Frame as Citable Parameter

Abstract

Computational translation of esoteric primary text presents a problem orthogonal to those addressed by general-purpose machine translation: the doctrinal content of mystical, contemplative, and apophatic literature is conditioned by the hermeneutic frame the translator brings to the source, and human translation history demonstrates that translators rarely declare the frame they are operating under. We present a single-passage three-reception experiment with the Hekhal Targum engine on the closing of Plotinus's Enneads VI.9.11. The same Greek source is rendered three times under three frame controllers (kataphatic-apophatic, zahir-batin, and a deliberately misapplied PaRDeS as anachronism control), with all other variables held fixed. We report four findings. (1) Orthographic disposition of the closing μόνου πρὸς μόνον tracks frame, with each capitalization choice tied in audit trail to a specific frame relation. (2) Apparatus density (range cards, cross-references, footnotes) varies systematically by frame and is itself a frame-tracking metric. (3) The deliberately misapplied frame (PaRDeS on Plotinus) elicits register-decorum awareness from the engine: the controller is suppressed in the body, the misapplication is explicitly flagged in the audit trail, and cross-tradition resonance surfaces only as homology in apparatus. (4) Cross-run experimental design caught a hallucinated MacKenna citation that single-run verification would not have detected, suggesting a generalizable check on LLM citation reliability for canonical public-domain translations of widely-paraphrased classical sources. The experiment articulates the methodological position that hermeneutic frame is a tunable, citable parameter of computational translation rather than a hidden default, and that frame-disclosure can be an empirical contribution rather than a philosophical posture.

Keywords: computational translation, hermeneutic frame, Plotinus, Pseudo-Dionysius, Akbarian Sufism, Kabbalah, apophatic theology, reception studies, large language models, audit-trail provenance.

1. Introduction

Most of the world's mystical, contemplative, and esoteric primary literature is functionally untranslated. Modern critical English translations exist for a small fraction of the Aramaic Zohar, the Heikhalot synoptic, the Lurianic primary texts, Ibn Arabi's Futuhat al-Makkiyya, the Ismaili da'wa literature, the Hesychast desert fathers beyond the Philokalia, the Hermetic corpus past Mead 1906, and entire traditions further afield. What translations do exist are largely paywalled or out of print. Serious primary-source access in these traditions is a privilege of the academic library system, not a public utility. The texts that go untranslated are precisely the texts where original-language manuscript culture, technical interpretive vocabulary, and the editorial discipline of a tradition matter most.

Large language models have made provisional translation of arbitrary text accessible at consumer scale. The translations they produce are, for esoteric primary text specifically, unreliable in characteristic ways: technical vocabulary drifts across passages within a single tradition (the Kabbalistic sefirot rendered "emanations" in one passage and "numerations" in another); doctrinally-live ambiguities that traditional commentaries preserve are smoothed away in favor of single-sense readings; the hermeneutic frame the translator imposes is hidden rather than declared. The last problem is the deepest. Every translator of esoteric primary text has a frame -- Catholic, Orthodox, academic-Platonist, Sufi, Theosophical, scholastic. The frame conditions the rendering. Pseudo-Dionysius's ὑπερούσιος reaches English through Parker (1897, scholastic super-essential), Rolt (1920, restrained beyond-being), and Luibheid (1987, modern-eminence higher-than-being); each rendering is recognizable as the product of a particular interpretive position. None of these positions is illegitimate. What is missing is the position's citation.

The Hekhal Targum engine is built around the position that the frame should be cited. The engine's frame controllers are explicit, version-controlled, swappable, and audited. Every rendering carries the frame in its metadata; every preserved ambiguity is flagged with the frame relation that produced the flag; every controlled-vocabulary choice is traced back to a glossary revision. This paper presents an experiment that tests the position empirically. We render the closing of Plotinus's Enneads VI.9.11 -- the most-cited passage in the Plotinian corpus and the locus classicus of the contemplative coincidence figure -- through three different frame controllers, holding every other variable fixed. Two of the frames model documented historical reception (Christian apophatic via Pseudo-Dionysius; Akbarian Sufi via the Arabic Theology of Aristotle); the third is a deliberate anachronism control (PaRDeS retrojected onto Plotinus, where no documented Kabbalistic reception of Plotinus exists). The three resulting renderings, their audit trails, and the engine's response to the deliberately misapplied third frame are the experiment's data.

Section 2 situates the experiment in the existing literature. Section 3 summarizes the engine architecture sufficient for non-specialist readers. Section 4 specifies the experimental design. Section 5 reports the results, organized around four findings. Section 6 discusses the philosophical and methodological consequences. Section 7 names the limitations and queues future work. Section 8 documents the reproducibility surface. Real Anthropic API calls were made on 9 May 2026 against the Targum production pipeline using Claude Opus 4.7; total Anthropic spend across the three runs was approximately 0.50 USD; all artifacts are preserved at paths named in §8.

