Research · Measurement before narrative

Make the signal defensible.

02 / Measurement
Separate systems from people

Delivery telemetry needs context.

Email-security gateways, reputation systems, link scanners, and prefetch behavior can touch campaign infrastructure before an employee does. Attribution rules should be documented before scores are compared.

Gateway effects · 12 Jul 2026R-01

Semantic dilution is a measurement problem.

The source note examines how gateway processing, feature extraction, and reputation systems can change what a campaign event means. Treat machine-generated touches as an observation to classify—not automatic evidence of human behavior.

Read the maintained Chinese source →

Scanner signals · 03 Jul 2026R-02

JA4 and cloud score are supporting signals.

The source note considers scanner and prefetch identification. Network fingerprints can support classification, but should not be presented as infallible proof of a person or a particular product.

Read the maintained Chinese source →

Human risk is a decision aid, not a verdict.

A defensible program preserves the underlying event types, the confidence of attribution, the exercise context, and the time window. A composite score can help prioritize follow-up, but should not erase uncertainty or be used as the sole basis for employment action.

Recommended interpretationUse delivery, view, interaction, reporting, and training evidence as distinct observations. Review trends at cohort and campaign level before drawing conclusions about an individual.

AiTM exercises require an explicit boundary.

An adversary-in-the-middle exercise can interact with authentication flows and sensitive data. The source note focuses on written authorization, contract and governance scope, data handling, stop conditions, and the distinction between demonstrating exposure and collecting unnecessary secrets.

Read the AiTM and MFA source note (zh-CN) →

Not product documentation. Research topics describe questions, methods, and operating constraints. They do not imply that a named technique or module is included in every TaiGong version or license.
03 / Operations
Sequence and entitlement

What is visible is not always enabled.

Channel design, tool discovery, server configuration, licensing, and external-platform approval are separate layers. An evaluation should name which layer supplies each capability.

Multi-channel · 20 Jun 2026R-04

Email and IM need one governed sequence.

The source note examines sequencing across email and instant messaging. Channel availability, identity mapping, timing, consent, and reporting should be treated as operational dependencies—not assumed from a single campaign label.

Read the maintained Chinese source →

MCP operations · 11 Jun 2026R-05

Tool discovery is not entitlement.

An MCP client may display a tool description while the server still enforces authentication, version, license, and policy. The source note separates client visibility from the authority to execute an operation.

Read the maintained Chinese source →

Commercial boundary · 02 Jun 2026R-06

SaaS and self-hosted are delivery modes—not identical scopes.

The commercial-license note separates deployment location from license entitlement, modules, service obligations, and infrastructure ownership. The signed order and version-specific support matrix remain the governing sources for a delivery.

Read the maintained Chinese source →

Turn a question into an evaluation.

Define the evidence first.

Tell us what you need to measure, your delivery channels, deployment boundary, and governance constraints. We will scope the relevant product and documentation path.