Not What She Said, but What They Used It For

Abstract. In denominations built on prophetic authority, quoting the founder’s writings is not mere transmission of content but an institutional act: it marks which claims are held to require external warrant, and what kind of warrant. This article measures that practice across a century of three official Seventh-day Adventist periodicals (Review and Herald, Ministry, Curierul Adventist; 7,133 issues, 177 million words, 78,091 marked Ellen G. White citations). The unit of analysis is not the article but the ±50-word neighbourhood of each detected citation, coded by a trilingual thematic lexicon and benchmarked against randomly sampled text positions drawn from the same channel and period, so that the resulting association measure is independent of a channel’s overall topical saturation. Between 83.8% and 89.2% of citations occur near none of the doctrinal themes examined: authority-citation as doctrinal reinforcement is a minority, targeted practice. Within that minority, point estimates for the two American channels rise across periods (1.25 → 1.72 → 2.02; 1.19 → 1.42 → 1.79) while the Romanian channel moves the other way (1.68 → 1.06). Year-by-year citation rates supply essential context: citation volume in the American channels roughly halves after 1980 while binding strengthens — authority-citation becomes a rarer yet more specialised practice — and in the 2010s the two measures diverge in the opposite direction: volume partly rebuilds while doctrinal binding recedes toward chance. In the Romanian channel the collapse is not gradual but a datable step (1947: 5.3 → 1949: 0.05 citations per 10,000 words), with recovery lagging the 1989 transition by four years. A dedicated pre-screen of the censored volumes shows, however, that what collapsed was naming, not the practice: between 1950 and 1980 name-substituting invocations of her authority (“the spirit of prophecy”, “the Lord’s servant”) are the paper’s dominant form of reference — 263 such places against 25 named citations. The two figures count different units (places of invocation versus source-identified citations), so their ratio is not a citation ratio. Sanctuary doctrine travels on that channel too, and on occasion — chiefly in anniversary issues — the unnamed authority is drawn into the regime’s rhetoric of loyalty. Named citation is restored only after 1989. Source selection is systematically patterned: visionary-chronological works cluster around historicist-prophetic themes, whereas institutional self-identification draws disproportionately on the church-disciplinary Testimonies — administrative rather than apocalyptic warrant.

Keywords: content analysis, corpus-assisted discourse studies, religious press, authority, citation practice, Seventh-day Adventism, Ellen G. White, censorship, Romania

1. Introduction

1.1 Citation as an Object of Study

In a denomination built on a founder with prophetic standing, quoting the founder’s writings is never a neutral operation. Whoever quotes does two things at once: carries a content across, and transfers authority onto it. The placement of a citation is therefore diagnostic. It shows which claims the speaker holds to be insufficient on their own — and what kind of confirmation is deemed fit to support them.

This study is not about Ellen G. White, but about those who quoted her. Its subject is citation as church practice: where the official denominational periodicals place White references, next to which themes, from which of her works, and how all of this changes across a century and three political-publicity environments.

The question also has a sharper formulation, and the study ultimately answers it. The three organs examined are not a random sample of Adventist speech: all three are official, authoritative periodicals representing the normative register of denominational publicity, and — in the Romanian case, for decades — appeared under state supervision. What these periodicals do with their citations is therefore not simply practice but suggestion: it signals to the reader what kind of authority White is. Prophetic proof for a disputed doctrine? A pastoral voice? Organisational admonition? The aim of the measurement is to read out this suggestion not from impression but from countable linguistic co-occurrence.

1.2 Prior Research

The literature on White’s writings organises itself around two questions, and both are well developed.

The first is theological: what did White teach, and how does this fit into the denomination’s system of belief. The history of this literature is also the developmental history of the belief system itself, whose standard summary is Knight’s A Search for Identity.¹ Sanctuary doctrine as the integrating element of the system, as well as its internal interpretive currents, have received dedicated monographs.²

The second is historical-critical: how original White’s texts are, how they came about, to what extent they depend on contemporary sources. The turning point here was Ronald L. Numbers’s 1976 book, which foregrounded the compositional history of the texts, and to which the denominational side responded with comprehensive apologetic surveys.³ The debates of the half century since — from the rediscovery of the 1919 Bible Conference to the reception history of Questions on Doctrine — largely move along this axis.⁴

