Prospective direct request-frontier comparison: stock Nanobot and reconstructed Mode50, physical construction caps 4/8/150, two stochastic blocks, all ten development identities, 120 original attempts. No library or helper. Read record/CHOICE.md, STUDY.md and INTEGRATION.md. Restore frozen-source into a new workspace with the external official baseline and unchanged Harness-Bench checkout. Pinned images and private metered Luna credentials are external. The nested peer archive contains the Mode50 source reconstruction and neutral runtime check. Current public source matches; historical image identity is unestablished. All ten identities have prior exposure. Full-cost complete-work measurement is enabling reference evidence, not a novel harness or transfer claim. Physical retries and native finalization share the construction cap; grading is outside the cap and still charged. Existing synthetic preparation records retain their original scope; ENGINEERING.json records the new analysis checks. No benchmark task contents, answers, provider traces, credentials or current outcomes are included. This archive is made before any new paid inference is launched.