How to Read These Heatmaps

What Are EAMs?

EAM stands for Emergency Action Message. These are coded messages broadcast over high-frequency (HF) radio by the United States military's High Frequency Global Communications System (HFGCS), a network of transmitter stations that can reach receivers anywhere in the world.

Each EAM is a string of uppercase letters and numbers. They are transmitted in the clear (unencrypted at the radio level), but their contents are encoded—only the intended recipients know what a given message means. Radio hobbyists routinely log these broadcasts, and over time a large collection accumulates.

These heatmaps are a tool for exploring structural patterns in that collection: recurring regularities that suggest the messages follow an underlying format rather than being purely random.

What This Tool Does

Each heatmap page shows all logged EAMs that share the same character length. The "55 Character EAMs" page, for example, shows every message that is exactly 55 characters long.

The tool analyzes these same-length messages as a group and looks for places where the same characters, sequences, or relationships appear again and again—across different messages or within individual ones—in ways that are unlikely to happen by chance. When it finds patterns, it highlights them with colors. Each color represents a different type of pattern.

The Heatmap Bar

At the top of each page is a horizontal bar that stays visible as you scroll. This is the aggregate heatmap. Each column represents one character position (position 1, position 2, etc.), and the brightness of each column reflects how much pattern activity was detected at that position across all messages on the page.

Dark columns indicate few or no patterns. Those positions likely hold unpredictable data—encrypted content, random identifiers, and so on.

Bright columns indicate many messages share structural patterns at that position. This suggests the position has a defined role in the message format: a fixed field, a check character, a predictable delimiter.

Pattern Colors

When a character is part of a detected pattern, it is highlighted in one of five colors:

• Blue — Exact Repeat. A sequence of two or more characters that appears more than once in the same message.
• Green — Gapped Repeat. A sequence that repeats within the message, but with one or more wildcard positions (characters that differ between the two occurrences).
• Purple — Hot Position Pair. Single characters at two positions that match each other across many different messages.
• Pink — Quadruple Repeat. Four or more identical characters in a row.
• Coral — Partial Quad. Two or three identical characters in a row, at a position where other messages tend to have quadruple repeats.

Exact Repeats (Blue)

The most common pattern type. The tool scans each message for substrings—sequences of consecutive characters—that appear more than once within that same message. In bioinformatics, a substring of length k is called a k-mer ("mer" just means "part," so a 3-mer is a three-character part).

Longer repeats are generally more significant. A 2-character match in a long message could easily happen by chance, while a 5-character match is far less likely to be coincidental. The tool uses a confidence score to account for this (see Confidence Filtering below).

Gapped Repeats (Green)

Similar to exact repeats, but with flexibility. A gapped pattern is a sequence that repeats in the message where most characters match, but one or more positions act as wildcards—they can differ between the two occurrences.

The wildcard characters (X and Y above) are not highlighted—only the matching positions get green. These patterns must be at least four characters long and often reflect structured fields where some positions are fixed and others vary, like a template with fill-in-the-blank slots.

Hot Position Pairs (Purple)

This is the only pattern type that looks across messages rather than within a single one. The tool compares all messages of the same length and asks: are there two positions that tend to hold the same character as each other, message after message?

The threshold is 30%: the characters at the two positions must match in at least 30% of messages within a group to qualify. The specific character changes from message to message, but the relationship persists—the two positions are structurally linked. This suggests they may be computed from the same input, or one is derived from the other.

Quadruple Repeats (Pink)

The simplest pattern: four or more of the same character in a row (e.g., AAAA). In a random 36-character alphabet (A–Z plus 0–9), a run of four identical characters is quite rare. When it happens, it likely reflects something deliberate in the encoding—padding, a separator, or a special marker.

Partial Quads (Coral)

Sometimes a message has a double or triple repeat (like AA or BBB) at a position where other messages of the same length tend to have full quadruple repeats. The coral highlight means: "this position is a known quad-repeat zone, and this message has a partial match here." A double repeat at an arbitrary position would not be highlighted—it only triggers when the position is structurally significant across the broader message set.

