Charakteres: los signos ilegibles de la magia judía

Charakteres: the unreadable signs of Jewish magic

Ring-letters moved from Roman curse tablets into Hebrew amulets, survived a thousand years of copying, and never once gave up a meaning. This is a provenance-first reconstruction of what the signs are, where they can be dated, how they traveled, and why every attempt to read them has failed.

Forensic Series, DMTCode Research. All figures are original schematic tracings that follow the documented compositional grammar. None reproduces the exact strokes of a single object; the originals are photographed in the cited collections.

---

What counts as a charakter

Open almost any Jewish magical manuscript from the Cairo Genizah, or a Greek curse tablet from Roman North Africa, and somewhere on the page sit rows of small marks that behave like writing without being writing. They keep a baseline. They come in sequences. They sit next to real words. And they belong to no alphabet used anywhere in the surrounding text.

The working definition, following Gideon Bohak, is deliberately narrow. A charakter is a letter-like or simple ideographic sign that belongs neither to the script of the surrounding text nor to any known, stable symbol system. The definition excludes pictures of demons and deities, magic squares and word triangles built from ordinary letters, scribal punctuation, astrological and alchemical notation with fixed genre meanings, and true ciphers whose users held a recoverable key. The narrowness is a research tool. If every strange mark on a page is filed under one label, the corpus stops being comparable, and comparison is the only method that works on material this silent.

The generative rule

The place to start is not history but morphology, because the signs were made by a rule any epigrapher can state in one sentence. Take a letter, usually a rectilinear Greek letter, and place a small ring at each free stroke end.

Richard Gordon's survey of over a thousand distinct signs found the great majority to be variants on the Greek alphabet. The ring, the annulet, does all the work. It converts a unit of language into a unit of something else, a mark whose visible foreignness is the point. Scribes across the whole tradition understood which letters the rule wanted. Rings sit naturally on straight stroke ends, so the rectilinear letters dominate the corpus, and the curved letters, omicron, sigma, omega, are correspondingly rare. The distribution is mechanical, a fact about pen strokes, and it is the first hint that we are looking at a craft convention rather than a code.

No ancient writer explains the ring. Three modern accounts compete. Gordon reads it as deliberate graphic deviance: the ring flags the sign as non-standard, supernatural script, writing that announces itself as non-human. David Frankfurter reads the signs as celestial writing, the script of heaven or of the angels, cousins of star glyphs. Campbell Bonner proposed a descent from Egyptian protective loop signs, a theory Gyorgy Nemeth rejects on blunt quantitative grounds: well over ninety-eight percent of catalogued charakteres resemble no hieroglyph at all. Kirsten Dzwiza's dissertation catalogue, which secondary summaries report at 699 distinct sign types across 943 occurrences in the ritual handbooks, treats the wider corpus geometrically, as combinations of strokes, junctions, rings, enclosures, and appended elements. Read that way, the reusable object here is not a font. It is a shape grammar.

The oldest witnesses, and how firmly they stand

Ask when the signs begin and the popular answer, Ptolemaic Egypt, arrives quickly and without an object behind it. The archaeologist's discipline is to split the question four ways: the earliest named example, the strongest excavation context, the most precise associated date, and the earliest unquestioned ring-letter. In this corpus those are four different artifacts.

The earliest named ring-letter is a lead tablet from a grave at Autun, Roman Augustodunum, in Gaul. It is opisthographic, written on both faces, and side B carries the name ABRASAX bracketed by two X-shaped signs with rings at all four stroke ends. Nemeth calls it the earliest known charakter. Its date, the first third of the second century CE, rests on Latin letter forms, a palaeographic argument rather than a sealed stratigraphic one, and that is the single caveat on its title. There is a small joke buried in the layout: the Greek letters of Abrasax sum to 365 by isopsephy, a solar year bound into a name, flanked on either side by marks that refuse to be read at all.

The strongest reported excavation context belongs to a black steatite amulet from the Athenian Agora, published by T. Leslie Shear in the 1938 campaign report and attributed to a dated second-century deposit. If the deposit attribution holds, it is the best stratigraphic anchor among the early amulets. The exact inventory identifier could not be independently confirmed for this report, so the object stands here as bibliographically verified with its identifier an open field, which is the honest way to hold it.

