Phase 2 integrations
GA4 · Planned
HubSpot · Planned
Apollo · Planned
LinkedIn Ads · Planned
Google Ads · Planned
Search Console · Planned
Phase 1 is document-driven: all data on this site traces to the research seed.
Agent Harness
Task queue, stage gates, operating rules, and required outputs
11 · Harness · seed 1.0.0
Execution contract, not a summary

Agent sessions in this repo are the research team. Each run reads data/harness-seed.json, works the task queue in order, writes documents to content/research/ and outputs to agentharness/outputs/, then updates data/harness-status.json. This page renders that live state.

Mission

Establish YUAN's Video-Native Edge AI category in the US market and increase qualified design-in opportunities, PoC-to-production conversion, and design wins.

12 · Queue
Task queue and stage gates

Every task must produce files, a diff report, and evidence links—never just a chat summary. A stage gate must pass before the next task starts.

T00

Source Audit+Claim Ledger
GATE ADone
claim_ledger.jsonsource_registry.csv
Marketing skills/research · /competitor-profilingGate A passed: 63 claims and a 60-source registry were validated. Official evidence covers all 30 seed companies; ambiguous founding years are explicitly scoped and unverified fields remain unknown.

T01

Competitor Deep-Dive Scorecard
after T00GATE BDone
competitor_scorecard.jsoncompetitive_gaps.md
Marketing skills/competitors · /competitor-profilingGate B passed: the database expanded from 30 to 74 companies (41 direct, 27 adjacent, 6 ecosystem), with a six-axis scorecard, evidence maturity, and competitive gaps.

T02

Messaging Architecture
after T00, T01GATE CDone
messaging_architecture.mdproof_point_backlog.csv
Marketing skills/product-marketing · /copywriting · /marketing-psychologyGate C passed: product-marketing context, messaging architecture, and a 14-item proof backlog are complete; unverified performance, Re-ID, and cross-platform claims remain restricted from external use.

T03

Website IA+Page Briefs
after T02GATE DDone
website_ia_and_page_briefs.mdconversion_map.json
Marketing skills/site-architecture · /cro · /seo-auditGate D passed: US website IA, eight page-brief types, a five-stage conversion architecture, and multilingual URL guardrails are complete; no external website changes were authorized.

T04

Content Operating System
after T02, T03GATE EDone
content_backlog.csvcontent_brief_schema.jsonreview_checklist.md
Marketing skills/content-strategy · /programmatic-seo · /ai-seoGate E passed: 300 unique bilingual ABM ideas, a content-brief schema, publication review checklist, priority model, and dashboard are complete; all ideas remain evidence-gated.

T05

ABM Account Model
after T01, T02GATE FDone
abm_target_model.csvcampaign_matrix.md
Marketing skills/prospecting · /cold-email · /revopsGate F passed: 100 unique-domain named accounts, fit scoring, readiness guardrails, and a campaign matrix are complete; all remain research_required, and high fit is not treated as buying intent.

T06

Measurement+Attribution
after T03, T04, T05GATE GDone
measurement_spec.mdevent_taxonomy.json
Marketing skills/analytics · /attribution · /ab-testingGate G passed:完成 measurement spec 與 21-event taxonomy,定義 MQL/MQA/SQL、design-in opportunity、PoC、design win、recognized revenue、account-level attribution、隱私與 data-quality gates。所有 connector 均維持 Phase 2 specification-only;Apollo 仍是唯讀公司層級研究輸入,沒有 CRM write、PII export 或外聯。
Non-negotiable rules
Operating constraints

1Official English sources first; keep unknown as unknown—never backfill.

2Separate verified, user_provided, inference, and hypothesis states.

3No “Top N” company rankings without consistent revenue data.

4Cross-camera tracking must split capture / sync / inference / Re-ID / VMS ownership.

5Every core website or content claim links to an evidence object.

6US-facing copy in English; research, decisions, and handoffs in Traditional Chinese.

7All content centers the ISV / OEM / SI design-in journey.

8Jetson TOPS is never the primary brand differentiator.

Stop conditions
When the agent must halt and ask

· A core claim lacks a direct official source.

· The task requires internal benchmarks, customer authorization, or product-roadmap data that has not been provided.

· Official sources conflict and the conflict cannot be resolved.

· The scope expands to unauthorized external publishing, customer contact, or system changes.

Evidence readiness

Market benchmark seed

Ready

Yuan public evidence

Partial

Internal benchmark data

Missing

Named customer proof

Missing

Messaging hypothesis

Testing

ABM named-account universe

Ready

Apollo current-client account audit

Ready

AgentHarness market and Apollo run manifest

Ready

Bilingual ABM content backlog

Ready

Measurement and attribution contract

Ready
Source document
YUAN-EDGE-AI-MARKET-HARNESS-20260824

Version 1.0.0 · research cutoff 2026-08-24 · 作為人類可讀市場報告與後續 AgentHarness 的共同起點

Open the original research + harness document (HTML) →