There is, however, a third question, less developed than either: what did those who quoted the texts use them for. This is not about the author, and not about the authenticity of the texts, but about reception. Knight is the one who treats this most clearly as a subject in its own right: he distinguishes the historical variant forms of reading White, and describes the institutionalisation of her posthumous “afterlife” — the editing of compilation volumes, the canonisation of the corpus, the use of the citation as proof — as an independent problem.⁵ The 2014 volume Ellen Harmon White: American Prophet performs the same turn institutionally: it treats White with a scholarly apparatus external to the church, within an American religious-historical context.⁶

The current state of this line of inquiry is most precisely captured by the 2025 essay collection Reclaiming the Prophet, which brings together the field’s leading historians. In his introduction, Eric Anderson formulates exactly where research stands: “the participants in the conference accepted that the critical scholarship of the last fifty years had established how Ellen White should not be used.” The volume’s aim is to move on from the “litany of negatives” — that White was “neither unique nor infallible”, that she “was not the final authority for all Biblical interpretation” — toward what can be affirmed, in the spirit of the old maxim abusus non tollit usum.⁷

This is where the gap becomes visible. The literature has worked out the normative question: how White should and should not be used. The descriptive question — how she was actually used, across the full run of the periodicals, over decades — has remained open. This is not a reproach: the literature’s method is case-study based, analysing prominent controversies, emblematic authors and turning-point documents, and generalising from these. But this leaves it unknown whether the citation pattern observed in the prominent controversies is characteristic of the periodical run as a whole, or exceptional.

The sharpness of the gap is well illustrated by the fact that the descriptive claim appears in the literature as perception. Niels-Erik Andreasen puts it this way in the volume’s afterword: “a popular perception exists in parts of our church that some of its spokespersons over the years have referenced E. G. White’s writings as a convenient substitute for Scripture, a kind of third testament” — and that in disputes over interpretation believers have turned “to Ellen White as the final doctrinal arbiter”, although “she would not have supported that role for herself.”⁸ White herself upheld the same norm: during the 1888 controversy she explicitly declined to have her writings used to settle a doctrinal question.⁹

A “widespread belief”, however, is exactly the kind of claim that can be measured. This is what the study does: it neither presupposes nor refutes in advance the “proof-text” model — that the White citation stands primarily where a disputed doctrine needs support — but turns it into a testable claim, and measures it across the full periodical run.

The same gap looks different on the methodological side. Content analysis of the religious press is a well-established genre, but the coding unit is almost without exception the article or the issue: it measures what a text is about. In a denominational periodical, however, document-level co-occurrence is close to tautological — nearly every issue both quotes White and discusses sanctuary doctrine somewhere. The rhetorical claim, by contrast, is local: it says that the citation stands at a doctrinally load-bearing point. A local claim can only be verified with locally measured data.

1.3 Contribution of the Study

The investigation goes beyond prior work on four points.

(1) The descriptive question alongside the normative. The subject is not the author and not textual authenticity, but usage. The literature has worked out how White should and should not be used; this study measures how she was actually used. This turns a long-voiced but never tested belief — that citation functions primarily as a doctrinal arbiter — into a testable claim.

(2) The unit of measurement. The coding unit is not the article but the ±50-word linguistic neighbourhood of each individual citation. This allows the study to measure citation practice, not topic frequency — to my knowledge this resolution has not previously appeared in quantitative analysis of the religious press.

(3) An empirical, within-channel reference. A bare proportion is uninterpretable on its own: if a periodical is saturated with sanctuary doctrine, its citations will automatically stand next to it too, without “attracting” anything. The study therefore compares every proportion against randomly sampled text positions from the same periodical and period. The measure thereby becomes independent of the channel’s topical saturation. The procedure is familiar in corpus linguistics, rare in content analysis of the religious press; combining the two is the study’s methodological proposal.

(4) A comparative design. The same measurement instrument runs across two free-press American channels and one Romanian channel that operated under censorship for decades. This makes it possible to examine whether political constraint does not merely reduce citation but also reorganises what it is called upon to support.