The most precisely dated of the early objects is a silver lamella from a burial in the Becsi Street cemetery at Aquincum, found rolled inside a bronze bulla with a coin of Trajan giving a terminus post quem of AD 116. It carries Egyptianizing magical writing, including a sign of the kA type, two upraised arms on a base, alongside a Hebrew word for spirit. It is exactly the kind of object that tempts a researcher to push the ring-letter chronology earlier, and it must not be used that way. The kA sign is a closed hieroglyphic form. It has no letter skeleton and no terminal rings. Nemeth files it as a distinct category, and the drawing below shows why the classification matters.

The largest early transmission archive is the Amathous cache from Cyprus, more than two hundred fragments of selenite and lead curse tablets acquired in the late nineteenth century, probably from a single deposit, and commonly dated to the late second and third centuries. The tablets repeat sign sequences, among them an hourglass form, bowtie strokes with ringed corners, across many fragments, the fingerprint of professional practitioners working from shared models rather than improvising. Its weakness is the mirror of its strength: the findspot is reconstructed from dealers' archives and joining fragments, not from a controlled excavation.

The chronological result is plain. No true ring-letter earlier than the early second century CE was confirmed anywhere in the reviewed evidence. The older Philinna Papyrus contains verse spells and no signs, so it serves as a control rather than a counterexample. Egyptian incubation of the broader magical culture is plausible; a third-century-BCE origin for the ring-letter is not demonstrated by any object. One securely stratified pre-100 find with genuine ring-ended signs would move this floor. Until such a find is published, the floor stands.

A methodological rule follows and is worth stating because museum records so often violate it. An object carries up to four different dates, acquisition, publication, palaeographic estimate, and excavation, and they must never collapse into one field. An artifact can be perfectly authentic and still be weak evidence for place and time. The Amathous tablets are real; their deposit history is reconstructed. The Autun tablet is real; its date is a judgement about handwriting. Keeping those distinctions visible is most of what provenance discipline means.

The word is a fossil

Transmission arguments built on visual resemblance are weak, because simple shapes get reinvented. Loanwords are strong, because technical vocabulary travels with the technology it names. Here the linguistics does real work.

Greek charakteres is an agent noun from charassein, to cut or engrave: first the engraving tool, then the stamped impress, then, in the magical papyri, the special signs themselves. Jewish Aramaic and Hebrew texts borrow the word more or less whole, as karaqtiraya. In transmission the form also surfaces as kalaqtiraya, the first r shifting to l, the ordinary dissimilation of two liquids in one word that any Semitist will recognize. And then something better happens: copyists re-hear the now-opaque foreign word as native Aramaic, kol qtiraya, all the knots, qitra being a perfectly good Aramaic word for knot. A folk etymology grows inside the loanword, handing the imported signs an internal explanation, magic as binding, that the Greek never carried. The word is a small archaeological section through the borrowing itself: Greek technology, Aramaic naturalization, and a Jewish reinterpretation, all preserved in four syllables.

Late antique observers knew the category by name. Augustine, warning Christians off amulets in De doctrina Christiana, condemns cures that use marks which they call characters. That testimony proves the label and the practice were recognizable across religious lines by around 400. It explains nothing about origin, and decodes nothing.

Adoption: the absence before the signs

For the Jewish evidence, the strongest chronological control is negative. The earliest Jewish amulets carry no signs at all. The silver scrolls from Ketef Hinnom, the magical fragments among the Dead Sea Scrolls, the early Jewish amulets from Wales and from Halbturn in Austria: all of them are sign-free. From the fourth or fifth century CE onward the picture reverses, and Hebrew and Aramaic amulets from Palestine and the western diaspora display charakteres routinely, with the borrowed Greek name attached. Somewhere in those intervening generations Jewish practitioners adopted the foreign technology, slowly and practically, the way workshops adopt any useful tool. No surviving source names the first importer, the manual he copied from, the city, or the authority under which it happened. The signs were adoptable precisely because they were unreadable: tied to no obvious pagan god, they could be copied, re-addressed to angels, and naturalized.