2. Background and related work

2.1 Apophatic theology and the Plotinian inheritance

The closing of Enneads VI.9.11 -- φυγὴ μόνου πρὸς μόνον -- is the locus classicus of Plotinian henosis, the contemplative terminus reached through progressive simplification (haplosis) past Soul, past Intellect, to coincidence with the One. Standard treatments include Rist (1967), Hadot (1993), and the Armstrong Loeb apparatus (1988). The passage's significance for the present experiment is that it carries documented historical reception into both the Christian apophatic tradition and into Islamic Neoplatonism, while carrying no documented Kabbalistic reception. This asymmetry is the experiment's premise.

The Christian apophatic reception runs through Pseudo-Dionysius the Areopagite (writing c. 500 CE, within a generation of Damascius and the closing of the Athenian Academy in 529). Pseudo-Dionysius synthesizes the Late Platonic apophatic register into a Christian theological framework that systematizes the kataphatic and apophatic moments as a contemplative methodology in the Mystical Theology and the Divine Names. The reception extends through Maximus the Confessor's commentary, Eriugena's Latin translation, and downstream Christian apophatic tradition (Eckhart, the Cloud of Unknowing, Lossky, Yannaras). Standard treatments: Rorem (1993); Perl (2007); Schaefer (2006); Wear and Dillon (2007); Louth (2007); Carabine (1995); Sells (1994); Turner (1995). The Plotinian-Christian distinction between henosis (contemplative-ontological terminus) and theosis (doctrinal-soteriological participation grounded in the Incarnation) is articulated by Russell (2004) and Bradshaw (2004); the distinction is load-bearing for any translation that respects the Plotinian register without retrojecting Christian doctrinal apparatus.

The Islamic Neoplatonic reception runs through the Arabic Theology of Aristotle (c. 832 CE), a paraphrase of Enneads IV-VI compiled in al-Kindi's circle and misattributed to Aristotle. The paraphrase reaches al-Kindi, the Brethren of Purity, Suhrawardi's Illuminationism, and ultimately the Akbarian register through mediated channels. Standard treatments: Adamson (2002); D'Ancona contributions to Adamson and Taylor (2005). The Akbarian apophatic register operates around wujud (being), tajalli (self-disclosure), fana/baqa (passing-away/subsistence), and the ahadiyya/wahidiyya distinction (non-relational unity prior to the divine names vs. relational unity through the names). Standard treatments: Chittick (1989); Chodkiewicz (1993); Corbin (1969); Schimmel (1975); Sells (1996).

No comparable Plotinian-Kabbalistic reception chain exists. The medieval Jewish reception of Neoplatonism reaches Maimonides and his successors through Aristotelian-Avicennan channels; the Renaissance Pico-Reuchlin synthesis of Kabbalah and Neoplatonism is a much later cross-pollination that does not constitute a Plotinian-Kabbalistic reception comparable to the Christian or Akbarian chains. The PaRDeS hermeneutic itself is a thirteenth-century Provençal-Catalan theosophical articulation operating on Hebrew and Aramaic source material; standard treatments include Scholem (1990), Idel (2002), and the methodological discussion in van der Heide (1983). Applying PaRDeS to a third-century Greek Platonist text is therefore a defensible move only as a methodological control variable, not as a claim of historical reception. This is how the third run of the experiment is framed in §4.

2.2 Translation studies and reception history

The translation studies literature has established for decades that translation is interpretation, and that interpretation is conditioned by the translator's position. Toury (1995) and Venuti (1995) provide the standard framing of translation as a culturally and ideologically situated act. Reception studies, classically applied to ancient sources, makes the same argument with diachronic empirical data: the same text is received differently across periods and audiences, and the differences are interpretively load-bearing rather than incidental. Hardwick (2003) and Hardwick and Stray (2008) summarize the classical reception studies methodology; Catana (2013) addresses the specific reception of Late Ancient Platonism.

Henry Corbin's lifelong articulation of l'imagination créatrice in Sufi Neoplatonism (1969) makes a stronger claim: that Plotinus-in-Greek and Plotinus-in-Arabic are not the same text under translation but distinct living texts whose distinctness is generated by the receiving tradition's hermeneutic frame. Corbin's position is the philosophical commitment the present experiment operationalizes computationally. Where Corbin demonstrates the claim through close reading and historical reconstruction, the present experiment demonstrates it through a controlled computational test: the same Greek source rendered through three frames produces three different English texts whose differences are traceable in audit trail to the controlling frame.

2.3 Computational translation of religious and esoteric text

Explicit prior work on hermeneutic-frame-aware large language model translation is sparse. The general-purpose machine translation literature has not engaged the apophatic-register problem; mainstream MT evaluation (BLEU, chrF, COMET) operates on assumptions of single-sense reference translation that do not hold for esoteric primary text. The digital humanities literature on critical text editing (TEI markup, prosopographical structuring, manuscript-variant encoding) operates at a different layer of the editorial pipeline than the present experiment. The closest existing work is in computational classics on Greek-to-English translation evaluation; cross-translator alignment work for ancient source material remains at an early stage. The position the present paper occupies -- computationally controlled hermeneutic frame as a first-class translation parameter -- is, to our knowledge, unrepresented in the published literature on machine translation of religious or esoteric primary text. The contribution of this paper should therefore be read both as an experimental result and as a methodological proposal for the field.

