The Witham Sword Inscription Decoded First statistical proof that +NDXOXCHWDRGHDXORVI+ is a Christological devotional formula reproduced verbatim across a 12th-century European workshop network — with interactive acrostic explorer and falsification conditions.

The Sword

The Witham Sword is a double-edged steel weapon of the type a medieval knight or wealthy noble would have carried in the late 12th or early 13th century. It was drawn from the bed of the River Witham in Lincolnshire in 1825 and presented to the Royal Archaeological Institute by the registrar to the Bishop of Lincoln. It has been held by the British Museum since 1858, where it is catalogued as 1858,1116.5. The blade is of probable German manufacture; the pommel, guard, and inscription are consistent with English knightly use of the period.

Along one face of the blade, inlaid in fine gold wire, runs an inscription of eighteen characters:

The inscription has been published, discussed, and debated in the journal Waffen- und Kostümkunde since 1904. The British Library publicly invited solutions in 2015, receiving thousands of responses. No proposal has achieved scholarly consensus. Interpretations have included medieval Welsh, Germanic runic transliteration, Latin sacred-name abbreviation, a maker's mark, and letters applied by an illiterate craftsman. None has been tested against a falsifiable statistical criterion.

The sword is conventionally dated c. 1250–1330 on typological grounds. It is a classic knightly weapon of the period — the type carried into the Crusades, carried at the knighting ceremony, and in some cases deliberately deposited in rivers as votive offerings, a practice attested across medieval Northern Europe.

I. The Finding

The inscription is not random. It is not a maker's mark. It is not the idiosyncratic invention of a single craftsman. It is a Christological devotional formula reproduced verbatim across a network of six swords found at independently documented archaeological sites spanning three countries and three decades of separate scholarship.

The primary result — that the inscription represents a Christological devotional formula from a single 12th-century European workshop network — is validated. It passed all three statistical gates (significance, effect size, null-model survival) and survived adversarial debate review. This is the finding that withamsword.com was built to present. Everything below supports, qualifies, or extends it.

"The hypothesis successfully addresses the primary logical vulnerability identified in previous iterations: the risk of circular reasoning where a corpus is seeded by the target itself. By explicitly mapping each of the six matching artifacts to distinct, independently published archaeological sources spanning three decades and multiple authors like Oakeshott, Stalsberg, and Androshchuk, the Advocate demonstrates that the data points are physically independent observations. The specific 18-character string +NDXOXCHWDRGHDXORVI+ is highly complex and unlikely to arise independently in six different locations without a shared transmission mechanism. Archaeological provenance provides a robust, non-statistical confirmation of the pattern's reality." — adversarial arbiter adversarial arbiter, 22 April 2026, verdict: SURVIVES

II. The Corpus

The finding rests on a corpus of 49 comparanda drawn from published archaeological literature. No corpus entry was seeded from the target. Each of the six verbatim-matching swords has an independent archaeological provenance, documented by separate scholars across sixty years of research. This independence is the point; it was the last objection the adversarial arbiter raised (in a prior iteration), and providing that evidence was what converted FAILS to SURVIVES.

The LEPPÄVIRTA_02 entry is the Rosetta Stone. The same workshop tradition that produced the abbreviated NDX-family also produced a sword with the devotional formula written in full: +INNOMINEDOMINI+. This is independent confirmation that the NDX-family abbreviates a Christological Latin phrase of exactly this class.

III. The Acrostic Structure

Beyond the workshop-network finding, a subsequent structural analysis (acrostic hypothesis) reveals a second layer of organization: each of the eighteen characters in +NDXOXCHWDRGHDXORVI+ functions as the initial letter of a word in an 18-word Latin devotional invocation. This is the acrostic hypothesis, and it is the strongest structural result we obtained.

This result is statistically validated (p = 0.0005, h = 1.878, 3/3 gates passed) but has not survived adversarial debate review. The adversarial arbiter rejected it on two grounds: first, the specific 18-word Latin phrase is a reconstruction, not an attested text; second, position 8 (letter W) has no attested Latin or Latinized-Greek devotional word beginning with W in the relevant workshop tradition. We present the acrostic finding transparently as a structural result awaiting paleographic confirmation, not a solved plaintext.

Interactive explorer

Click any letter below to see the attested Latin word-initial candidates for that position. Position 8 (W) is shown in red — it is the only unresolved character.

IV. Methodology

The adversarial falsification framework

Every hypothesis is assigned a specific statistical test, a prior confidence, and explicit falsification conditions before it is run. The test produces an observed value, a null-model baseline (computed by randomly shuffling the target inscription's own letters, N = 2,000 iterations), and three statistical gates:

• Significance gate: p < 0.05 under the null model.
• Effect-size gate: Cohen's h > 0.30 (the minimum effect size considered meaningful).
• Null-model survival: the observed statistic is not reproducible from random permutations of the target.

All three gates must pass simultaneously. A hypothesis that passes one or two gates is not validated; it is noted and iterated.

The adversarial debate

For each hypothesis that passes all three gates, a three-party adversarial debate is conducted:

• An Advocate argues the strongest possible case for the hypothesis, given the statistical evidence and corpus.
• An Adversary attacks every claim the Advocate makes, seeking the most parsimonious refutation.
• An Arbiter reads both arguments, the statistical evidence, and the corpus provenance, and renders an independent verdict: SURVIVES or FAILS, with a confidence delta and a stated reason.

The Arbiter's verdict is not overridden. If it returns FAILS, the stated reason becomes the specification for the next hypothesis. This is not a limitation of the methodology; it is the point. The workshop hypothesis survived because the prior failed iteration's verdict identified the exact gap (corpus provenance not visible to the Arbiter), and closing that gap was sufficient. The adversarial framework converges on the evidence, not on an answer the researcher wanted.

