Historical policy review and current v32 contracts
The released S27 review and its S28/S29/S30/S34/S36 follow-ups remain available as the byte-identical v31.1-roots.6 historical review. Its source is the exact released commit e93a8cca047c12ce9cab97181dd30705ed0e7e83, original Git blob 1837f64c27e05e1c3749ea82fac344007fe17dca. Embedded relative paths refer to that original v31 document/source context; the pinned released document preserves those historical links. The inventory, raw-measurement context and 32,768-slot baseline are historical, not current v32 qualification or defaults.
The unpublished v32.1 candidate starts directly from Core f0de87f040621468b956f01f526a43a010310501. Current behavior is described by the preservation catalog, port review and owning policy grammar, effective-policy RPC and subdust contracts. Their current-source tests and exact build/run evidence govern the port; the old review is not substituted for them. Native v32 graph/package/replacement/HTTP interfaces remain native where the carried contract does not require an intentional adaptation.
Admission, relay and mining-template restrictions remain local policy. Consensus-valid blocks containing locally rejected transactions remain valid. Extra compact-reconstruction references do not admit transactions to the mempool or authorize relay. No RDTS/BIP110 or new consensus/privacy behavior is introduced by restoring the review or tooling. The production cache retains Core's 100-entry count and the Roots 10 decimal MB charged-dynamic-byte limit; slot/object overhead, shared ownership and process RSS remain separate budgets.
The restored offline cache experiment preserves the S29 multi-cell exact-Core-versus-Roots comparison, injected insertion and InitData CPU/wall metrics, Linux process memory/CPU samples, reference snapshots, observed reconstruction requests/hits/wire bytes and raw paired repetitions. Only disposable exports receive instrumentation. Its explicit v31.1 and v32.1 adapters require exact source/configuration/API boundaries. Current native cache microbenchmarks complement this experiment; they do not replace it. Whole-vector snapshots warm the reconstruction input, and there is no uninstrumented production latency control or mainnet representativeness claim. A bounded smoke or a unit-test pass cannot justify a default change or qualify an unrun campaign.