One preliminary conceptual clarification. The study measures thematic co-occurrence, not rhetorical function. From the fact that a citation stands in a sanctuary-doctrine linguistic neighbourhood, it does not follow that it proves, illustrates or disputes. The study therefore speaks throughout of “thematic binding”, and leaves functional claims to future manual coding. This is not an excuse but a limit that follows from the measurement design, and stating it precisely marks out the scope of validity of the conclusions.

2. Theoretical Framework

Citation as a transfer of authority is not a new object either in literary theory or in the sociology of religion; the present study draws on these traditions and renders them measurable.

The citation as an operation. In the literary-theoretical tradition, citation is not transmission but intervention: it lifts something out of the source, places it in a new environment, and binds the receiving text to the source’s authority. Compagnon’s classic analysis describes citation explicitly as an operation — the double gesture of cutting and pasting — while Genette provides a typology of the relations texts bear to one another.¹⁰ From this tradition the present study isolates and renders measurable a single dimension: the location of the operation.

Marked and unmarked borrowing. In research on biblical intertextuality, Hays distinguishes citation, allusion and echo according to how explicit the borrowing is.¹¹ This distinction supplies the present study’s most important scope limitation: the measurement sees only marked citation — that in which a name, a title, or some other recognisable anchor is present. Unmarked borrowing and anchorless paraphrase remain, in principle, invisible. This limitation has a direct consequence for the interpretation of one result (5.7).

Routinised authority. In sociological terms, the object of the study is what is called the routinisation of charismatic authority: the transformation of a personal prophetic claim, after death, into an authority that is institutionally managed, bound to a text and dispensed through official channels.¹² Within this framework, citation practice is not incidental: it is precisely the surface on which routinised authority becomes visible and measurable. The question — “what kind of authority does the official periodical suggest?” — is thus not a matter of journalism but an empirically tractable question of routinisation theory.

The pastoral axis: from fear to assurance. Among the eleven themes, the four rhetorical emphases — imminence, apologetics, judgment-fear, assurance — call for separate justification, because they measure not doctrinal content but mode of address. This axis maps one of the best-documented strands of Adventist internal developmental history. The movement’s preaching in its beginnings was built on the imminence of judgment and the urgency of preparation for it; since the 1888 Minneapolis General Conference session, regarded as a turning point, with its “righteousness by faith” message, the denomination’s historiography has traced a long, contested shift from performance- and fear-based religiosity toward centring evangelical assurance.¹³

The doctrine of the investigative judgment is the focal point of this tension, and specifically as a pastoral question: the same doctrine can be read as unsettling (“every sin is recorded”, “your case comes up in turn”) and as assurance-giving (“we have an advocate with the Father”, “the judgment speaks in favour of the saints”). The two readings organised themselves into schools in the twentieth century: one pole’s codification is M. L. Andreasen’s “last generation” theology, with its requirement of character perfection; the other’s is the direction that derived the believer’s assurance from the high-priestly intercession — the Questions on Doctrine and the Ford controversy (5.6) are, in part, a clash of these two poles.¹⁴

The instrument therefore treats fear and assurance as two separate poles, not as the two ends of a single scale: a text can score high on both at once — this is the classical “fear, then comfort” preaching arc — and a combined index would erase this. In the present study this axis carries the question of whether White’s authority was invoked more in the older, urgency-fear register or in the newer, assurance register. The instrument was originally built, within the author’s broader research, to interpret the denomination’s growth dynamics — its baptismal waves — that is, the question of what rhetorical repertoire the church worked with, and how this changed; for the purposes of the present analysis, the significance of this origin is that the emphasis themes were not built with knowledge of the White-citation result (4.3.2). The content background of the themes and the linguistic operation of the coding are presented theme by theme in Appendix G.

3. Data

3.1 The Three Channels

The choice is not one of convenience. The three channels differ along two axes. The two American periodicals share the same political environment but address different audiences: Review and Herald the membership, Ministry the clergy’s professional publication. This pair makes the effect of audience visible. Curierul Adventist is the official organ of the same world church, but in a different language and, for decades, under radically different conditions of publicity — this pair makes the effect of political constraint visible.

All three periodicals are official publications. This is both a limit and part of the study’s subject: the results pertain to the normative register of denominational publicity, not to Adventist speech as a whole.