3. The Targum engine

We summarize the engine architecture sufficient to make the experiment intelligible to non-specialist readers. Full architectural specification is at docs/TARGUM-ENGINE.md in the project repository.

Targum is a seven-layer pipeline. Layer 0 resolves canonical references across upstream corpus providers (OpenITI for Arabic, Sefaria for Hebrew/Aramaic, Perseus for Greek/Latin, GRETIL for Sanskrit, Hekhal-internal for unaligned material). Layer 1 runs the source through a multilingual morphology router (CAMeL Tools, Dicta, CLTK, Stanza, with a Unicode-aware naive fallback for any language). Layer 2 is the controlled glossary system: per-corpus YAML with hard-constraint injection at generation time, drift audit against versioned revisions, and bidirectional sync with the project's lexicon. Layer 3 is hybrid retrieval-augmented generation over a curated scholarly corpus: BM25 lexical retrieval and dense retrieval (multilingual e5-base, 768-dim, ChromaDB persistent index) combined via reciprocal rank fusion at the standard k=60 constant. Layer 4 is the generation orchestrator, which assembles a prompt that includes the corpus's frame controller, calls the language model via tool-use mode with a JSON Schema that enforces the output shape, validates the response, and regenerates on scope violations up to two retries. Layer 5 is the canonical output schema: every chunk produces a JSON object with multi-pass translation (crib / faithful / fluent / sod-commentary), per-term semantic-range cards, ambiguity-preservation entries with frame notes, structured apparatus (footnotes, manuscript variants, cross-references, decision notes), comparison-to-PD-translations array, and a queryable audit trail. Layer 6 is editor sign-off and publication: machine-assisted drafts require human review before flipping to verified.

For the present experiment the load-bearing layer is Layer 2 -- specifically, the frame controller selection in backend/frames.py. The function select_active(frames, corpus, *, explicit=...) returns the controllers active for a chunk, with the explicit argument overriding automatic corpus-trigger matching when provided. Each controller is a YAML file specifying frame-id, version, display-name, triggers (corpus-in list and source-language signals), interpretive senses, named relations, output validators, and worked examples. For the experiment, we set the corpus assignment per spec to fire each desired frame; the existing engine architecture supports this without modification.

4. Experimental design

4.1 Source

The closing sentence of Plotinus, Enneads VI.9.11. Twenty-one Greek words. The Greek text follows the Henry-Schwyzer critical edition convention (Oxford Classical Texts, 1964-1982); the closing phrase φυγὴ μόνου πρὸς μόνον is universally attested across editions:

Καὶ οὗτος θεῶν καὶ ἀνθρώπων θείων καὶ εὐδαιμόνων βίος, ἀπαλλαγὴ τῶν ἄλλων τῶν τῇδε, βίος ἀνήδονος τῶν τῇδε, φυγὴ μόνου πρὸς μόνον.

The MacKenna 1930 public-domain English (verified against Sacred-Texts via Internet Archive Wayback Machine, snapshot 2024-05-25 of sacred-texts.com/cla/plotenn/enn714.htm):

"This is the life of gods and of the godlike and blessed among men, liberation from the alien that besets us here, a life taking no pleasure in the things of earth, the passing of solitary to solitary."

4.2 Three reception lenses

Three benchmark specifications were authored at benchmarks/specs/plotinus-vi-9-11-{christian-reception,akbarian-reception,pardes-anachronism}.yaml. The specifications are byte-identical in source text and differ only in the corpus field, which simultaneously fires the corresponding frame controller (via the controller's triggers.corpus_in list), loads the corresponding glossary (which is keyed by (corpus, school)), and filters the dense retriever to the corresponding scholarly corpus collection. Table 1 summarizes the design.

Run C is published with the anachronism flagged in spec metadata (the experimental.anachronism: true and experimental.anachronism_kind: deliberate-control-variable fields), in the audit trail (the model's own sod_commentary pass opens with a FRAME-CONTROL NOTICE when the frame is misapplied), and in the editorial banner at the head of the rendered text page. The methodological purpose of Run C is articulated in §6.2.

4.3 Held variables

Identical model (claude-opus-4-7), identical schema, identical morphology pipeline, identical hybrid retriever (BM25 + multilingual e5-base via reciprocal rank fusion). All three runs single-attempt to schema-valid output (no regeneration retries fired). The runs are reproducible from the spec files; all audit packages are preserved at benchmarks/plotinus-vi-9-11-{...}/run-20260509T10*Z/ in the project repository.

4.4 Cost and ethics

Total Anthropic API spend across the three runs was approximately 0.50 USD. The experiment was self-funded; no external grant or institutional funding supported the work. Run C's deliberate anachronism is a methodological control variable, not a claim of historical reception, and is flagged as such in every artifact the run produces. No fabricated primary text was authored; the Greek source follows standard critical-edition convention with editor verification recommended in the artifact's experimental.source_verification field.

5. Results

5.1 Run summaries

Table 2 summarizes the three runs along the schema-output dimensions that proved differentially responsive to frame.

