The initial handoff fixture displays seventeen hundred estimated tokens. Tracked-file estimates sum to fourteen hundred ten, and fixed weights add ninety for two completed milestones, eighty for two constraints, seventy for two pending tasks, and fifty for one error. Adding one nonempty milestone adds forty five, producing seventeen hundred forty five regardless of its real token length. At point one five pixels per estimated context token the bars span two hundred fifty five and two hundred sixty one point seven five. This is metadata budgeting, not actual tokenizer output, file inspection, or transferred model context. The readiness score starts at one hundred and subtracts thirty if no pending tasks remain. Removing the two fixture tasks therefore yields seventy. At three pixels per authored readiness percentage point the bars shrink from three hundred to two hundred ten. Removing all architectural constraints would subtract another twenty. These point rules are a planning heuristic rather than measured context-loss probability. The exported project brain preserves fields and task lists, but a high badge does not prove another model loaded them or understood the work. The handoff button only refreshes local diagnostics and artifact text. When an active error exists, the source subtracts fifteen unless any completed milestone contains isolated, spec, or pkce after lowercasing. With no matching milestone the score is eighty five. Adding an arbitrary milestone containing spec restores one hundred, even without actually fixing or isolating the error. At three pixels per readiness point the bars span two hundred fifty five and three hundred. This explains the heuristic, not an instruction to misrepresent progress. The download is a JSON project-brain artifact despite the button mentioning zip, and the generated CLI or MCP strings are not executed. Preserve genuine evidence separately from these source labels.