3.2 Source and Preparation

The English-language material comes from the denomination’s public digital periodical archive (documents.adventistarchives.org); the Romanian material from the public online archive of Curierul Adventist (curieruladventist.ro); both offer issue-level PDFs. Text extraction was performed with pdftotext, and, on failure, pdfplumber.

Every document carries an OCR health score (0–1), based on the dictionary-match rate of the recognised text; per-cell values are reported in Table A1 (0.90–0.95). The texts were normalised before analysis: the normalisation standardises diacritics, handles Romanian orthographic reforms (1953, 1993: sînt / sunt / sânt), obsolete Hungarian orthography, and two characteristic OCR errors — end-of-line hyphenation and, measured separately in the Romanian material, the systematic loss of word-final ă.

The two-column layout of the pre-war issues is often merged by OCR into a single run of text; a dedicated pre-processing step separates the columns before any window is cut. Omitting this step would systematically create false neighbourhoods.

The post-2005 volumes of the American material were added to the corpus in a supplementary collection round: 465 Adventist Review issues (2005–2020) and 180 Ministry issues (2006–2020). The Ministry series is complete — four misdated filenames in the archive had to be corrected to the actual month of publication, and the 2010 and 2015 volumes appeared with a combined July–August issue (11 issues per year constitutes a full volume); the Review series is missing a single issue (2014/30). The overlapping 2005 volume was deduplicated by issue identifier. The post-2005 PDFs carry a born-digital text layer; the effect of the extractor switch is ruled out by a 30-document sample, stratified across periods, on which the text from the two extractors and the detector’s accepted hit count are exactly identical.

3.3 Coverage

Table A1. The Nine Analysed Cells

The nine cells together comprise 163.0 million words and 74,520 accepted citations. The full processed corpus is larger still — 177.5 million words, 78,091 citations — because it also contains material outside the period boundaries (Review and Herald 1912–1919: 388 issues, 13.2 M words, 3,162 citations; Curierul Adventist 2021–2026: 69 issues, 1.3 M words, 409 citations). These do not appear in the analysis.

All three channels’ third period covers the same calendar interval (1990–2020), so the cross-period comparison is symmetrical. One structural feature must nevertheless be flagged: the American print periodicals thin out toward the end of the period — Adventist Review from 2016 onward, and while Ministry remains formally monthly throughout, the Review’s annual word volume falls from 1.1 million in 2005 to 0.25–0.42 million per year after

1. The consequences of the small-denominator late years are discussed in

sections 5.1 and 7.

One asymmetry remains: the Romanian middle cell. Between 1946 and 1989, Curierul carries only 194 accepted citations in total. This is not a sampling shortfall but one of the results itself (5.7) — but it has a statistical consequence: this cell is weak for thematic analysis.

3.4 Periodisation

The three major periods — 1920–1945, 1946–1989, 1990–2020 — form a common comparative grid. Its boundaries follow the political caesuras of Romanian history (the end of the war, and the regime change); in American history the natural fault lines do not fall in the same places. The purpose of the common grid is not to model each side’s internal development optimally, but to make directions comparable across sides. The internal American controversies are therefore also examined separately, in a finer periodisation (5.6), in an exploratory capacity.

4. Method

4.1 The Unit of Measurement

The analysis cuts out a ±50-word neighbourhood around every accepted White reference, and codes this. The choice reflects three considerations.

Why local? Because the claim under examination is local. Document-level co-occurrence yields a near-tautological result in a denominational periodical.

Why words, and not characters? Because the word is the more stable reading unit across languages. A fixed character count would — owing to the longer average word length of Romanian — systematically give a shorter neighbourhood than for English.

Why fifty? Because this is roughly a paragraph’s worth of text in each direction, i.e. the citation’s immediate argumentative context. The number remains a free parameter nonetheless, so every major result is also reported at ±25 and ±100 words (5.8). The three widths are built from the same detected citations and the same random positions, and are therefore directly comparable, not three independent experiments.

4.2 The Citation Detector

The study’s precondition is that White references can be reliably found. The detector searches for the title variants of seventeen White works (or work groups) in all three languages — The Great Controversy, The Desire of Ages, Steps to Christ, Testimonies for the Church, Patriarchs and Prophets, Prophets and Kings, The Acts of the Apostles, The Ministry of Healing, Christ’s Object Lessons, Education, Thoughts from the Mount of Blessing, Early Writings, Messages to Young People, the Sabbath School Testimonies, Gospel Workers, the health-counsel volumes, and White’s articles published in the periodicals themselves — together with variant forms of the author’s name and the formula “the spirit of prophecy”.

