Who Is Working for Whom? LLM Quota Resets, Time-Zone Asymmetry, and Heavy Users' Temporal Autonomy

Quota resets can replenish a scarce computational resource while creating a new demand on human scheduling. This paper distinguishes project-directed intensive AI use from quota-directed use motivated by exhausting an allowance before it refreshes. It reconstructs a bounded public archive of Codex/ChatGPT Work and Claude/Claude Code reset events, separates announcement, public completion notice, individual notification, account arrival, redemption, and expiry, and models how public notice windows intersect with local sleep schedules. The corpus contains 95 distinct event records: 81 automatic resets, three banked offers with documented expiry terms, and 11 banked offers whose exact offer-specific expiry remains unresolved. Fifteen records have separate announcement and completion-notice anchors, including 13 automatic resets. No individual-account arrival or health outcome is observed. Assumed weights rank automatic, expiring banked, and explicitly non-expiring resets in descending order; this is a modeling choice, not a result. Four figures present event history, weighting sensitivity, sleep overlap, and rolling potential pressure across time zones. The contribution is an auditable documentary and conceptual framework, not a causal estimate of sleep loss, exhaustion, or changes in facial appearance. Independent working paper No. 07, Version 1.0 (2 November 2026). Not peer reviewed. The English PDF includes the event ledger, source links, paired-time tables, and model assumptions. No participant-level health data or private personal case history is published.