The faithful pass of each run, byte-identical to the engine output:

5.2 Finding 1 -- capitalization tracks frame disposition

The closing pair (μόνου ... μόνον, genitive and accusative of the identical lexeme) produces three different orthographic decisions across the runs. Each run's apparatus.decision_notes field records the rationale explicitly. Code Block 1 reproduces the relevant decision-note JSON from each run's output.json:

{
  "anchor": "monou pros monon",
  "decision": "Render symmetrically as 'the alone to the alone' rather than disambiguating to 'soul to One'. The morphological symmetry is doctrinally operative; preserving it in English keeps the contemplative coincidence visible.",
  "by": "editor-targum"
}


{
  "anchor": "alone-to-alone-capitalization",
  "decision": "Rendered 'alone to the Alone' with capital A on the second monon to mark the aspectual shift the Akbarian reception reads into the line, while preserving the Greek surface's identical lexeme. This is a minority editorial choice in English Plotinus translation (Armstrong uses lowercase throughout); flagged for editor review.",
  "by": "editor-targum"
}


{
  "anchor": "monon-capitalization",
  "decision": "Capitalized 'Alone' in the body to mark the second monon as functioning as a divine name analogue (the One as Alone), parallel to the editorial convention of capitalizing 'the Infinite' for Ein Sof. Preserves the apophatic register without importing Kabbalistic vocabulary.",
  "by": "editor-targum-engine"
}

Three different rationales for the orthographic choice, all citable. The Hekhal editorial position would recognize Run A's lowercase as the philologically preferable rendering of Plotinian source per the henosis-and-theosis discipline (do not import the reception's frame into the body); Runs B and C are coherent within their respective frames but represent reception readings rather than philologically primary readings. The frame produces the orthographic discipline. The same Greek source becomes incommensurable with itself at the level of capitalization once the frame is named -- and the audit trail makes the incommensurability traceable to specific frame relations rather than to opaque editorial taste.

5.3 Finding 2 -- apparatus density tracks frame articulation

Range card population varies systematically across runs: 2 / 6 / 0. The differential is not arbitrary. The kataphatic-apophatic frame operates around a small set of load-bearing technical terms (apophasis, hyperousios, henosis, gnophos, plus the hyper-compound system), and the closing of VI.9.11 directly invokes only one (henosis as the contemplative terminus, performed but not lexically named). The engine populates a card for henosis and one for the monou pros monon formula itself, with explicit rationales. No fabrication.

The zahir-batin frame operates around a richer term-pair lattice (zahir/batin, fana/baqa, ahadiyya/wahidiyya, tajalli, hijab) where each pair generates two range cards. The frame's hermeneutic rule is to surface both poles even where only one is lexically present in the source. The engine populates cards for fana (implied via apallage / phyge), baqa (implied via monou pros monon), ahadiyya, wahidiyya, tajalli, and hijab; the last two are explicitly tagged "frame-internal, not in surface" in the rationale field, with senses surfacing only in the sod_commentary pass. Code Block 2 reproduces the tajalli range card from Run B as illustration:

{
  "term": "tajalli (frame-internal, not in surface)",
  "selected_sense": "self-disclosure",
  "active_senses": ["manifestation", "epiphany", "self-disclosure"],
  "rationale": "Surfaced only in the sod_commentary, not in the primary translation. The Akbarian reception reads the 'flight of the alone to the Alone' as the very structure of self-disclosure: the disclosing remains hidden in the disclosure. Range card populated because the term is operative in the batin reading even though absent from the Greek surface.",
  "lexicon_ref": "lexicon/tajalli"
}

The PaRDeS frame, deliberately misapplied to a non-Hebrew non-scriptural source, produces zero range cards. The engine refused to fabricate Kabbalistic content where no Hebrew lexical material exists in the Greek source. This is the frame's articulation made measurable: apparatus density is itself a frame-tracking metric, and the engine's behavior under deliberately misapplied scaffolding is to produce no apparatus content rather than fabricated apparatus content.

5.4 Finding 3 -- register-decorum awareness in Run C

Run C is the experiment's load-bearing methodological control. The hypothesis going in: a sufficiently scaffolded engine should recognize the register-decorum violation produced by deliberately misapplying PaRDeS to Plotinus, and respond by suppressing the misfired controller in the body while still surfacing the structural cross-tradition resonance the source happens to carry. The actual response across four observed dimensions:

(i) Body translation produced by Greek philological decorum, not by PaRDeS. The body is essentially the same as Run A in syntactic register, with Run B's capital-A "Alone" choice adopted (with the Run-C-specific rationale documented above). No Hebrew technical vocabulary appears in the body. No sefirot-system reading is imposed. The four PaRDeS senses (peshat, remez, derash, sod) are not performed on the Greek surface.

