44-Character Block Vocabulary

1. Approach

The block structure analysis established that EAM messages are composed of discrete 44-character blocks, not governed by a periodic cipher. This page takes the next step: extract every block, cluster similar blocks into “block types,” and look for patterns in how blocks are ordered within messages.

1. Extract: Chop each message into 44-character segments starting at position 0. The final segment may be shorter (a “tail”).
2. Cluster: Within each prefix group, compare every pair of full-length blocks. Two blocks are the “same type” if ≥35% of their characters match at corresponding positions (well above the ~3% random baseline for a 32-character alphabet). Merge transitively.
3. Cross-prefix check: After within-prefix clustering, compare consensus sequences across prefixes to detect system-wide shared blocks.
4. Catalogue: For each block type, record its position within messages, temporal span, and relationships to other block types.

2. Key findings

Blocks are prefix-specific

Zero block types were found across different prefix groups. Every shared block exists strictly within a single prefix. This means the 2-character prefix genuinely marks distinct message families with distinct block vocabularies — shared blocks encode prefix-specific content, not system-wide structural elements.

Most blocks are unique

91% of block types are singletons — they appear only once in the entire corpus. The 52 multi-instance types (9%) represent structurally significant shared blocks: headers, trailers, and recurring content segments. Singleton dominance increases with block position (deeper in the message = more likely unique), suggesting that earlier blocks carry more reusable structural content while later blocks carry unique payload.

Blocks form chains

Same-prefix messages that share blocks typically share 2–3 consecutive blocks, forming chains. A message from prefix BV might share blocks at positions 0, 1, and 2 with another BV message from weeks later, while their remaining blocks differ entirely.

All tail lengths are even

Of the 238 tail blocks (shorter than 44 characters), every single one has an even length. This is striking confirmation of the 2-character pairing subunit identified in the alignment explainer. The smallest structural element is a character pair, and message lengths always contain a whole number of pairs.

3. Block chains: worked examples

Below are three block chains from different prefix groups. Each shows a set of messages that share consecutive blocks. Within each block type, characters that match the consensus are green; variant characters are dimmed.

4. Block ordering

Block types are position-locked. Header blocks (position 0) always appear at position 0; trailer blocks always appear at the final position. The block-type bigrams (which type follows which) form consistent chains within each prefix, suggesting a fixed ordering grammar.

Example chain orderings

The pattern is consistent: the first 2–3 blocks of a message are drawn from a small prefix-specific vocabulary (the “header chain”), while subsequent blocks are unique. This suggests the header chain encodes semi-persistent operational parameters (routing, addressing, key material?) that remain stable for weeks to months, while the deeper blocks carry per-message payload.

5. Tail length distribution

Message lengths are not arbitrary. The remainder after dividing by 44 (the tail length) clusters at specific values, with all tails having even length:

The two dominant tail lengths are 32 (51 messages) and 10 (41 messages), together accounting for 39% of all messages. Combined with the all-even constraint, this implies message construction follows rules that produce predictable total lengths — consistent with a block-assembly model where each component contributes an even number of characters.

6. Singleton position distribution

The 524 singleton block types (blocks appearing only once) are concentrated at earlier positions, simply because there are more blocks at earlier positions (every message has a block 0, fewer messages are long enough to have a block 4 or 5). But the rate of singleton-ness is actually fairly constant — unique content appears at every position:

Positions 1 and 2 have the lowest singleton rates (74% and 72%), meaning they are the most likely to be shared across messages. Positions 4+ are always unique. This aligns with the chain model: the shared “header chain” occupies positions 0–2, and everything beyond is per-message content.

7. Implications

• Block vocabularies are prefix-specific. No block types cross prefix boundaries. The prefix is a genuine message-family identifier, not just a routing label.
• Messages have a header–payload structure. The first 2–3 blocks form a semi-persistent header chain shared across messages in the same prefix over weeks to months. Blocks at position 3+ are unique per-message.
• The 2-character pair is the atomic unit. All message lengths (and therefore all tail lengths) are even, confirming the character-pair substructure.
• Header chains are temporally bounded. The most persistent chain (BV) spans 66 days; most span under a month. This suggests periodic key rotation or operational parameter changes.