Encoded Message Catalogue

Encoded Message Catalogue

Executive Summary

The 128 × 128 Anna Matrix has been searched for embedded messages using four decode methods (XOR-127, column-pair XOR, mod-26 stream, deviation palindromes). An independent audit on 2026-05-12 re-derived the empirical statistics:

No encoded-message claim survives as significant. The K-e-y values at the diagonal positions [8,74][9,75][10,76] are arithmetically real, but under a fair null model the pattern is not unusual (185 of 2,000 controls, p ≈ 0.09). The earlier p ≈ 0.0005 came from a test that normalised decoded strings to "KEY" and compared them with "Key", so a hit was impossible, and that searched horizontal rather than diagonal windows.

The earlier-claimed clustering of messages in the 68 exception cells does NOT hold after audit: none of the falsified messages appear at exception-cell positions either.

Methodology

1. Decode Methods Applied

Four primary decode methods were applied systematically to the full 128x128 matrix:

Method A -- XOR-127 Transform: Each cell value is XOR-ed with 127 (the maximum positive value in 7-bit signed representation). The resulting byte stream is interpreted as ASCII characters.

decoded_char = chr(matrix[r][c] ^ 127)

This method exploits the matrix's point-symmetry identity (M[r,c] + M[127-r, 127-c] = -1), which in two's complement arithmetic is equivalent to M[r,c] XOR 0xFF = ~M[r,c]. The XOR-127 transform converts the symmetric complement into a readable character space.

Method B -- Column-Pair XOR: For each of the 64 mirror column pairs (C, 127-C), the XOR of corresponding cell values across all 128 rows is computed and interpreted as ASCII.

xor_value = matrix[row][col_a] ^ matrix[row][col_b]
decoded_char = chr(xor_value) if 32 <= xor_value <= 126 else '.'

This method targets the 4 exception column pairs where point symmetry is violated, since these pairs produce non-trivial XOR outputs.

Method C -- Modulo-26 Stream: Cell values are reduced modulo 26 and mapped to alphabetic characters (0=A, 1=B, ..., 25=Z). The resulting character stream is scanned for English words of length 3 or greater.

letter = chr((matrix[r][c] % 26) + ord('A'))

Method D -- Deviation Palindrome: The 68 cells that violate the symmetry rule produce deviation values. When ordered by position and interpreted as absolute ASCII codes, these deviations form a palindromic sequence.

2. Validation Protocol

Each candidate message was tested for statistical significance using the following protocol:

1. Baseline generation: 1,000 to 10,000 random matrices were generated with the same symmetry constraint as the Anna Matrix (99.58% point-symmetric with 68 exceptions)
2. Decode application: The same decode method was applied to every random matrix
3. Frequency counting: The number of occurrences of the candidate word in random matrices was recorded
4. p-value computation: p = (count of random matrices with >= observed occurrences) / (total random matrices)

A message is classified as validated if p < 0.05 and highly significant if p < 0.01.

3. Multiple-Testing Correction

With multiple decode methods and many possible word matches, multiple-testing correction is essential. Bonferroni correction was applied: for k independent tests, the significance threshold is adjusted to alpha/k. With approximately 20 candidate words tested, the corrected threshold is p < 0.0025. After the 2026 audits, no message meets it.

Validated Message Catalogue

K-e-y at M[8,74][9,75][10,76] (values real, not significant)

The bytes (75, 101, 121) spell "Key" in ASCII. All three cells also satisfy the complement rule, so they are ordinary bot replies, not exception cells. Reproduce the values:

import json
M = json.load(open('apps/web/public/data/anna-matrix-min.json'))['matrix']
print(chr(abs(M[8][74])) + chr(abs(M[9][75])) + chr(abs(M[10][76])))
# Output: Key

✗ Falsified: GAME (XOR-127)

The original Monte Carlo claim ("p < 0.0001, 0 of 1,000 random matrices matched") is technically true in the trivial sense that the BASE rate is also zero, neither Anna nor random matrices contain "GAME" in their XOR-127 streams. The relative claim "GAME appears in Anna's stream" is false. Verifiable by:

stream = [abs(M[r][c]) ^ 127 for r in range(128) for c in range(128)]
text = ''.join(chr(b) if 32 <= b <= 126 else '·' for b in stream)
print(text.count('GAME')) # Output: 0

✗ Falsified: CFB at Row 11

Neither the |cell| nor XOR-127 transform yields "CFB" at Row 11. The originally-claimed Bonferroni-surviving p-value 0.0058 was based on a transform that doesn't decode to "CFB" at this position. The "Population B / period-4 conductor" architectural narrative around this position remains correct as a structural observation, but the "CFB signature" label is unsupported.

✗ Falsified: MEGA (XOR-127)

✗ Falsified: AI MEG GOU (Column-pair XOR)

The "AI", "MEG", "GOU" tokens may individually appear (any 2-3 letter sequence is likely to occur in any 16K-byte stream by chance, e.g. "AI" appears 25 times in the mod-26 stream alone), but the specific assembly into "AI MEG GOU" is not reproducible at the claimed positions. The original p < 0.0001 claim used a multiple-test-uncorrected statistic; after correction for the ~20 candidate words × 4 decode methods ≈ 80 tests, the marginal occurrences are not significant.

⚠ Weak: RISE (Mod-26 stream)

The single occurrence is real, the p-value is real and matches the original claim, but it does not survive multiple-testing correction. Treat as anecdote, not evidence.

