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	14945
title	Scale changed-topology slot assignment before Demo B
state	Open
labels	enhancementaiarchitectureperformance
assignees	neo-gpt
createdAt	7:57 AM
updatedAt	7:57 AM
githubUrl	https://github.com/neomjs/neo/issues/14945
author	neo-gpt
commentsCount	0
parentIssue	13158
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]

Scale changed-topology slot assignment before Demo B

Open Backlog/active-chunk-5 enhancementaiarchitectureperformance

neo-gpt commented on 7:57 AM

Context

PR #14925 makes the changed-topology reconciler correct and fail-closed, but its exact-head DockTopologyReconciler.assignSlots() implementation explores every skip/pair branch in the window-assignment space. Direct dense probes at 72da2822baad921c86404ae8a4d7be59fe748783 measured approximately 431 ms for 8×8 and 5.4 s for 9×9. That cost is not a merge-safety defect while there is no runtime consumer; it becomes one before Demo-B scene G4 consumes the result.

Live latest-open sweep: checked the latest 20 open issues plus GitHub search, repository issue/discussion artifacts, Knowledge Base tickets, and the latest 30 A2A messages at 2026-07-10T05:57:03.906Z; no equivalent successor was found.

The Problem

The current exhaustive search is a useful correctness oracle, but its combinatorial growth makes ordinary multi-window restore latency unpredictable. Replacing it carelessly would reintroduce the greedy trap already disproved during review or weaken the deterministic tie contract that makes topology restore permutation-stable.

The Architectural Reality

src/dashboard/DockTopologyReconciler.mjs owns a pure Body-side assignment boundary. Its ordered objective is part of the contract:

maximize mapped-slot cardinality;
maximize summed Jaccard affinity;
maximize summed structural affinity;
choose the lexicographically smallest content signature;
use live-index order only for content-identical candidates.

The solver must remain id-free, preserve workspace-global uniqueness and captured-item conservation, create no windows, and leave the planner/reconciliation result vocabulary unchanged. ADR 0029 §2.2 and #14668 remain the authorities; #14590 is the first planned consumer.

The Fix

Replace the production exhaustive traversal with a polynomial/scalable assignment solver that preserves the full ordered objective. Keep a bounded form of the current exhaustive algorithm in tests as an independent oracle for small matrices. Add randomized equivalence and permutation probes, plus a reproducible dense-topology benchmark that demonstrates the runtime curve before the Demo-B integration is enabled.

Contract Ledger

Target Surface	Source of Authority	Proposed Behavior	Fallback	Docs	Evidence
`DockTopologyReconciler.assignSlots(capturedDocs, liveDocs)`	ADR 0029 §2.2 + #14668	Same mapping/unmapped/unmatchedLive contract and exact objective ordering, implemented with a documented polynomial bound	Existing fail-closed validation and reason classes remain unchanged	Update method JSDoc with the chosen algorithm and complexity	Exhaustive-oracle equivalence, permutation properties, focused dashboard specs, dense benchmark

Decision Record impact: aligned-with ADR 0029.

Acceptance Criteria

Production assignment no longer enumerates the full skip/pair search tree; the chosen algorithm and worst-case complexity are documented.
For bounded randomized matrices, the new solver is output-equivalent to the exhaustive oracle across cardinality, both affinity sums, content-signature ties, and content-identical live-index ties.
Captured/live input permutations preserve the content-canonical result established by #14668.
Existing uniqueness, conservation, reason-class, pass-through, and no-window-creation specs remain green.
A reproducible dense 8×8 / 9×9 / 12×12 benchmark is recorded, demonstrating that the 9×9 case no longer incurs multi-second assignment latency.
Demo-B scene G4 does not enable the reconciler as a runtime path until this ticket lands.
Cross-family review.

Out of Scope

Changing affinity semantics; adding IDs; changing restore result shapes; UI work; window spawning; weakening validation; broad dashboard refactors.

Avoided Traps

Do not restore the greedy captured-order algorithm; the existing greedy-trap spec is authoritative.
Do not gain speed by dropping deterministic content ties or workspace-global invariants.
Do not turn a benchmark into a hardware-fragile wall-clock CI gate; correctness properties remain the merge gate and benchmark evidence records the scale curve.
Do not broaden this into Demo-B UI delivery.

Parent: #13158
Related: #14668 #14925 #14590

Origin Session ID: de713f27-0e82-4960-b4c6-f281e0c36449

Retrieval Hint: "DockTopologyReconciler exhaustive assignment 8x8 9x9 polynomial solver content-stable tie conservation"