Adoption was also uneven in space, and the unevenness is diagnostic. The western branch of Jewish magic, Palestine and the Mediterranean diaspora, absorbed the signs thoroughly, and that branch flows directly into the Cairo Genizah and the later manuscript tradition. The eastern branch looks different. Among the Jewish Babylonian incantation bowls, Bohak counts only three bowls bearing signs in a published catalogue of 122 bowl images, with the caveat that most bowls remain unpublished. So the sentence characteristic of Jewish magic is true of the tradition as a whole and false of its Babylonian medium, and any generalization has to say which branch it is standing on.

What the signs did on the page

Function has to be read from context, sign by sign, and the contexts refuse to converge on one job. Some texts address the kalaqtiraya directly, as holy powers commanded to act. Some present a sign or sequence as the seal of a named demon or angel. Some signs mark beginnings, endings, and divisions, doing the work of punctuation with none of its meaning. Some are plainly there to impress, mysterious marks as an advertisement of expertise. And everywhere the signs are copied ritual technology: the recipes instruct the practitioner to draw, engrave, or write them onto an object, and locate the operative moment in the act of inscription. The power was held to discharge when written, never when spoken. That is an attested historical belief about graphic efficacy. A historian can document the belief and the prescribed act. Whether anything further happened is outside what evidence of this kind can establish, and this report makes no claim about it.

The manuals show what a working line of signs looked like. The great London handbook, British Library Papyrus 121, known to scholarship as PGM VII, a third or fourth century ritual book purchased in 1888 out of the Theban antiquities trade, prescribes charakteres to be written on tablets cut from lead water pipes. The tablets themselves were never recovered. We hold the recipe and not the product, which is its own provenance category: proof of a prescribed practice, proof of nothing manufactured. The line below is built in the corpus grammar from six attested sign families, and one mechanical fact about such lines matters more than any of their shapes. Signs were typically used once or twice within a single spell and never redeployed as a reusable alphabet. Writing systems do not behave that way. Codes do not behave that way. This is the first structural proof that no code is present.

The rare signs that state their purpose

Against the general silence, a handful of figural devices carry their intent in their own words, and they are worth drawing precisely because they show what stated meaning looks like when antiquity actually provides it.

The best documented is the Chnoubis device of the stomach gems: a triple S crossed by a bar, paired on the stones with the written command pesse, pesse, digest, digest, addressed to the stomach itself. Galen independently records the use of such amulets for the stomach. And Galen supplies the twist that keeps the whole subject honest: he credits the green jasper, not the carving, and says the stone works just as well uncarved. Even where the ancient world documents an intent, the ancient authority may locate the power in the material rather than the glyph.

The recipe literature gives the ouroboros its job description outright. PGM VII directs that names and charakteres be enclosed by the tail-eating serpent as a phylactery, a bodyguard against daimones, against phantasms, against every sickness and suffering, on gold, silver, tin, or hieratic papyrus. The stated mechanism is the boundary itself, a drawn circle that seals what it surrounds.

The Testament of Solomon, a tradition of roughly the first to fifth centuries, says the seal on Solomon's ring is a pentalpha, five alphas in one continuous line, and that with it he bound the demons who built the Temple. The pentagram is the Solomonic seal in the same-period tradition; the six-pointed hexagram is a medieval substitution. Because the statement comes from a contemporary text rather than from an inscribed object, it sits one evidentiary tier down from the gems.

And one famous letter device is older than the whole ring-letter tradition and mute about itself. The Rotas square, a two-dimensional palindrome reading identically across, down, forward, and backward, is scratched on a column of the Palestra Grande at Pompeii, which fixes its existence before 79 CE. Its structure is ancient beyond argument. Every purpose ever attached to it, childbirth aid, protection from house fire, the Paternoster anagram, is a medieval accretion. The pagan original states no purpose at all. Existence, tier A; purposes, tier C; meaning, unsolved.

Copying that worked: two continuity chains

The strongest positive finding in the Jewish material is that sign sequences could survive centuries of hand copying recognizably intact. Two chains demonstrate it.

