P.3/P.4 frozen as P.5 inputs

An independent rebuild of p13 passed. A reviewer unpacked XVI-P-source-only.zip (sha256 2f85a566) in a fresh environment and ran the source-only entrypoint. All reference values matched (112/112, no mismatch), 77 values were recomputed by separate code, and a hand-written spot check of the conformal boundary identity agreed. The report is archived as archival/p13/reproducibility/independent_rebuild_report_p13.json (sha256 pinned). Caveat: the rebuild ran on the same box (shared kernel, TeX Live and poppler) in a fresh virtual environment from the source-only zip; it was not on a second machine.

The flag is bound to the p13 source tree 2b1baf33 and to the p13 zip; p14 changes cannot inherit it, and p14 has its own (pending) independent flag. P3_P4_frozen_as_P5_inputs is recomputed in every build by freeze_p13.py as the conjunction of three conditions: F1, the archived independent report has its pinned hash, pass=true, no mismatch, and is bound to the p13 tree and zip; F2, the 54 frozen files (classical P.3/P.4 code, the inherited XVI-O code and notes, pins and inherited data; archival/p13/frozen_p13_inputs.json) are byte-identical; F3, the regenerated classical values (export hashes, the seven classical certificates and two ledgers, residual records, pins) equal archival/p13/reference_values_p13.json. The classical code is therefore unchanged in p14. Its certificates keep their P13_* names, and the p14 gate recomputes the p13 acceptance from them (Section sec:frozen-classical). P.5 entry is stated against this freeze: if any of F1–F3 fails, P.5 entry is not earned and publication is rejected.