Why does well-designed embedded firmware rarely need the general heap?

Because most real allocation patterns are not arbitrary-size-and-lifetime, they're uniform-size or batch-scoped, which the specialized allocators handle better. A fixed number of tasks, peripherals, and buffers → static arrays. Many same-size objects (packets, nodes, control blocks) cycling through alloc/free → a pool, O(1) and fragmentation-free. Bursts of temporary parsing/processing state → an arena reset per batch. Once you map each need onto the right strategy, the residue that genuinely requires arbitrary-size, long-lived, unpredictable allocation, the only thing general malloc is actually needed for, turns out to be small or nonexistent. And avoiding the heap buys determinism (bounded WCET), no fragmentation, link-time-known memory, and no unrecoverable exhaustion, exactly the properties real-time and safety-critical systems require, and why standards like MISRA restrict dynamic allocation. So it's less that the heap is forbidden and more that, pattern by pattern, a better-fitting allocator almost always exists.