(ii) Misapplication explicitly flagged in the sod_commentary pass. The pass opens with the literal text reproduced in Code Block 3:

"FRAME-CONTROL NOTICE: The PaRDeS controller has been deliberately applied to a third-century Greek Platonist text as an anachronism control. PaRDeS is a Kabbalistic hermeneutic for reading Hebrew scripture; Plotinus is neither scripture nor Hebrew. No peshat/remez/derash/sod registers properly fire here. What can be said cross-traditionally and only in apparatus: Plotinus's monou pros monon (the flight of the alone to the Alone) is the structural-apophatic analogue of the ein-sof / ayin lineage in Kabbalah ... But Plotinus's One is not Ein Sof: there is no sefirotic articulation, no shevirat ha-kelim, no Hebrew root-play, no scriptural substrate to perform sod against. The body translation here is governed by Greek philological-philosophical decorum; the Kabbalistic frame is suppressed except for this notice and the cross-references in apparatus. To render this passage as Kabbalistic sod would be exactly the methodological incompetence the PaRDeS controller forbids: importing one register's content into another register's text."

(iii) Cross-tradition resonance surfaced in apparatus, not body. The apparatus cross-references include lexicon/ein-sof, lexicon/ayin, and lexicon/devekut as structural-apophatic homologies, with each cross-reference's relation field stating "structural homology, not historical identity." The asymmetry between Plotinian henosis (which tends toward identity-language at its strongest formulation) and Hebrew devekut (which preserves the soul-divine distinction even at the contemplative apex) is named explicitly in the apparatus footnote on monou pros monon: "the parallels arise independently in traditions that take the divine prior to predication seriously, not historical dependencies (though the Plotinus → Pseudo-Dionysius → medieval Christian apophatic line is direct, and the Akbarian and Kabbalistic developments are mediated through the Arabic-Neoplatonist heritage)."

(iv) One productive drift incident surfaced. The engine cited lexicon/ayin as a cross-reference; ayin did not exist in the Hekhal lexicon at run time. This is a fabricated-lexicon-ref incident in the audit's terms. Code Block 4 reproduces the drift report:

{
  "chunk_id": "plotinus.enneads.vi-9-11.closing.pardes-anachronism",
  "clean": false,
  "incidents": [
    {
      "type": "fabricated-lexicon-ref",
      "message": "apparatus.cross_references[1] references 'lexicon/ayin' which does not exist on the Hekhal side",
      "location": "apparatus.cross_references[1]",
      "proposed_fix": "either build the lexicon entry at src/content/lexicon/ayin.mdx, or change the ref to point at an existing entry, or drop the ref"
    }
  ]
}

The drift is productive. The cross-tradition apophatic-priority triangle (ein-sof / ayin / hyperousios / ahadiyya) is genuinely worth a Hekhal lexicon entry, and the engine surfaced the gap by trying to cite something that should have existed. The drift is closed in the experiment's publication artifact by adding the entry (now live at lexicon/ayin).

The Run C output is not an error. It is the engine's mature response to a frame outside its proper domain: rendering by philological decorum, surfacing cross-tradition resonance only as homology in apparatus, explicitly flagging the methodological move in the audit trail, and surfacing genuine lexical gaps where present. Every translator has a frame. Targum's frame is citable in a way human translators' historically have not been -- and the deliberate-anachronism stress test demonstrates that the engine respects register-decorum rather than producing fabricated cross-tradition output under arbitrary frame application.

5.5 Finding 4 -- cross-run experimental design caught hallucinated MacKenna citation

The engine's comparison_to_pd_translations field cited MacKenna 1930 in all three runs. The cited renderings of the closing phrase μόνου πρὸς μόνον differ across runs:

• Run A's MacKenna citation: "a flight of the alone to the Alone"
• Run B's MacKenna citation: "the passing of solitary to solitary"
• Run C's MacKenna citation: "the passing of solitary to solitary"

The divergence is not stylistic variation. We verified the canonical MacKenna 1930 text against the Sacred-Texts archive of the public-domain MacKenna edition, accessed via the Internet Archive Wayback Machine snapshot dated 2024-05-25 (web.archive.org/web/20240525171551/sacred-texts.com/cla/plotenn/enn714.htm, page identified as Sixth Ennead, Ninth Tractate, Section 11). The canonical reading:

"This is the life of gods and of the godlike and blessed among men, liberation from the alien that besets us here, a life taking no pleasure in the things of earth, the passing of solitary to solitary."

Runs B and C cite MacKenna correctly. Run A fabricated. The form "flight of the alone to the Alone" is the most-quoted English form of the closing of VI.9.11 in popular and scholarly reception, but it is not MacKenna's rendering; the form belongs to Armstrong's 1988 Loeb (Armstrong 1988) and to Hadot's French "fuite du seul vers le Seul". The model defaulted in Run A to the famous-by-paraphrase form rather than to MacKenna's actual text. The methodological observation: cross-run experimental comparison of canonical-translator citations surfaces hallucination by divergence even when single-run verification is unavailable. Without the experiment's three-run structure, Run A's hallucinated MacKenna would have shipped as if real. The editorial-law requirement that every translation has provenance, every claim is footnotable is what made the divergence catchable; the divergence itself proves the requirement is load-bearing rather than ornamental.

