Long-task drift field reports: what changed before the first wrong action?

[1052326311]

Plan Lattice is built around a narrow execution boundary: the first protected mutation whose accepted intent or fact basis is incomplete, stale, or no longer authoritative.\n\nI am looking for real long-running Harness tasks that cross that boundary. Useful reports include:\n\n- the exact Plan Lattice and Harness versions;\n- whether the session compacted, resumed, delegated, or received new user input;\n- which contract, plan, target file, ownership chain, or external precondition changed;\n- the first wrong mutation, or a legitimate mutation that was blocked;\n- a sanitized Session event sequence or minimal reproduction.\n\nPlease do not post API keys, signing material, private benchmark assets, or workspace secrets.\n\nThe current public mechanism test covers 12 engineered stale-basis hazards and seven matched legitimate controls. It does not establish general coding-quality uplift. The frozen external-model study will report every result, including failures, before any broader claim is made.\n\nReports that expose a new invariant or a missing matched control will be turned into a focused reproduction before implementation.