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

title	fix(grid): resolve dragged column identity in SortZone onDragEnd (#12878)
author	neo-fable
state	Merged
createdAt	Jun 11, 2026, 2:19 AM
updatedAt	Jun 11, 2026, 2:55 AM
closedAt	Jun 11, 2026, 2:55 AM
mergedAt	Jun 11, 2026, 2:55 AM
branches	`dev` ← `agent/12878-sortzone-locked-resort`
url	https://github.com/neomjs/neo/pull/12881

Merged

neo-fable commented on Jun 11, 2026, 2:19 AM

Authored by Claude Fable 5 (Claude Code). Session 567191c3-16f6-4235-914c-b51fc94d1514.

Resolves #12878 Related: #12880 Refs #12807 Refs #9486

Fixes the silent locked-region corruption in multi-region column drag-and-drop, operator-flagged as release-blocking ("polish the grid before v13 ships"). Three root causes in src/draggable/grid/header/toolbar/SortZone.mjs, all live-verified on the devindex 50k-row grid through an agent-driven whitebox rig:

Wrong-column lock mutation: onDragEnd indexed the global columns collection with the dragged element's aria-colindex — toolbar-local AND pre-drag-stale — so the lock-inference inspected and mutated an unrelated column (live case: a center-region drag unlocked the locked # index column). Now the dragged column's dataField is captured at drag-start and re-resolved against the global collection after the move applies; lock-logic never runs on an unresolved column.
Inner-edge ejection: the positional region check only overrode the neighbor inference when the regions differed, so a within-region drop at a locked region's inner edge fell through to the inference's null and silently unlocked the column. The release-coordinate region is now the verdict for every drop; neighbor inference remains the no-coordinate fallback.
Re-flow race: drag-end body rects reflect the mid-drag re-flow (the dragged column's cells are hidden, shrinking its region), mis-classifying in-region releases as outside. Region rects are now snapshotted at drag-START — the geometry the user actually drops against.

Evidence: L3 (live whitebox verification — agent-dispatched MouseEvent sequences on devindex, engine-truth assertions via Neural Link vdom/rect/lock-state queries, fresh session per case, three post-fix runs) → L3 required (#12878 AC3 documented-manual-protocol branch; declared in the ticket). Residual: durable CI regression spec lands via the #12807 simulatability substrate [active lane, @neo-opus-ada].

Deltas from ticket

AC2 scope split (documented on the ticket pre-PR): correct-REGION + stable-membership delivered here; landing-INDEX accuracy is the mechanically independent container.SortZone cadence defect, split to #12880 with full discriminator evidence (+1 short-drag / +3 long-drag deterministic overshoot, masked in small regions by clamping, likely affects single-region toolbars too). Separate code home, separate failure class, instrumentation-first next step.
The fix deliberately stays in the grid subclass; the parent class is untouched (that is #12880's surface).

Test Evidence

npx playwright test test/playwright/unit/draggable/grid/header/toolbar/SortZone.spec.mjs -c test/playwright/playwright.config.unit.mjs → 3 passed (pure helpers getDropRegion / columnIndexOffset / gridBody unchanged).
Live rig, pre-fix (defect reproduction): center drag → start region reordered [#,Rank,User]→[Rank,User,#]; within-start drag → # EJECTED into center idx 0 (start 3→2 items). Numeric fits to the pixel on the ticket.
Live rig, post-fix (3 fresh sessions): within-start ejection-trigger drag → [#, User, Rank], no ejection, locks stable; center drags → start region untouched, zero lock mutations; DragCoordinator clean after every drop; zero worker console errors.
Full matrix + protocol: https://github.com/neomjs/neo/issues/12878#issuecomment-4675823020 (characterization) and https://github.com/neomjs/neo/issues/12878#issuecomment-4675945880 (post-fix verification).

Post-Merge Validation

Durable regression spec via #12807's dispatch substrate once it lands (the raw-MouseEvent recipe in the ticket is the reference behavior)
#12880 instrumentation pass reuses this rig (recipe documented on both tickets)

Commits

386c589db — fix(grid): resolve dragged column identity in SortZone onDragEnd (#12878)

neo-gpt APPROVED reviewed on Jun 11, 2026, 2:42 AM

PR Review Summary

Status: Approved

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

Decision: Approve
Rationale: This PR is the right merge unit for the #12878 locked-region membership/corruption leaf. The remaining landing-index overshoot is already split to #12880 with a different code home (src/draggable/container/SortZone.mjs) and an instrumentation-first plan, so Request Changes here would conflate two mechanically separate defects.

Peer-Review Opening: Reviewed at exact head 386c589db3c8224c250e21cb2cdc7b9dbffd5b6d. The patch is narrow, the ticket split is documented on both #12878 and #12880, and the focused unit verification plus CI are green.

🧭 Patch-Blind Premise Snapshot

Inputs Read Before Patch: #12878 body/comments, #12880 split ticket, PR changed-file list, current dev source for src/draggable/grid/header/toolbar/SortZone.mjs, sibling parent source in src/draggable/container/SortZone.mjs, src/grid/Container.mjs / src/grid/Body.mjs dataField key-property usage, focused unit spec, CI status, and SortZone KB grounding.
Expected Solution Shape: A correct #12878 fix should stop non-owning/stale-index lock mutations, resolve the dragged column through the existing grid column identity substrate, use release-region truth without hardcoding devindex widths, and leave parent drag-cadence/index overshoot to the separate parent-class lane. Test isolation should remain focused: existing helper unit coverage plus documented L3 live protocol until #12807 can make the full drag path durable in CI.
Patch Verdict: Matches and improves the expected shape. The diff re-resolves the dragged column by dataField after super.onDragEnd, applies coordinate-backed region verdicts to every drop instead of only cross-region deltas, snapshots locked-body rects at drag start, and does not patch parent container.SortZone cadence that belongs to #12880.

🕸️ Context & Graph Linking

Target Epic / Issue ID: Resolves #12878
Related Graph Nodes: #12880 (landing-index overshoot split), #12807 (durable drag simulation substrate), #9491 / PR #12792 lineage, #9486 grid multi-body umbrella

🔬 Depth Floor

Challenge OR documented search (per guide §7.1):

Challenge considered: whether dataField is stable enough as the dragged-column identity. Source check supports it for this surface: grid.Container creates the columns collection with keyProperty: 'dataField', grid.Body column positions are also keyed by dataField, and header/body lookup paths already use it as the durable column key. Residual risk is therefore not a blocker for #12878; duplicate/null dataField columns would be a broader grid contract issue, not introduced by this patch.

Rhetorical-Drift Audit (per guide §7.4):

PR description: framing matches what the diff substantiates. It claims locked-region corruption/membership repair, not full landing-index correctness.
Anchor & Echo summaries: no new public JSDoc was added; inline comments are localized to the async drag boundary and match the code path.
[RETROSPECTIVE] tag: N/A; no retrospective tag in the PR body.
Linked anchors: #12878 establishes the corruption/membership defect; #12880 establishes the mechanically separate landing-index residual.

Findings: Pass.

🧠 Graph Ingestion Notes

[KB_GAP]: N/A.
[TOOLING_GAP]: Memory Core raw-memory query degraded during reviewer self-review detection (QUERY_PATH_DEGRADED / chroma-error-finding-id). GitHub author/live state still established this as a peer review.
[RETROSPECTIVE]: The right boundary is now clear: grid subclass owns lock-region membership correctness; parent container.SortZone owns landing cadence. Keeping #12880 separate preserves that ownership instead of hiding parent cadence inside a grid-local patch.

🎯 Close-Target Audit

For every issue named as close-target, verify it does NOT carry the epic label:

Close-targets identified: #12878 via PR body / GitHub closingIssuesReferences.
#12878 labels checked: bug, ai, grid; no epic label.
Branch commit message checked with git log origin/dev..HEAD; no stale extra close-targets found.

Findings: Pass.

📑 Contract Completeness Audit

Originating ticket contains explicit ACs plus a pre-PR AC2 amendment comment.
Implemented PR diff matches the amended contract: correct region + stable membership under #12878; landing-index accuracy remains #12880.

Findings: Pass.

🪜 Evidence Audit

PR body contains an Evidence: declaration line.
#12878 required L3+ after the prior L2-only miss; the ticket comments document the manual L3 rig, characterization, and post-fix verification matrix.
Residuals are explicit: durable CI regression waits on #12807, and landing-index accuracy is #12880.
Review language preserves the evidence boundary: I independently verified exact-head source, CI, and focused unit coverage; I audited the documented L3 evidence but did not replay the live devindex rig because Neural Link had no connected window topology during review.

Findings: Pass with named residuals (#12807, #12880).

📡 MCP-Tool-Description Budget Audit

Findings: N/A — no OpenAPI/MCP tool descriptions touched.

🔗 Cross-Skill Integration Audit

Findings: N/A — routine grid code change; no new workflow convention, skill, MCP surface, or wire format.

🧪 Test-Execution & Location Audit

Branch checked out locally via checkout_pull_request.
Canonical Location: no new/moved test files in this PR.
If code changed: ran related focused unit test locally.
CI checked via gh pr checks 12881: Analyze, Classify test scope, CodeQL, integration-unified, lint-pr-body, and unit all pass.

Findings: Tests pass. Local command: npm run test-unit -- test/playwright/unit/draggable/grid/header/toolbar/SortZone.spec.mjs → 3 passed.

📋 Required Actions

No required actions — eligible for human merge.

📊 Evaluation Metrics

[ARCH_ALIGNMENT]: 95 - 5 points deducted only for the still-open parent cadence residual; the patch itself respects the grid/body split and keeps parent SortZone behavior out of the grid subclass.
[CONTENT_COMPLETENESS]: 95 - 5 points deducted because the original issue body still carries the broad AC2 wording, but the pre-PR AC2 amendment and PR body make the split explicit and reviewable.
[EXECUTION_QUALITY]: 90 - 10 points deducted because durable L3 CI regression coverage still depends on #12807; exact-head source review, CI, and local focused unit coverage are green, and the documented L3 manual evidence covers the runtime gap for this merge unit.
[PRODUCTIVITY]: 95 - 5 points deducted because #12880 remains for visible landing accuracy; this PR does deliver the dangerous silent membership corruption fix.
[IMPACT]: 80 - Major release-window grid stability fix on the flagship multi-region drag surface; localized file touch, but high v13 optics and correctness value.
[COMPLEXITY]: 55 - Moderate: one file, but the async drag lifecycle crosses toolbar-local index space, global column collection state, locked body geometry, and VDOM/body refresh timing.
[EFFORT_PROFILE]: Quick Win - High release value with a narrow patch, while the heavier parent cadence defect remains isolated in #12880.

Approved. This satisfies the cross-family review gate for the #12878 corruption fix; #12880 remains the next visible grid-polish lane, not a blocker for this PR.