5.6 Per-Greek-term divergence

Table 3 maps eight key Greek tokens from the source across the verified MacKenna 1930 rendering and the three Targum runs. Where the columns diverge, the divergence is doctrinally rather than incidentally significant. The clearest sites of frame-disposition are the rendering of φυγή (where Run A uses the definite article, marking the contemplative-terminus typification, and Runs B and C use the indefinite article in Plotinian descriptive register) and the orthographic decision on the closing μόνον.

Three observations follow from Table 3. First, all three Targum runs are more philologically conservative than MacKenna at multiple sites: Targum chose release over MacKenna's interpretive liberation; Targum read τῶν ἄλλων strictly as "the other things" rather than expanding to MacKenna's "the alien that besets us"; Targum preserved flight against MacKenna's softening to passing. Second, MacKenna's symmetric lowercase rendering (solitary to solitary) matches Run A's discipline of preserving Plotinian morphological symmetry against the Christian-reception capitalization convention. Third, the differences between runs are concentrated at exactly two sites -- the article on φυγή and the capitalization on μόνον -- both of which the engine's audit trail ties to specific frame relations.

6. Discussion

6.1 Frame as citable parameter

The four findings together support a methodological position that has not, to our knowledge, been articulated in the published literature on machine translation of religious or esoteric primary text: hermeneutic frame is a tunable parameter of computational translation, not a hidden default. The position is not novel as philosophy -- Corbin (1969), Idel (2002), Sells (1994), and the broader reception studies tradition (Hardwick and Stray 2008) have argued versions of it for decades. The position is novel as engineering practice: prior LLM translation systems have not exposed the hermeneutic frame as a first-class, audited, swappable parameter, and the implicit-frame default that human translation has historically operated under has been imported uninspected into computational practice. The Targum architecture and the present experiment together demonstrate that the implicit-frame default is not necessary, and that the explicit-frame alternative produces translations whose differences are traceable in audit trail to specific named relations rather than to opaque editorial taste.

6.2 Frame-disclosure ethics

The deliberate misapplication of the third frame -- PaRDeS on Plotinus -- requires explicit methodological framing. Run C is not comparative religion. It is not a claim that Plotinus is best read through Kabbalistic categories, or that the apophatic-priority triangle (Ein Sof / hyperousios / ahadiyya) is operative in the Plotinian source. It is a stress test of the engine's response to a frame outside its proper domain. The result of the test is that the engine recognizes the register-decorum violation and produces output that explicitly flags its own historical incoherence -- the body rendered by Greek philological decorum, the cross-tradition resonance acknowledged only as structural homology in apparatus, the misapplication called out by name in the sod_commentary pass.

The published artifact that results is a contribution to the ethics of translation, not to comparative religion. Every translator has a frame. Historical comparative-religion writing has been bad at flagging implicit frames; the discipline has improved with the rise of explicit reception studies, but the move from implicit to explicit frame-disclosure remains incomplete in published English translation of esoteric primary text. Targum's audit-trail-flagged frame is what makes frame-choice citable in a way human translators' historically have not been. The deliberate-anachronism control demonstrates this empirically: the engine produced a reading and explicitly flagged its own historical incoherence in the apparatus, in a way that no human translator imposing PaRDeS on Plotinus would automatically do. The flag is the contribution. The methodology is the discipline. The frame is the citable parameter.

6.3 Apparatus density as a frame-tracking metric

The systematic variation in range card population (2 / 6 / 0 across the three runs) is generalizable beyond this experiment. Apparatus density is a frame-tracking metric available at the engine's audit trail without additional instrumentation. Frames with denser term-pair architectures (zahir-batin, with its five named relations and term-pair lattice) proliferate cross-references on a given source. Frames with leaner technical-term sets (kataphatic-apophatic, organized around the hyper-compound system and a small set of apophatic terms) stay restrained. Frames misapplied to non-cognate source material produce no apparatus rather than fabricated apparatus. The metric is queryable per chunk and stable across runs of the same source through the same frame; it could be developed into a quantitative measure of frame-source compatibility that complements existing translator-agreement metrics.

6.4 Cross-run experimental design as hallucination check

The hallucinated MacKenna citation in Run A was caught by cross-run divergence rather than by single-run verification. This generalizes. For any LLM translation system that produces canonical-translator citations as part of its output, cross-run comparison of those citations across varied prompt conditions is a low-cost check on citation reliability. The check is particularly valuable for canonical public-domain translators of widely-paraphrased classical texts (MacKenna of Plotinus, Mead of the Hermetica, Westcott of Sefer Yetzirah, Parker of Pseudo-Dionysius), where the most-famous English form of a passage may not be the canonical translator's actual rendering but a later translator's rephrasing. We propose that future Targum versions add a canonical-translator-citation drift type that flags engine-generated comparison_to_pd_translations entries as draft pending editor verification by default for canonical PD translators of widely-paraphrased classical texts. This addition would catch the failure mode at audit time rather than at cross-run-comparison time.

