White Paper

.SYSTEMX white paper

Portable project memory for agentic work is now a 50-page expanded implementation reference, covering the complete coordination format and its operating boundaries for founders, engineering teams and AI tooling practitioners.

Edition 1.1 | 4 October 2026 | 29 chapters plus reference appendices

Download the expanded PDF · Read the editable Markdown · Metadata and checksums

Listen to the paper

Download or play the full audio reading · Spoken-text transcript

This is a synthetic voice reading. It retains the prose, questions, code and file inventory, converts tables into header/value sentences and diagrams into ordered labels, and omits long web addresses, Markdown markup and inline reference numbers. The PDF and Markdown preserve exact spelling and formatting for commands and paths. Audio is a separate release asset, outside the portable install bundle.

Reading routes

Need	Sections
Understand the purpose and architecture	1-3, 15-16
Follow a complete coordination turn	4-7, 15, 17-21
Inspect exact data and review contracts	17-21 and appendices A/B
Handle workers, locks, failures and recovery	19, 22, 25
Install, update, remove or integrate tooling	10, 23-24 and appendix C
Coordinate child projects, chats and cloud sessions	8-9, 28
Walk one real local task from failure to acceptance	26-27
Assess efficiency, evidence and practical limits	11-14, 29
Find every baseline file and component	Appendices D/E

What expanded

The new edition adds field-level task, focus, event, schedule, review and policy contracts; every permitted task transition; dependency and reopening rules; context bounds; Agent 0 assignment/return responsibilities; X/Z identity and replay mechanics; atomic-write and local-lock limits; recovery decisions; command plans; version selection; exact casing; multi-project ownership; and all 100 default Agent Z questions, copied directly from the examined policy.

The executable walkthrough creates a disposable project, observes a failed local HTML check, applies the specific correction, records Agent X evidence, retains and compares Agent Z reports, verifies task acceptance and closes a tracked due item. It checks replay behavior and refusal paths. It starts no subagent runtime or external job and makes no browser, hosting or production-readiness claim.

The PDF includes numbered contents, bookmarks, selectable text, two vector diagrams and clickable references. It is not a tagged PDF/UA export; Markdown is available for accessible reading and editing. Optional publishing tools are in docs/media.

Editions and interpretation

Edition 1.1 examines the v1.8.3-alpha.1 implementation and ships with the v1.8.4-alpha.1 documentation patch. Runtime behavior and record schemas are unchanged. The paper distinguishes implementation, tested local examples, illustrative planning, efficiency hypotheses and proposed future extensions. It does not claim measured savings, AGI, vendor endorsement or certification.

The original 17-page edition 1.0 remains unchanged: PDF · Markdown. Its technical baseline is v1.8.2-alpha.1.

For hands-on adoption, use First-time setup and the New-chat setup prompt. For visual workflow cards, open Use-case infographics. Downloads, version pins and release evidence remain in Versions and changelog.