Clay shards excavated at Horvat Rimmon in southern Judaea, sixth or seventh century, carry a love incantation with a line of magical characters. A Cairo Genizah fragment of the eleventh or twelfth century transmits the same recipe, and its signs remain recognizably the same signs, roughly five hundred years downstream. Independently, a Genizah cloth amulet of the twelfth or thirteenth century, Cambridge T-S AS 142.174, matches a recipe copied in 1468 into a large compendium now in New York, NYPL Heb. 190, with near-identical instructions and a near-identical sign order, two or three centuries apart. In both chains the unit that survives is the sequence bound to its recipe, not the isolated sign. Copyists treated the marks as content worth exact preservation, which is itself evidence of how seriously the graphic package was taken.

The Genizah scribes also left behind the clearest evidence of how signs could be pressed into local service. In some fragments Hebrew letters are written above a row of charakteres, one letter per sign, a gloss that let a practitioner spell a target's name in signs on an amulet. The device is attested; the particular pairing drawn below is illustrative, because the manuscripts do not agree with one another.

Faithful copying is only half the transmission story. Set parallel copies of one recipe side by side and the drift becomes visible: a similar-looking sign substituted for another, a sign rotated in copying, a terminal ring dropped, a sign duplicated. Family resemblance survives while individual signs mutate, which is what hand transmission does to any graphic material, and it sets the tolerance any comparison method must build in. Match too loosely and every ringed X on three continents becomes one sign; match too strictly and genuine five-hundred-year chains like Horvat Rimmon dissolve into noise.

Sorting the strata: not everything ringed is a charakter

By the later Middle Ages a Jewish magical manuscript is an archaeological site in its own right, with several distinct graphic technologies deposited in layers, and the field's own habit of calling everything charakteres flattens exactly the history the signs preserve. A working taxonomy has to keep at least these apart: the standard Greek-derived ring-letters, the dominant imported family; Islamicate string letters, horizontal lines with pegged letter and number forms; the Arabic seven-sign seal of Solomon series, usually ringless, borrowed into Jewish books from Muslim ones; Latin sigilla and figurae, elaborate framed seals arriving from Christian Europe; planetary seals with their own framed circular conventions; archaizing Paleo-Hebrew-like signs deployed for an air of sacred antiquity; and outright false signs, miscopied Coptic, Demotic, Greek, crosses, and Christograms that later scribes carried along as meaningless marks. Where a manuscript labels its own signs, the practitioner's label comes first and the modern morphological class second.

One late stratum deserves its own drawing, because it inverts the founding rule. Medieval and early modern Jewish scribes eventually produced Hebrew charakteres, actual Hebrew letters given the ring treatment, and the practice reaches its most striking form in the Lurianic prayerbook tradition, where Siddur ha-ARI and Siddur ha-RASHASH write the Tetragrammaton itself in ringed letters, a device Christian Kabbalists such as Agrippa then copied. After a millennium in Jewish hands, the borrowed convention had become native enough to apply to the Name of God. It never displaced the old signs. As Bohak observes, the standard charakteres still look like Greek letters with ringlets at their tips, and still do today.

The decoder dream, and how it fails

From the Middle Ages forward, practitioners tried repeatedly to turn the inherited signs into readable alphabets of angels, and the surviving tables are the best evidence that no such reading ever existed. The alphabets carry magnificent attributions, the Alphabet of Metatron, the Script of Michael, the Writing of Gabriel. Set them side by side and they disagree on nearly everything: inventories of 22, 23, or 27 signs, repeated signs within one table, correspondences that contradict the next manuscript's, and, decisively, no ability to decode the sign sequences standing in the very books that transmit them. One notebook in the John Rylands Library, Gaster Hebrew MS 177, copies nineteen different angelic alphabets across four consecutive folios, and its scribe admits in places that he does not know whose alphabet he is copying. That is the profile of collectors amassing competing claims to authority. It is not the profile of a community preserving a key.