6.5 Position of the experiment in the methodology paper

This experiment is the methodology paper's qualitative single-passage chapter. The argument articulated here -- frame as citable parameter, frame-disclosure ethics, apparatus density as frame-tracking metric, cross-run hallucination detection -- is qualitative across one rich passage. Quantitative replication across the spec-§11-named held-out evaluation set (~500 passages where modern critical translations agree) is the methodology paper's load-bearing experimental apparatus and remains unbuilt as of this writing. The qualitative argument is necessary but not sufficient; the quantitative replication is queued.

7. Limitations and future work

Five limitations of the present experiment are worth naming explicitly.

Single-passage qualitative argument. The four findings are demonstrated on one twenty-one-word Greek passage. The argument's generalization to other passages, other source languages, other frame controllers, and other model versions remains to be tested. The methodology paper's eval set replication will address this.

Manually authored frame controllers. The three frame controllers tested (kataphatic-apophatic, zahir-batin, pardes) were authored by the project editor against the standard scholarly literature in each tradition. Auto-induction of frame controllers from existing scholarly editorial summaries is an open question; it is not addressed here.

No embedding-space metric layer. The findings reported are categorical (capitalization choice, range-card count, presence/absence of FRAME-CONTROL-NOTICE in sod_commentary) rather than continuous-distance. An embedding-space layer that measures semantic distance between scaffolded outputs and unscaffolded baseline, or between Targum outputs and historical translator outputs, would produce continuous-valued frame-tracking metrics complementary to the categorical ones reported here. This is queued.

No output validator yet for the orthographic discipline. Finding 1 documents that Run A's lowercase symmetric rendering of monou pros monon is the philologically preferable choice for Plotinian source. The kataphatic-apophatic frame controller currently does not enforce this; capitalization fires as a free editorial choice. A controller-level output validator preserving the symmetric lowercase rendering by default, with capitalization requiring an explicit decision-note, would close the gap.

Canonical-translator-citation drift type unimplemented. Finding 4 surfaces a failure mode (the model defaulting to famous-by-paraphrase forms over the canonical translator's actual text) that is not currently caught by the engine's audit. The proposed canonical-translator-citation drift type addresses this and is queued.

Future work, in approximately the order it should be undertaken: held-out evaluation set construction (the load-bearing experimental apparatus the methodology paper depends on); the canonical-translator-citation drift type and the kataphatic-apophatic orthographic-discipline output validator (engine improvements that close the gaps surfaced here); cross-tradition validation experiments such as the apophatic-priority triangle experiment that operationalizes Hekhal's strongest cross-tradition lexicon link; further frame controllers (tafsir-ta'wil for Ismaili, Quadriga for Christian-esoteric exegesis, peshat-derash for Heikhalot transmission); and ultimately the methodology paper itself.

8. Reproducibility

All artifacts referenced in this paper are preserved at named paths in the project repository. The benchmark specification files, the assembled prompt snapshots, the engine output JSONs, the drift reports, and the comparison.md scaffolds are at:

benchmarks/specs/plotinus-vi-9-11-{christian-reception,akbarian-reception,pardes-anachronism}.yaml
benchmarks/plotinus-vi-9-11-{christian-reception,akbarian-reception,pardes-anachronism}/run-20260509T10*Z/
benchmarks/frame-conditioning-vi-9-11/analysis.md

The frame controllers tested are at backend/frames/{kataphatic_apophatic,zahir_batin,pardes}.yaml. The glossaries loaded are at glossaries/{christian-apophatic,akbarian-sufism,kabbalah}/general.yaml (with an additional Ibn-Arabi-specific glossary at glossaries/akbarian-sufism/ibn-arabi.yaml). The scholarly corpus retrieved is at corpus/scholarly/{christian-apophatic,akbarian-sufism,kabbalah}/; the christian-apophatic collection (six editorial summaries on apophasis, hyperousios, henosis-and-theosis, gnophos, kataphatic-apophatic-pair, plotinus-pseudo-dionysius-reception) was seeded specifically for this experiment.

Reproduction commands (PowerShell on Windows; equivalent on POSIX shells):

$env:ANTHROPIC_API_KEY = "sk-ant-..."
cd L:/Creative/hekhal-targum
python -m backend.cli benchmark benchmarks/specs/plotinus-vi-9-11-christian-reception.yaml --client anthropic
python -m backend.cli benchmark benchmarks/specs/plotinus-vi-9-11-akbarian-reception.yaml --client anthropic
python -m backend.cli benchmark benchmarks/specs/plotinus-vi-9-11-pardes-anachronism.yaml --client anthropic

Engine specification: docs/TARGUM-ENGINE.md. Model: claude-opus-4-7 via Anthropic API in tool-use mode with a JSON Schema enforcing the output shape. Glossary revision: as named in each run's audit_trail.lexicon_revision field. Frame controllers version: 1.0. Retriever: hybrid BM25 + dense (multilingual-e5-base, 768-dim) via reciprocal rank fusion at k=60. Date of execution: 9 May 2026 (UTC timestamps in audit packages). Total Anthropic API spend: approximately 0.50 USD.

The three rendered Targum text pages this experiment publishes are at Run A, Run B, and Run C. The lexicon entry seeded to close Run C's productive drift incident is at lexicon/ayin. The full per-Greek-term analysis with verified MacKenna alignment is at benchmarks/frame-conditioning-vi-9-11/analysis.md in the project repository.

