_{ACCESSION RECORDS}

_{FIG. 01 — PRODUCT SYSTEM SPECIMEN}

Question — How can visual and spatial context survive the handoff to AI?

Canvas Prompt is a local-first canvas that preserves visual structure and context with marks, speech notes, revision history, and local Prompt Packages. It lets a workspace continue across sessions through explicit continuation, rather than requiring a full regeneration of scene assumptions.

→ Open repository

→ Open website

→ 3:50 live demo

_{FIG. 02 — RESEARCH SPECIMEN}

Question — What happens when humans are repeatedly trained by AI feedback?

A longitudinal self-case study based on three years of GPT conversations, examining how AI feedback reshapes questioning strategies, judgment frameworks, and identity coordinates.

→ Read the paper

_{FIG. 03 — PROTOCOL SPECIMEN}

Question — How can long-running AI projects preserve judgment without reloading raw conversation history?

A local project-state context engine that keeps decisions, rationale, pending changes, and current project state traceable across sessions and agents.

→ Open repository

_{FIG. 04 — ANALYSIS SPECIMEN}

Question — How do we audit messy project conversations without smoothing away evidence and disagreement?

Turns raw conversations into supported facts, participant claims, influence chains, judgment audits, and next actions.

→ Open repository

_{FIG. 05 — FIELD SYSTEM SPECIMEN}

Question — How do we turn an ambiguous customer problem into a delivery-worthy AI operating loop?

A reusable operator playbook for qualifying AI opportunities, modeling operational reality, defining human-in-the-loop boundaries, and compressing broad ideas into testable delivery loops.

→ Open repository

_{CURATOR'S NOTE}

Research first. Build what survives contact with reality.

Writing and field notes at aizhiz.com.

_{FORMERLY — Tencent Games publishing, ten years in marketing and content.}