# Controlled synthetic product demonstration — method and limits September 8, 2026. Fictional company, job and people. Not customer evidence. ## Actual method Four operator-written fictional files supplied company context, a job note, an old crew guide and a supplier note. A separate scripted owner reply clarified which extras rule applied. The primary file-capable Codex session followed the released Existing Business Starter Kit's document workflow in an isolated workspace and saved a bounded company record, job record, workflow, source register, open questions and pause state. The operator authored both the fictional inputs and the generated business records through Codex. This was not an independent customer usability trial. A standalone startup attempt reported insufficient write capability and stopped. Its logs show some policy-blocked read commands but no write attempt, so the underlying write limitation was not independently tested. No policy restriction was bypassed. A later, separate fresh AI process used read-only access to the saved records and produced the handoff without receiving earlier conversation or business facts in its task prompt. It was a context-consumption task, not a full Starter Kit restart or resume test. Codex operator review found that a record had compressed a room planned to be clear by Thursday into “reported clear.” The original output was preserved in private audit evidence. The saved record was corrected directly from the original source, and a fresh AI reader ran again. The final handoff is the actual unedited response of that second consumer run. No independent human review or real-world use occurred. Capture review also corrected a stale source-register field to reference the scripted owner clarification. That register-only metadata repair happened after the final handoff; it did not change the AI output or any business fact. Earlier source-register evidence was retained. ## Inspect the public-safe sample - [Owner orientation](inputs/01-owner-orientation.md) - [Office note](inputs/02-office-job-note.md) - [Old guide](inputs/03-old-crew-guide.md) - [Supplier note](inputs/04-supplier-note.md) - [Scripted owner clarification](inputs/05-owner-clarification.md) - [Company record](repository/knowledge/company-snapshot.md) - [Job record and correction history](repository/knowledge/job-cs-104.md) - [Workflow record](repository/knowledge/scope-to-crew-visit.md) - [Source-register excerpt](sources-excerpt.md) - [Open-item table excerpt](open-items-excerpt.md) - [Actual final AI output](outputs/crew-handoff.md) ## Release identity Existing Business Starter Kit: V1.0 Package.4, 44 released files. Released ZIP SHA-256: `69aab4b361441d864acfcd8281afb33d5d5b0f790612d366d46e0d5e24348b28`. New Business Starter Kit: V1.0 Package.6, 45 released files. Released ZIP SHA-256: `5964cc62a8888af0ea9a4b6c727b905849fd08a9870aa1489ab8f6f5f109a746`. Both ZIPs and instructions were inspected without changing their contents. This Existing Business demonstration does not verify the separate New Kit concept-building workflow. Neither the paid Kit instructions nor reusable templates are distributed in this sample. ## Screenshot and export treatment Real Chromium screenshots render the actual Markdown files in a plain editorial reading frame. The frame is not included software or a Viewer. The office-note and final-output captures show their complete text. The job and workflow images show named complete sections. The source table shows all five rows but only Source, Origin and Currentness columns. Captioned excerpts do not add business values. Exported record links point to the sample's source and open-item excerpts; business wording is unchanged. Table-header blank lines are normalized only for rendering. ## What is not established No real customer result, revenue, measured time saving, flawless AI, automatic source ingestion or updates, complete five-stage Kit build, universal provider compatibility, independent human onboarding success or superiority over ordinary AI is established. The AI draft needs human review before real use. The sample is not technical flooring, safety, legal or pricing guidance. The Repository is a portable set of inspectable business files that compatible AI can use. Environment permissions and capabilities matter. Original sources, clarification, uncertainty and review remain necessary. Prepared for local review. Publication and distribution require separate authorization. ## Follow-up technical investigation A later clean startup test reproduced the write denial with an actual temporary-file probe. The test launcher had omitted its native Windows sandbox selection while ignoring user configuration. Selecting that sandbox restored ordinary file writing while keeping workspace restrictions enabled; no released Kit instructions or global security settings changed. Both Kits then reached their opening questions. This corrects the earlier unresolved environment limitation; it does not turn the operator-directed Cedar & Slate build into a customer trial or establish universal compatibility. Separate, small synthetic tests then ran the unchanged Existing and New Kits through initial records, a deliberate pause and a fresh-session resume. Each fresh session recovered the saved next question without receiving the earlier conversation. Existing preserved a requested extra separately from accepted work. New preserved a founder's access preference separately from untested feasibility and demand. Review caught an unsupported payment-model detail in the New record; it was corrected with history retained, and a fresh reader recognized the correction. These tests support bounded continuity with review. They do not establish complete Kit coverage, equal performance between Kits, independent human usability, actual customers or commercial outcomes.