References

1. Adamson, P. (2002). The Arabic Plotinus: A Philosophical Study of the Theology of Aristotle. London: Duckworth.
2. Adamson, P., and R. C. Taylor, eds. (2005). The Cambridge Companion to Arabic Philosophy. Cambridge: Cambridge University Press.
3. Armstrong, A. H. (1988). Plotinus, with an English Translation, Volume VII: Enneads VI.6-9. Loeb Classical Library 468. Cambridge, MA: Harvard University Press.
4. Bradshaw, D. (2004). Aristotle East and West: Metaphysics and the Division of Christendom. Cambridge: Cambridge University Press.
5. Carabine, D. (1995). The Unknown God: Negative Theology in the Platonic Tradition, Plato to Eriugena. Louvain: Peeters.
6. Catana, L. (2013). "Late Ancient Platonism." In The Routledge Handbook of Neoplatonism, edited by P. Remes and S. Slaveva-Griffin. London: Routledge.
7. Chittick, W. C. (1989). The Sufi Path of Knowledge: Ibn al-'Arabi's Metaphysics of Imagination. Albany: SUNY Press.
8. Chodkiewicz, M. (1993). An Ocean Without Shore: Ibn Arabi, the Book, and the Law. Translated by D. Streight. Albany: SUNY Press.
9. Corbin, H. (1969). Creative Imagination in the Sufism of Ibn 'Arabi. Translated by R. Manheim. Princeton: Princeton University Press. [Original French: L'Imagination créatrice dans le soufisme d'Ibn 'Arabi, Flammarion 1958.]
10. Hadot, P. (1993). Plotinus, or The Simplicity of Vision. Translated by M. Chase. Chicago: University of Chicago Press. [Original French: Plotin ou la simplicité du regard, 1963.]
11. Hardwick, L. (2003). Reception Studies. New Surveys in the Classics 33. Oxford: Oxford University Press for the Classical Association.
12. Hardwick, L., and C. Stray, eds. (2008). A Companion to Classical Receptions. Oxford: Wiley-Blackwell.
13. Idel, M. (2002). Absorbing Perfections: Kabbalah and Interpretation. New Haven: Yale University Press.
14. Louth, A. (2007). The Origins of the Christian Mystical Tradition: From Plato to Denys. 2nd edition. Oxford: Oxford University Press.
15. MacKenna, S. (1917-1930). Plotinus: The Enneads. 5 volumes. London: Medici Society. Public-domain digital text accessed via Sacred-Texts (sacred-texts.com/cla/plotenn/) and Internet Archive Wayback Machine.
16. Perl, E. D. (2007). Theophany: The Neoplatonic Philosophy of Dionysius the Areopagite. Albany: SUNY Press.
17. Rist, J. M. (1967). Plotinus: The Road to Reality. Cambridge: Cambridge University Press.
18. Rorem, P. (1993). Pseudo-Dionysius: A Commentary on the Texts and an Introduction to Their Influence. Oxford: Oxford University Press.
19. Russell, N. (2004). The Doctrine of Deification in the Greek Patristic Tradition. Oxford Early Christian Studies. Oxford: Oxford University Press.
20. Schaefer, C. (2006). The Philosophy of Dionysius the Areopagite: An Introduction to the Structure and the Content of the Treatise On the Divine Names. Philosophia Antiqua 99. Leiden: Brill.
21. Schimmel, A. (1975). Mystical Dimensions of Islam. Chapel Hill: University of North Carolina Press.
22. Scholem, G. (1990). Origins of the Kabbalah. Translated by A. Arkush. Edited by R. J. Z. Werblowsky. Princeton: Princeton University Press / Jewish Publication Society. [Original German: Ursprung und Anfänge der Kabbala, 1962.]
23. Sells, M. A. (1994). Mystical Languages of Unsaying. Chicago: University of Chicago Press.
24. Sells, M. A. (1996). Early Islamic Mysticism: Sufi, Qur'an, Mi'raj, Poetic and Theological Writings. New York: Paulist Press.
25. Toury, G. (1995). Descriptive Translation Studies and Beyond. Amsterdam: John Benjamins.
26. Turner, D. (1995). The Darkness of God: Negativity in Christian Mysticism. Cambridge: Cambridge University Press.
27. van der Heide, A. (1983). "PARDES: Methodological Reflections on the Theory of the Four Senses." Journal of Jewish Studies 34 (2): 147-159.
28. Venuti, L. (1995). The Translator's Invisibility: A History of Translation. London: Routledge.
29. Wear, S. K., and J. Dillon. (2007). Dionysius the Areopagite and the Neoplatonist Tradition: Despoiling the Hellenes. Aldershot: Ashgate.

Citation

Hekhal Editorial. (2026). "Frame as Citable Parameter: A Three-Reception Experiment on Plotinus Enneads VI.9.11 with the Targum Translation Engine." Hekhal Targum Phase-2 Methodology Paper, v1.0, 9 May 2026. hekhal.org/targum-experiments/frame-conditioning.