Other claims (not re-audited in this pass)

The following were in the original document. They have not been independently re-verified by audit script and should be considered UNVERIFIED until tested:

• Message 5: ">FIB" at column pair (22, 105), Rows 27-30, original disclosed it was never Monte-Carlo tested. Architectural narrative; weak evidence.
• Message 7: ASCII Deviation Palindrome K 8 u Z x ~ ~ x Z u 8 K structural property of the 34 deviation magnitudes. Not a statistical claim; it either is or is not a palindrome. Re-verification deferred.

The palindrome reinforces the evidence that the 68 exception cells were deliberately placed. A random perturbation of a symmetric matrix would not produce a palindromic deviation sequence.

Structural Context of Validated Messages

Population Alignment

The validated messages cluster overwhelmingly in Population B of the matrix's neural architecture:

Population B (43 neurons) serves as the conductor in the period-4 attractor cycle, flipping during transitions 1-2 and 3-0 while Populations A and A' remain static. The concentration of creator and system-identity messages in the conductor population suggests intentional placement in the architecturally most significant component.

Exception Column Distribution

The two column pairs that yield readable messages (30/97 and 22/105) together account for 91% of all exception cells. This concentration is itself statistically notable and suggests that the exception cells serve a dual purpose: structural correction for the neural network dynamics and information encoding.

XOR Triangle at Position [22, 22]

The cell at matrix position [22, 22] is the only self-matching asymmetric cell -- it and its mirror position [105, 105] both contain the value +100. This produces a closed arithmetic relationship:

100 XOR 127 = 27
100 XOR 27 = 127
27 XOR 127 = 100

The three values 127 form a closed XOR triangle. The number 27 (= 3 cubed) is one of the matrix's dominant cell values, appearing as part of the 26/-27 mirror pair that satisfies the symmetry rule (26 + (-27) = -1). The number 127 is the maximum positive value in 7-bit signed representation and defines the column-pair architecture (C + C_mirror = 127).

This relationship is an arithmetic fact and does not require statistical validation. Its significance lies in connecting three structurally important constants through a single cell position.

Gematria Checksums

Row-Level Character Sums

When cell values at specific row positions are interpreted as alphabetic indices and summed (A=1, ..., Z=26), certain rows produce character sums that coincide with structurally significant constants:

These sums were identified through systematic scanning. However, the number of possible row subsequences is large (of order 10 thousand for 2-character combinations), and no Bonferroni-corrected significance test has been applied to these particular sums. They are documented as observations.

Negative Results: Messages That Failed Validation

The following candidate messages were tested and did not achieve statistical significance. They are documented to prevent re-investigation of known dead ends.

Failed Message Candidates

Key Negative Finding: Bitcoin Timestamps

A specific test was conducted for Bitcoin-era timestamps (2009-2010 date ranges encoded as Unix timestamps in adjacent cell pairs). The Anna Matrix contains 389 such timestamps, compared to a random baseline mean of 479.7 (standard deviation 30.0). The matrix has significantly fewer Bitcoin timestamps than expected by chance (z-score = -3.02), definitively ruling out embedded Bitcoin temporal data.

Composite Message Interpretation

Combining all validated messages in order of statistical significance yields a coherent, if minimal, composite message:

CFB (creator signature) --> AI MEG[A] (system identifier) --> GAME (concept)

Proposed reading: "This is CFB's creation. An AI system. The GAME."

This composite interpretation is internally consistent but should be treated as Tier 3 (interpretation). The individual message validations are Tier 1 (reproducible). The semantic synthesis connecting them into a narrative is an inference that cannot be statistically tested.

Limitations

1. Decode Method Selection

The four decode methods (XOR-127, column-pair XOR, mod-26, deviation palindrome) were not specified a priori by the matrix's creator. They were discovered through exploratory analysis, which introduces the risk of method selection bias: given enough decode methods, some will produce seemingly meaningful patterns by chance. The Monte Carlo validation partially addresses this concern but cannot eliminate it entirely.

2. Multiple-Testing Burden

Approximately 20 candidate words were tested across 4 decode methods, yielding roughly 80 individual tests. The Bonferroni-corrected threshold of p < 0.000625 is met by GAME, MEGA, CFB, and AI MEG GOU, but not by RISE or the gematria sums. Future analyses should pre-register hypotheses before testing.

3. Semantic Interpretation

The validated messages are short (3-8 characters each). Statistical methods can confirm that specific character sequences appear more frequently than chance, but they cannot determine the intended meaning. "GAME" could refer to game theory, a literal game, or be a coincidental arrangement. "AI MEG GOU" could represent "AI MEGA GOLD" or be an unrelated sequence that happens to contain recognisable fragments.

4. Context Dependency

Several messages gain significance partly from their architectural placement (exception columns, Population B membership). While the spatial correlation is independently testable, the interpretation of why messages cluster in architecturally significant regions is inherently model-dependent.

5. Reproducibility Requirements

All findings require:

• The public 128x128 matrix file anna-matrix.json (compiled by hand from bot replies; it is not part of the Aigarth-it repository)
• Python 3.11+ with NumPy
• The validation scripts referenced in this document

Without access to the original matrix data, no finding can be independently verified.

Verification Scripts

Summary Table

Related Documentation

Analysis date: February 2026, revised May and September 2026. The p-values in the historical tables are the original claims; the verdicts in the callouts replace them.