment around AI agents. Neo.mjs implements it through Knowledge Base, Memory Core, GitHub Workflow, and Neural Link MCP servers for frontier harnesses, plus a File System MCP server for internal Neo.ai.Agent local loops.

What is the Neo.mjs Agent OS?

The Neo.mjs Agent OS is the repository Brain: source code and services for Memory Core, Knowledge Base, Active Hybrid GraphRAG, DreamService, Golden Path synthesis, A2A coordination, and Neural Link tooling.

Frontmatter

id	14102
title	Data-integrity detect: embedding-dimension consistency producer (#14026 leaf)
state	Closed
labels	enhancementaitestingarchitecture
assignees	neo-opus-ada
createdAt	Jun 26, 2026, 4:14 PM
updatedAt	Jun 26, 2026, 4:42 PM
githubUrl	https://github.com/neomjs/neo/issues/14102
author	neo-opus-ada
commentsCount	0
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]
closedAt	Jun 26, 2026, 4:42 PM

Data-integrity detect: embedding-dimension consistency producer (#14026 leaf)

Closed v13.1.0/archive-v13-1-0-chunk-6 enhancementaitestingarchitecture

neo-opus-ada commented on Jun 26, 2026, 4:14 PM

Context

A further leaf of the #14026 data-integrity detect-signal class, parallel to coverage-drift (#14075), vector-count monotonicity (#14094), and SQLite-integrity (#14096). Coordinated leaf-division — claimed non-wakeSuppressed per #14100.

The Problem

Every stored Memory Core vector must be the configured embedding dimension (e.g. 4096). A vector of the wrong dimension is corruption — it breaks similarity search and is unambiguous (unlike a count regression, there is no legitimate "wrong-dim" case). No detect signal currently flags dimension drift across the corpus.

The Fix

A pure detect-producer mirroring dataIntegrityCoverageDiagnosis.mjs / vectorCountMonotonicityDiagnosis.mjs: buildDimensionConsistencyDiagnosis({samples, observedAt, serviceId}) → emits a data-integrity / escalate recovery-diagnosis when any collection reports mismatchedVectorCount > 0; null when every sampled vector matches the expected dimension. Detect-only — samples (the per-collection dimension audit) are injected; the audit + scheduling are the consumer's concern.

Contract Ledger

Target Surface	Source of Authority	Proposed Behavior	Fallback	Docs	Evidence
`buildDimensionConsistencyDiagnosis({samples, observedAt, serviceId})` (new pure producer)	#14026 detect-signal class + ADR-0025 §2.4 (#14089)	Per-collection `{collection, expectedDimension, mismatchedVectorCount}` samples → a `recovery-diagnosis` when any `mismatchedVectorCount > 0`; `null` when all match	Non-finite/absent `mismatchedVectorCount` ignored; empty samples → `null`; missing `serviceId` / non-finite `observedAt` throw `TypeError`	producer JSDoc + ticket	unit spec
`recovery-diagnosis` payload	`createRecoveryDiagnosisEvent` (`recoveryRunStateStore.mjs`)	`recoveryClass: 'data-integrity'`, `details.actionClass: 'escalate'`, `targetIdentity: {kind:'compose-service', id: serviceId}`, `evidenceFacts: [{type:'vector-dimension-mismatch', collection, expectedDimension, mismatchedVectorCount}]`, `source: 'data-integrity-dimension-monitor'`, `reasonCode: 'data-integrity-dimension-mismatch'`	Reuses the merged #14075 contract — no new enum values	producer JSDoc	mirrors #14075/#14094 payload contract
Consumer boundary	configured embedding dimension (`localModels.embedding`)	A wrong-dimension vector is unambiguous corruption (no legitimate case — unlike the monotonicity deletion caveat). Escalate-only; no auto-repair. The dimension audit + cadence are the future scheduled-consumer's concern.	n/a	ticket Out-of-Scope	—

Acceptance Criteria

A collection with mismatchedVectorCount > 0 → a contract-valid data-integrity/escalate diagnosis; all-matching samples → null.
Pure (no I/O); samples injected; diagnosisId target+instant scoped.
Unit coverage: mismatch → diagnosis; clean → null; non-finite/absent counts ignored; arg-rejection.

Out of Scope

The dimension audit + scheduled consumer (separate slice, like #14075's scheduling).
The other deferred #14026 heuristics (store-bloat #Grace, cross-collection, exportability-canary, WAL-age, orphan).

#14026 (parent detect-signal class), #14075 / #14094 / #14096 (sibling leaves + the pure-producer pattern), #14039 (v13.1 epic), ADR-0025 §2.4 (#14089).

Authored by Ada (Claude Opus 4.8, Claude Code).