Bare title search on this material is not merely noisy, but noise-dominated, and the direction of the bias would produce misleading results. A few real examples: the Romanian mărturii (“testimonies”) is also the standing rubric title for personal conversion accounts, so a raw search would make the Testimonies by far the most frequently cited work — an artefact of the rubric, not of citation practice. Faptele apostolilor is far more often the biblical book of Acts than White’s volume of the same title. educația, marea luptă, and viața lui Isus are, moreover, ordinary word combinations precisely within this literature.

The detector therefore assigns an ambiguity class to every title, and accepts homographs only when an anchor licensing citation stands beside them. Four kinds of anchor count: the author’s name, a pair of quotation marks, the formula “the spirit of prophecy”, and citation apparatus (page number, volume number). For biblical homographs, a chapter:verse-form reference is an absolute veto — it overrides every anchor. A periodical’s mention of its own name counts as a citation only alongside a strong anchor.

Hits fall into three confidence grades. High: an unambiguous title, or a homograph with a strong anchor. Medium: a homograph with a weaker anchor. Rejected: an anchorless homograph, or an ordinary common-noun title with a weak anchor. Only the high and medium grades enter the analysis; the rejected hits are also recorded, so that the acceptance/rejection ratio remains verifiable.

The detector’s window is deliberately narrow (100 characters), because on the two-column pre-war scans, text further away is often physically adjacent but content-independent, coming from a different column: a wide window here does not gain recall, but generates false anchors.

Finally: every run of the thematic analysis compares its own confidence-grade distribution against the output of the citation counter’s independent run, and aborts before writing anything out if they diverge. This rules out the possibility that the re-implementation that also returns positions could drift unnoticed away from the original detector.

4.3 Operationalising the Themes

This section answers the question of what makes a passage of text count as the neighbourhood of a given theme.

4.3.1 Basic Principle

The measurement does not interpret the text. Every theme is associated with a few dozen expression patterns; if at least one of these occurs in the ±50-word window, the theme’s marker switches on, otherwise not. There is no gradation and no judgement call.

This has both a cost and a benefit. The cost is that the measurement gauges the linguistic surface: a paragraph that circumscribes the ministry performed in the heavenly sanctuary while avoiding the fixed terminology is not captured. This is a systematic, not a random, distortion — more sophisticated, circumlocutory prose is undermeasured relative to prose that repeats the set terms. The benefit is that the measurement is transparent and replicable: the list can be published (Appendix F; the content background of the themes and examples of borderline coding cases are given in Appendix G), and the run is reproducible.

This is the classic trade-off of dictionary-based content analysis, ever since the General Inquirer mechanised the procedure.¹⁵

4.3.2 The Eleven Themes and Their Three Sources

The study does not work with a new lexicon built for this question, but draws its themes from three toolsets that existed independently of it and were built for other purposes. This matters: the lists were not tailored with knowledge of the result.

a) The sanctuary complex. The inclusion filter was originally a gate: it decided whether an article was on-topic at all. A single list suffices for a gate. For an emphasis question, however, it must be split apart, because it conflates three distinct things: the judgment (what happens), the place (where), and the date apparatus (since when). The distinction matters historically: a periodical can keep the date and drop the judgment, or vice versa, and it is precisely such a shift that the comparison across periods seeks. The split is thus a partitioning of the list, not an expansion: the union of the three compartments returns the original core. A fourth theme, high-priestly intercession, is built from the filter’s peripheral-term list (high priest, intercession, day of atonement, blotting out of sins, scapegoat, heavenly books/records) — the doctrine’s liturgical-christological facet, which can appear even without the core terminology. I call the union of the four themes the “sanctuary complex”.

b) The remnant axis. “Remnant” is by itself immeasurable: it is both an everyday word (a remnant of material) and a neutral biblical noun. The ecclesiological content lies not in the word but in the accompanying head noun, so every pattern requires it:

• institutional: remnant church, Ro. biserica/bisericii rămășiței

(“the remnant church”) — the denomination is identical with the remnant;

• biblical: remnant people, remnant message, Ro. poporul rămășiței,

solia rămășiței (“remnant people”, “remnant message”) — mission, not membership roll;

• contested: phrasing that explicitly denies the institutional

identification.

