Actual CODAL heap alignment diagnostic

This prospective diagnostic preserves the selected raw CODAL heap and the existing ten service cases. It is not allocator admission or native Arcade qualification. Execution belongs to the source-bound hosted workflow; local manufactured witnesses are predicate controls only.

Runtime caller mode5 comes from CrispStrobe/codal-samd source 92e99cc881db3e5d8f0186821d8c83bc241eb3b2, with actual core 2ed19d64ec3359143409a4b0a9758b4aeef3ee7d. The workflow uses a sixth clean core checkout. Existing modes0–3 keep their four-million-step bounds; mode4 keeps its separately named twelve-million-step bound. Mode5 is bounded by four million actual machine steps in each scheduler profile.

The caller first completes its existing free/reuse heap control, then retains two simultaneous eight-byte allocations before constructing Timer, bus or fiber services. The expected old-source bug witness is exactly 0xe0000c60, stage1, checks1 and heap-reused1. Actual RAM addresses must be distinct, four-byte aligned and twelve bytes apart; at least one must be four modulo eight. They are observed guest values, never injected expected addresses. Actual exception, SysTick entry/return and assembly-switch observations must all remain zero, with PRIMASK1 and no pending SysTick at the terminal witness.

Each image retains unique aligned BSS witnesses, poisoned BSS/data startup verification, vector verification and source/core/image/map provenance binding. The nonignored manufactured controls reject wrong status, stages, sizes, address domains/alignment/stride/overflow, service activity and architectural state. They do not execute the guest or prove the bug happened.

Preserve the sole official original result before changing allocation semantics. A passing expected-bug diagnostic is evidence of missing fundamental alignment, not successful allocator behavior. A later adapter must preserve the actual raw heap path, explicitly separate pointer domains and qualify allocation/free, reallocation/content/failure and real eight-byte object accesses. Full image licensing, native Arcade integration, performance and installed-app adoption remain separate pending gates.