Machine-only exploratory pilot — v1.2

This package executes MACHINE-ONLY-PILOT-PROMPT.md (authored by the Codex research director) as an exploratory, machine-only instrument-debugging study. It does not modify, overwrite, or reclassify FE-BOUNDARY-2026-07-24 (v1.0) or FE-BOUNDARY-2026-07-24-v1.1, both of which remain:

Stage A: incomplete
Stage B: blocked-before-stage-b
classification: inconclusive
stage_b_authorized: false

Read pilot-gate-checklist.md for acceptance conditions, research-journal.md for the run log, and (once complete) the final report roles M8/M9 produce under analysis/ and validation/.

Provider-family limitation, stated prominently per the prompt's own requirement: every role in this run (M1–M7) is executed by an isolated Anthropic/Claude agent context. This satisfies role isolation (no agent curates both sides, no curator reviews mappings) but does not satisfy independent external replication across model-provider families — Codex (OpenAI) authored the prompt and ran v1.0/v1.1, but did not participate in this run's M1–M9 roles.