Bare occurrences of “remnant” remain in a separate, “unmarked” category, and are never assigned to either pole.

The distinction did not grow out of the corpus, and does not stem from a single author. In László Gallusz’s formulation, the Book of Revelation speaks of a remnant message and a remnant people, but not of a remnant church — and what is at stake is whether remnant-identification functions as a privilege or as a duty of service. Jenő Szigeti inverts the same point historically: it is the self-definition that follows from historicist prophetic interpretation that grounds the “often unspoken sense of superiority” against which he himself sets the formula “remnant of the disappointed” — the historically contingent, humbled self-understanding.¹⁶

The axis also has a living, contemporary articulation, and this is not incidental for the purposes of this study. Sigve K. Tonstad’s 2025 volume explicitly proposes a self-understanding that goes beyond remnant-identification: it understands Adventist identity not as a badge to be defended but as a calling to be tested, purified and renewed. At its centre stands the christological turn that Tonstad’s earlier monograph worked out concerning the pistis Iēsou construction in Revelation: Rev. 14:12 places not primarily the believers’ correct faith but the faithfulness of Jesus at the definition of the remnant — which turns the concept from a boundary-marking criterion into a christological claim. Tonstad’s key term, “chastened” identity, sits strikingly close to Szigeti’s formula.¹⁷

The present study’s remnant axis is therefore not a reconstruction of a closed theological-historical dispute, but the operationalisation of an ongoing distinction. This will matter in section 5.5.

c) The rhetorical emphases — a change of mode and its consequence. The three emphases were originally built as weighted, document-level rates: every expression receives a weight of 1–3 (3 = unambiguous, 2 = strong, 1 = suggestive), and the dimension’s score is the sum of weights per thousand words. In this mode, a weight-1, everyday word (“grace”, “critics”, “objection”) contributes little mass, and the rate does the substantive work.

The present analysis, by contrast, examines binary presence in a hundred-word window. There, that same single weight-1 word tips the whole theme; one observed consequence was that a Romanian article on church music was coded as “assurance rhetoric”. For the presence decision, therefore, only expressions with a weight of 2 or higher count — this is not a new judgement on the instrument, but the application of its own calibration in a changed mode. It does have a consequence, however, and it must be stated: the four emphasis themes measured here are narrower than the same dimensions measured in the weighted-rate version. The two sets of figures are not interchangeable.

From the same family, but graver for validity, is the following decision: every expression referring to Ellen White has been removed from the apologetics list. The original reason for this was that, in an earlier version, 75.4% of the signal in the 1930s Romanian material came from a single White-term — meaning the dimension was in fact measuring citation-of-White, not doctrinal defence. For the purposes of the present study this carries particular weight: without this cleaning, the result “apologetics stands beside the White citation” would be circular, since the theme would in part be measuring the citation itself.

4.3.3 Calibration: What It Means in Practice

Three observed false hits, and the response to each. These show what lexicon maintenance actually involves.

The year 1844. This is the one element of the core list that is not by itself unambiguously doctrinal. In a citation neighbourhood it also turns up in bibliographic year ranges — the observed case was an English book title (Ellen G. White’s Prophetic Gift, 1844–1889) in a Romanian article on an entirely different subject. The pattern therefore now excludes hyphenated year pairs: a date range is apparatus, a standalone year is a claim.

The Romanian cu siguranță. siguranță is a weight-3 theological term (“certainty”), whereas cu siguranță is the everyday adverb “surely”, which in a short window on its own switched on the assurance theme. The pattern now excludes the form preceded by cu.

The surname “Ford”. The apologetics list includes Desmond Ford as a named disputant. The earlier pattern, however, also matched inside Oxford, Bradford, Rutherford, as well as mentions of the Ford Motor Company, President Gerald Ford, and a periodical contributor of the same surname: of 1,220 flagged English documents, 999 predate Desmond Ford’s entry onto the scene. The pattern now requires the form desmond ford or dr. ford.

This is not a list of errors, but an illustration that the quality of a dictionary-based measurement depends on the systematic hunt for false hits, not on the length of the list. