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	14435
title	Temporal-pyramid L3–L5 dynamic synthesis path (weekly/monthly/quarterly)
state	Open
labels	enhancementaiarchitecture
assignees	[]
createdAt	Jul 2, 2026, 6:09 AM
updatedAt	3:04 PM
githubUrl	https://github.com/neomjs/neo/issues/14435
author	neo-fable-clio
commentsCount	1
parentIssue	12679
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]

Temporal-pyramid L3–L5 dynamic synthesis path (weekly/monthly/quarterly)

Open Backlog/active-chunk-2 enhancementaiarchitecture

neo-fable-clio commented on Jul 2, 2026, 6:09 AM

Context

Leaf C of Epic #12679's implementation decomposition (map: https://github.com/neomjs/neo/issues/12679#issuecomment-4862079789). Filed pre-Accepted per ADR 0005 §2.3. Sequenced after Leaves A (#14433) + B (#14434) — synthesis reads L2 aggregates.

⛔ Merge gate: merge-blocked until ADR 0028 is Accepted (PR #14428), per #11376 §6 and ADR 0028 §5.

The Problem

"What happened this week / month / quarter?" has no cheap answer path. The graduated design answers it with query-time synthesis over the durable tiers — deliberately NOT precomputed, so the photocopy-of-a-photocopy degradation and the tier×partition cron explosion never exist.

The Architectural Reality

Bound by ADR 0028 (re-verify Accepted at claim):

§2.2: L3 (weekly) / L4 (monthly) / L5 (quarterly) are computed on demand over L2 aggregates + query_recent_turns (chronological) + query_raw_memories (semantic). No durable output above L2 — nothing this leaf produces is written back as summary-of-summary. (The Gemini-family objection incorporated as the Cycle-3 shape; liveness carry in ADR §8.)
§2.5 fidelity: higher tiers cite sessions impact >= 90 and accepted ADRs directly; PR fidelity via named sources (accepted-ADR links / epic labels / high-impact review metadata). No invented universal PR impact score.
Query-time cost: the corrected Cycle-3 math (~33–34 LLM calls/week) is the accepted budget — re-verify at implementation (ADR §4).
Surface boundary: includes only the minimal Memory Core query entry point that invokes synthesis. Dashboards, cockpit panes, and richer consumer surfaces stay out (epic Out-of-scope; #11375 territory).
ADR §6 consume-direction (pre-declared): once the #14422 shared tier contract lands, synthesis over weak-digest sessions propagates a fidelity signal using that contract — adopt, don't invent.

The Fix

One PR: the L3–L5 synthesis path (window resolution → L2 + recency + semantic retrieval → cited synthesis with §2.5 fidelity rules) + the minimal MC query entry. No scheduled jobs, no durable writes, no consumer UI.

Contract Ledger Matrix

Target Surface	Source of Authority	Proposed Behavior	Fallback	Docs	Evidence
Synthesis path (new service logic)	ADR 0028 §2.2	On-demand window synthesis over L2 + existing retrieval; zero durable output above L2	No bird's-eye answers (status quo)	ADR 0028 §2.2	Tests assert no writes above L2
Citation discipline	ADR 0028 §2.5	impact≥90 + accepted-ADR direct-citation; named PR sources	—	ADR 0028 §2.5	Citation-rule assertions
Minimal MC query entry	epic problem-scope ("navigable") + ADR §6 boundary	One entry point; consumer surfaces excluded	—	Tool/service JSDoc	Entry-point contract test

Decision Record impact

depends-on ADR 0028 (Accepted-gate) · consumes Leaves A (#14433) + B (#14434) · future-adopts the #14422 tier contract (ADR §6 consume-direction).

Acceptance Criteria

L3/L4/L5 windows synthesize on demand over L2 + query_recent_turns + query_raw_memories; no durable artifact above L2 is written (asserted by test).
§2.5 fidelity enforced: impact≥90 session citation + accepted-ADR direct-citation + named PR sources.
Per-agent + unified partition views honored (§2.6) at synthesis time.
Minimal MC query entry point shipped; no dashboard/cockpit surface included.
Query-time cost measured against the ~33–34 calls/week budget and recorded in the PR.
PR held un-merge-ready until ADR 0028 Accepted; gate stated in the PR body.
Cross-family review before merge.

Out of Scope

Durable precompute of any higher tier (anti-anchor). Consumer surfaces beyond the minimal entry (#11375). The #14422 vocabulary work itself (adopt-when-landed only). Schema (A) and aggregation (B).

Avoided Traps

Durable L3–L5 cascade — rejected (photocopy degradation; the design's own center).
Per-tier crons — rejected (batch-explosion; dynamic-by-design).
Inventing a provenance vocabulary before #14422 lands — rejected (ADR §6: adopt the shared contract; don't fork a dialect).

Parent: #12679 (native link on file). Authority: ADR 0028 (PR #14428). Depends-on: #14433, #14434. Adjacent: #14422 (tier contract), #11375 (consumer surfaces).

Live latest-open sweep: latest-10 at 2026-07-02T04:06Z + this filing window; no equivalent. A2A sweep: all-states at 04:06Z; no competing claim. Filed unclaimed — self-select; claimer runs ticket-intake + structural-pre-flight.

Origin Session ID: 2251c81c-1446-4723-86b3-479322bbcc95 Retrieval Hint: "temporal pyramid Leaf C L3 L5 dynamic synthesis fidelity citation no durable cascade"