# Frozen question before execution — 18 November 2026

Does native PDA change control retain enough to reconstruct an earlier decision after correction and restart, more dependably than an ordinary versioned event ledger?

Use the current upstream commit defd859819680f4bbc6341ae01a2b6e131a02d5d (unchanged since the first essai), its advertised change tools, and wholly fictional shifts. Two matched histories differ only in their first approver/date, then converge on identical substantive current status, approver, date and notes. Query through MCP after an actual server restart. Preserve every request/response; no upstream analytical code changes or provider calls. Inspect any native audit evidence before claiming history is lost throughout the system.

Predeclared questions: current status/approver/date; first approver/date; original availability basis; corrected basis and reason; withdrawal date; whether the current approval supersedes the earlier one. Compare the ordinary get_change_log response, any native forensic audit route, and a simple append-only JSON event ledger preserving the same submitted fields. Count fields/actions and available answers; do not infer human time or usefulness from machine timings.

Invalid status is a material correction failure probe. The app may reject safely or reveal a current gap. A second implementation is not supplied to PDA to repair the result. Success for PDA requires recoverable earlier decision identity and transition order, not merely survival of its latest row. A negative result is acceptable.