How the Analysis Works

1. Group by length. All logged EAMs are sorted by character count. Each unique length gets its own page. Position-based analysis only makes sense when position 15 in one message corresponds to position 15 in another, so same-length grouping is essential.
2. Identify hot position pairs. Within each group, the tool checks every possible pair of positions to see if they tend to hold matching characters. Pairs that match in at least 30% of messages are flagged as "hot." This is a cross-message analysis.
3. Scan each message for within-message patterns. Each individual message is checked for exact repeating substrings, gapped repeats, quadruple runs, and partial quads.
4. Score confidence. Short repeats in long messages are often coincidental—the birthday paradox makes collisions surprisingly likely. The tool estimates the probability of each pattern occurring by chance and assigns a confidence weight accordingly. Low-confidence patterns appear dimmer; very weak ones are filtered out.
5. Consolidate and prioritize. When multiple patterns overlap at the same position, the most informative one is displayed. Priority order, highest to lowest: Quad Repeat > Partial Quad > Gapped > Exact > Hot Pair. In certain cases, a strong hot pair can override a weaker exact repeat when the exact repeat isn't fully corroborated by cross-message evidence.
6. Render. The results are assembled into the interactive page, with the aggregate bar at top and per-message colored characters below.

Interacting with the Pages

Click a highlighted character in any message. Every other message that shares the same pattern at that position will light up with a white outline, and non-matching messages will dim. This is the fastest way to see how widespread a pattern is. For purple (hot pair) patterns, clicking is especially useful: it shows every message where the two linked positions match.

Click a dot (the colored circle to the left of each message). This highlights all messages in the same group, so you can see which messages were classified together.

Filter controls let you show or hide specific groups using checkboxes. Useful for isolating one group at a time.

Sort controls let you switch between grouped ordering (messages organized by group, then by time) and chronological ordering (all messages in time order).

Navigation: arrows move to adjacent message lengths, the dropdown jumps to a specific length, and the "Index" link returns to the overview page.

Groups and Dots

Messages of the same length are further divided into groups based on shared characteristics. Grouping helps identify families of messages that may follow different sub-formats within the same overall length.

● Group 1 ● Group 2 ● Group 3 ● Group 4 ● Group 5 ● Unclassified

The dot next to each message reflects its group. Messages that couldn't be confidently assigned get a gray dot and are labeled "Unclassified."

Confidence Filtering

Not all detected patterns are equally meaningful. A 2-character substring repeating in a 142-character message is not very surprising—with that many characters and only 36 possible values, short collisions are statistically expected. A 5-character repeat in the same message is far more noteworthy.

The tool handles this with two tiers of scoring:

Tier 1 — Statistical baseline. The tool estimates how likely a repeat of a given length is to occur by chance in a message of a given size (using a concept from probability theory called the birthday paradox). Shorter repeats in longer messages score lower.

Tier 2 — Cross-message corroboration. If the positions where a repeat occurs also show up as hot pairs across the full set of messages, the tool boosts confidence. A repeat that aligns with a known structural relationship is less likely to be a coincidence.

Low-confidence patterns appear at reduced opacity (more transparent). Brighter, more opaque highlights are the ones most worth paying attention to.

Tips for Reading the Data

Start with the heatmap bar. Look for bright columns—those are the structurally active positions. Positions that stay dark across many message lengths likely carry unpredictable data.

Look for consistency across messages. A pattern that appears in one message might be coincidence. A pattern at the same positions across dozens of messages is almost certainly structural. Click highlighted characters to check how many messages share the pattern.

Compare across lengths. Some structural features may persist across different message lengths (e.g., the first few positions might always serve as a callsign field). Jump between length pages to check.

Pay attention to opacity. Bright, solid highlights are high-confidence. Faint, transparent highlights are statistically weaker.

Purple is special. Purple highlights are the only ones validated across the entire message corpus rather than within a single message. They represent structural relationships between positions—arguably the most interesting finding, because they suggest the encoding scheme has internal dependencies where one part of the message is determined by another.