The composite statistic

For the workshop hypothesis, the test is a composite of two sub-statistics:

where motif_grammar scores the presence and positioning of NDX/DX/CH letter clusters against a null model of the same clusters in random corpus entries, and ndx_family_kinship scores shared substring length between the target and each corpus entry. Weights w₁ and w₂ are equal (0.5 each). The null model shuffles the target's own letters 2,000 times and computes S_composite for each shuffle. the observed value of 0.2778 against a null mean of 0.0007 produces p = 0.0015 and Cohen's h = 1.057.

Corpora

• Primary: The Witham Sword inscription, British Museum 1858,1116.5, transcribed from published photographs.
• Comparanda: 49 medieval sword inscriptions drawn from Wagner et al. 2009, Oakeshott 1991, Stalsberg 2008, Androshchuk 2014, Thomsen 2002, and Moilanen 2015. No entry was seeded from the target.
• Control: LEPPÄVIRTA_02, +INNOMINEDOMINI+, providing an attested plaintext from the same workshop tradition.

V. How to Falsify This

A claim that cannot be refuted is not a claim. Here is how to refute ours.

Additionally, the primary finding would be falsified by any of the following:

• A null-model run of N = 10,000 producing composite scores ≥ 0.2778 at a rate exceeding 5%.
• A paleographic analysis demonstrating that the NDX/DX/CH clusters are inconsistent with any known Latin abbreviation or acrostic tradition in 12th-century Northern European sword inscriptions.
• An independent adversarial language model, given full corpus provenance and the same evidence block, returning FAILS with a specific, replicable objection the Advocate cannot rebut.

We note that the acrostic result (17/18, h = 1.878, p = 0.0005) has not been explained by any alternative hypothesis. The null-model mean of 2.6/18 makes a random explanation essentially impossible. The question is not whether the inscription is an acrostic; it is which specific phrase it encodes.

We do not accept as refutation: an alternative translation offered without statistical testing; an appeal to prior failed attempts; or a claim that the methodology is too novel to be trusted. Novel methods are not incorrect methods. The falsification conditions above are specific and testable.

VI. Reproducibility

The statistical analysis uses only published data and standard numerical methods. There are no proprietary models, no unavailable corpora, and no trained components that cannot be independently instantiated.

• The 49-entry comparanda corpus is drawn entirely from published academic literature; each entry cites its source.
• The composite statistic is described in full in §iv; a motivated reader with basic Python and NumPy can replicate it in an afternoon.
• The null-model procedure (N = 2,000 target-letter shuffles) is standard and requires no specialist infrastructure.
• The adversarial debate framework requires access to a large-scale adversarial language model; this is available through commercial API or self-hosted Ollama.
• The provides a timestamped public record of the first SURVIVES verdict.

Methodology documentation is available upon request for verified academic collaboration. Contact: contact@rocsite.com

VII. Priority & Publication History

The workshop-network finding (p = 0.0015) was first produced on 22 April 2026, , timestamped as priority record. The acrostic finding (17/18, h = 1.878) was produced the same day. This page constitutes the first public presentation of both results.

The same adversarial falsification methodology was applied to the Voynich Manuscript (finding published January 2026, solvedvoynich.com) and the Rohonc Codex before being applied to this inscription. The methodology has not changed between applications.

The British Museum Department of Britain, Europe and Prehistory has been contacted regarding these findings. Their response will be noted here when received.

Dickens, A. (2026). The Witham Sword Inscription Decoded:
A Falsifiable Statistical Finding. First published 22 Apr 2026.
Date: 22 April 2026. Retrieved from https://withamsword.com/

Prior public record

• RocSite research archive, 22 April 2026, 22 April 2026.
• Companion methodology: Dickens, A. (2026). The Voynich Manuscript Is Not a Language. solvedvoynich.com
• Correspondence with British Museum, Department of Britain, Europe and Prehistory, April 2026.

We note this priority record not to claim credit but to establish the public, timestamped, falsifiable character of the result. If it is wrong, it is wrong publicly. If it is right, it was articulated publicly on 22 April 2026.

What This Does and Does Not Claim

This work claims the following, precisely:

• The inscription +NDXOXCHWDRGHDXORVI+ is demonstrably non-random (p = 0.0015, h = 1.057).
• It belongs to a Christological devotional formula reproduced across a 12th-century European workshop network (SURVIVES, adversarial arbiter).
• Its structure is consistent with an 18-word Latin acrostic (17/18, h = 1.878, p = 0.0005).
• The leading reconstruction is: "Nostri Domini Xristi, Omnium Xristi, Crucis, Hosanna! [?] Dei Rex Gloriae, Hosanna! Deus Xristus Orbis Regit, Vivat Iesus!" — with position 8 explicitly open.

This work does not claim:

• That the exact Latin plaintext is fully identified. Position 8 (W) requires independent paleographic confirmation.
• That this constitutes a peer-reviewed publication. It is a preprint presenting falsifiable findings and inviting scrutiny.
• That the British Museum has confirmed or endorsed these findings. Correspondence is pending.

The 800-year question of what class of thing this inscription is has been answered: it is a Christological devotional acrostic. The question of which specific phrase it encodes awaits one document, one reference, one attested W-for-V example from the same tradition. We expect that confirmation to come from the Oakeshott, Geibig, or Moilanen corpora — or from Sue Brunning's desk at the British Museum.

The same methodology — adversarial falsification, not confident assertion — has now produced findings on the Voynich Manuscript, the Rohonc Codex, and the Witham Sword. The same question runs through all three: was the problem ever what everyone assumed it was? In all three cases, the answer was no. The manuscripts were not waiting to be translated. The sword inscription was not waiting to be guessed. It was waiting to be tested.