The seven seals of Solomon show the same process operating on borrowed material. The signs enter Jewish books from the Arabic tradition, and their canonical descriptions come from a poem attributed to Ali, more plausibly the work of al-Sharif al-Murtada, popularized by al-Buni, and received into Jewish literature through works like Toldot Adam. The poem describes shapes: rods with something like a lance above their heads, a blind and maimed mim, a ladder to all that is hoped for yet not a ladder, four fingers without the rest of the hand, an upside-down waw like the siphon of a blood-letter. The planet and day assignments belong to the standard Islamic scheme; the Jewish planetary table that exists is idiosyncratic; the per-seal meanings are, in Graham's words, rarely given and inconsistent across sources. The forms are older than every meaning attached to them.

Only once does the tradition build an actual mechanism from name to shape, and it is Renaissance, not ancient. Agrippa's De occulta philosophia of 1533 documents the kamea procedure: convert a name to Hebrew letters, the letters to their number values, and trace the path connecting those numbered cells on the planet's magic square; the resulting polyline is the sigil, with a small circle marking the first letter and a bar marking the last. It is a genuine, reproducible generation rule, arriving fourteen centuries after the Autun tablet, and its very novelty measures what antiquity lacked. The signs finally acquire a grammar of production at exactly the moment a Renaissance author invents one.

The verdict on decoding does not rest on the failed alphabets alone. It rests on how the signs behave in use. The same abstract sign recurs across love recipes, curse tablets, healing amulets, and protective phylacteries. Nemeth's study of the Hadrumetum tablets shows one workshop reusing four fixed sign sequences across its chariot curses regardless of target. The identical favor charm appears in two related Gaster manuscripts equipped with entirely different charakteres. A sign that floats freely between love and curse, and a charm that keeps its wording while swapping its signs, cannot be carrying its meaning in the glyph. Intent lives in the recipe's rubric. No ancient or medieval sign glossary ever existed to be lost; the medieval alphabets are letter keys at best, and never meaning keys.

Persistence without a key

The tradition never actually ended. Printed editions of Sefer Raziel, from 1701 onward, carried charakteres into typography, sometimes simplified or thinned by the medium. Handwritten Jewish magical manuscripts stayed dense with signs through the nineteenth and twentieth centuries, and Bohak records modern sign-bearing manuscripts and amulets from Kurdish, Yemenite, Moroccan, and Iraqi Jewish communities. These are witnesses to continuity of practice, and they carry no unbroken semantic knowledge with them; the last people who could have explained a sign, if anyone ever could, left no explanation.

The signs resist modern infrastructure the same way they resisted medieval decoders, and the resistance is diagnostic. Digital editions have encoded the same class of mark as a named glyph in one project, a generic figure in another, and an untranscribed drawing in a third, which quietly guarantees that no corpus-wide digital search is complete. A 2026 digital humanities project by Ortal-Paz Saar and Korshi Dosoo approaches the material as, in their phrase, text that was not text, requiring image and object detection where transcription fails. Almost a century of modern scholarship has not produced a successor to Bohak's survey with a complete updated corpus, and the basic data problem, connecting image, tracing, geometry, sequence, recipe, date, place, and uncertainty in one structure, remains open. For a project like this one, that gap is the opportunity: a provenance-linked vector corpus would be worth more than another speculative alphabet.

What cannot currently be recovered

The disciplined negatives, stated as findings, because in this field they are the findings.

No securely stratified ring-letter before roughly 100 CE. No ancient source naming an inventor, workshop, date, or design rule. No ancient explanation of the ring itself; deviance, celestial script, and Egyptian derivation are all modern reconstructions. No universal original alphabet; no stable inventory spans periods, languages, or regions. No corpus-wide decoder; local name encodings and late letter mappings exist, and none decodes independent sequences. No document of transfer into Judaism; the importing hands are anonymous. No complete object corpus, no complete survey of the Babylonian bowls beyond the published three-in-122 sample, and no complete account of the European Jewish manuscript signs, which Bohak singles out as a neglected class. And no access, by historical method, to whether any of it worked; the sources establish what practitioners expected of the signs, and expectation is where the evidence stops.

Two published thresholds would change the picture. A stratified pre-100 object with unmistakable ring-ended signs would move the chronological floor. A single early key that repeatedly and prospectively decoded independent sign sequences would overturn the decoder verdict. Neither exists in the record as reviewed.

