Document Center

I. Symmetric fixed-length frames — the full optimum for constrained devices.

1 · Static memory, zero heap allocation. Ask 128, answer 128 — a node allocates two fixed buffers at compile time; the hot path has no variable-length parsing, no TLV walking, no malloc. On MCU-class devices "no heap" directly means no fragmentation, no leak surface, predictable real-time behavior. This is the engineering delivery of the zero-parser claim: verification needs only the SHA-256/HMAC primitives plus fixed-offset reads.

2 · One review technique, two carriers. Both frames are fixed-layout, so reading them uses the same trivial method — constant offsets and hash primitives — and each reader can be audited in isolation. The carriers remain independent codecs (the specification requires that they do not parse each other); what is shared is the discipline, and that is what keeps every audit small.

3 · Fixed length → constant-time-friendly. The MAC comparison runs over a fixed layout with no variable-length loops, shrinking the side-channel surface; fields are O(1) reads — standing at offset 80 is an array index, not a "parse then extract". The validator is also total: every 128-byte input is either accepted or rejected whole, never partially parsed, and the rule set is small enough to reason about exhaustively, rule by rule.

4 · 128 is backed by hard constraints. A power of two (aligned allocation) and exactly two cache lines; above all, one datagram, no fragmentation: 128 bytes plus roughly 60 bytes of IPv6/UDP/CoAP header overhead fits comfortably inside the 1280-byte minimum MTU — fragmentation means reassembly buffers, reassembly means heap, and heap is a death sentence on constrained nodes. The upgrade runway is proven, not promised: the pulse frame tail (0x66–0x7F) went from zero bytes to the v1.2.6 extension block without changing the total length, and AE-128 reserves its own zero bytes (reserved + extension) for the same future.

II. The shared shard — the cryptographic handshake of the Q&A.

The amber link on the map is the most valuable stroke: the intent_shard at 0x36 in the question and at 16 in the answer are the same 16 bytes. Three engineering layers:

1 · One round trip. Addressing (WHERE + WHEN) and standing (WHETHER) ride the same exchange: one question, one answer — one round trip does what used to take two. On duty-cycle-powered LPWAN links, round trips are the dearest resource.

2 · The answer is bound to its subject. The MAC covers the intent_shard, so an answer cannot be re-targeted: swap in another subject’s shard and the bytes no longer authenticate — fail-closed. Freshness is carried separately, by the lease and the organ’s per-verdict nonce.

3 · From document pairing to byte pairing. "One datagram carries WHERE, WHEN and WHETHER" is no longer prose: two 128-byte frames sharing one constant is the physical implementation.

III. What is saved, against current practice.

Carrier	Size	Parser required
JSON envelope (the original AE route)	hundreds of bytes	JSON parser + field walking
CBOR encoding	compact	CBOR decoder (TLV logic)
AE-128 / PulseHeader128	128 bytes each, capped	none — fixed offsets + hash primitives only