D70: reproduce native metadata

Both native builders, Duncan and the six selected CRing spans, run from the exact source archive plus 26 additional frozen inputs. Two isolated executions of the packaged code reproduce all 13 shipped metadata artifacts byte for byte.

Run it yourself

  1. Download the dependency and editable replay-source ZIP. Extract it; keep the original ZIP beside the extracted files.
  2. Download the unchanged native source ZIP from its original public record. Do not rename it. Native book sources remain in that lineage, not in the shared adapter.
  3. With Python 3.10 or later, run the command below from the extracted directory. No package installation, network connection during replay, translation model or TeX engine is required. Historical translation scripts are only evidence inputs and are not executed.
python -B scripts/replay-d70-native-metadata-v1.py --bundle D70_NATIVE_METADATA_REPLAY_V1.zip --source-archive 05_o013-sumber-backend-1.0.0.zip --work-dir replay-work --output replay-result.json

Expected result: state “pass”, 13 matching metadata artifacts. Inspect replay-result.json for each artifact and validator result.

Exact download identities

D70_NATIVE_METADATA_REPLAY_V1.zip
3001559 bytes; SHA-256 cde3a053dc0c16e6056c6827f05bf955c1f511c375d02e4e8d404db084bf850c
05_o013-sumber-backend-1.0.0.zip
1295518 bytes; SHA-256 6273206ffb42277f3040d638e1a0f0870596b823a239fa9d3947d964aa094ef9

What this does not prove

This closes the metadata dependency gap, not reproduction of the full Li reader or all four PDFs. No new semantic terminology/canon review, tagged-PDF claim, screen-reader test or whole-program completion is implied. Native component rights, exact selected boundaries and translation credits are unchanged.

Machine-readable execution evidence · All forty roles and remaining requirements