Method note: how these figures were made, and how the symbol registry will use them

Every figure above is an original SVG schematic, drawn stroke by stroke from published descriptions, plates, and the documented shape grammar, and each carries its status in its caption: schematic reconstruction, described arrangement, attested device, tier of stated meaning. None is a facsimile, and none assigns a meaning its sources do not assign. Each figure exists as a standalone file with a manifest entry recording family, status, date floor, provenance summary, and sources, with the semantic and phonetic value fields deliberately null. The pipeline for any future addition runs capture, trace, segment, link, compare, review, in that order, and meaning is never inferred from geometry alone. The acceptance test is reproducibility: a second researcher must be able to open the cited witness, find the sign, tell tracing from facsimile, and reconstruct why two sequences were proposed as related. Unicode is not used as a sign's identity, because most of these marks have no codepoint, and forcing them into existing letters would erase the one distinction this whole subject depends on, the distance between an artifact's own form and our approximation of it.

Conclusion

The ring-letters are the master key to recognizing this technology and no key at all to reading it. They are securely on objects by the early second century CE, they crossed into Jewish practice in Late Antiquity carrying their Greek name with them, and Jewish scribes copied their sequences faithfully enough that a sixth-century potsherd and a twelfth-century Genizah fragment still agree. What never existed, on the full weight of the surviving evidence, is the thing every medieval alphabet-maker and more than one modern enthusiast has gone looking for: a fixed inventory, a pronunciation, a dictionary. The recoverable object is the transmission itself, shapes, sequences, media, recipes, and copying relationships, and it rewards exactly the kind of patient, provenance-first cataloguing this series exists to do. For any downstream summary, the defensible formula is short: a Greco-Egyptian magical sign technology, securely named by the early second century CE, ultimate origin unknown.

---

Sources

Bohak, "The Charakteres in Ancient and Medieval Jewish Magic," Acta Classica Universitatis Scientiarum Debreceniensis 47 (2011); Bohak, Ancient Jewish Magic: A History (Cambridge, 2008); Nemeth, "Magic Symbols (Charakteres) on North African Curse Tablets as a Regional Feature," Acta Classica 57 (2021); Nemeth, "Hieroglyphs in Greek Magical Texts?", Acta Classica 53 (2017); Gordon, "Signa nova et inaudita," MHNH 11 (2011); Dzwiza, Schriftverwendung in antiker Ritualpraxis (Erfurt dissertation, 2013); Marcillet-Jaubert, "Tabella defixionis Augustodunensis," ZPE 33 (1979); Shear, "The Campaign of 1937," Hesperia 7.3 (1938); Feher, "A Coptic Magical Name in a Pannonian Phylacterion," Acta Classica 52 (2016); Wilburn, "The Selenite and Lead Curse Tablets from Amathous, Cyprus," Religion in the Roman Empire 7.1 (2021); British Library Papyrus 121 (P. Lond. I 121, PGM VII) and PGM VII.579-590; Roman Inscriptions of Britain, Brit. 46.42 and Brit. 40.97; Posen Library record of the Horvat Rimmon potsherd incantation; Cambridge Genizah T-S AS 142.174; NYPL Ms. Heb. 190; Gaster Hebrew MS 177, John Rylands Library; Augustine, De doctrina Christiana II; Galen, Simple Drugs 9.2.19; Testament of Solomon; CIL IV 8623 (Pompeii Rotas square); Winkler, Siegel und Charaktere (1930); Graham (2011, 2014); Agrippa, De occulta philosophia II (1533); Delatte (1914); Bonner, Studies in Magical Amulets (1950); Frankfurter; Naveh and Shaked; Saar and Dosoo, "Magical Signs: The Text that Was not Text," DH Benelux 2026, Zenodo DOI 10.5281/zenodo.19659726; Campbell Bonner Magical Gems Database.

Reproducibility. Figure files and the machine-readable manifest accompany this article for the site's symbol registry. Dzwiza's corpus counts are reported here from secondary summaries, since direct collation of the dissertation was not possible in this build; the count of underlying handbooks is given variously as 87 or 94 and is flagged accordingly. The Athenian Agora amulet's inventory number is held as an open field rather than an invented identifier.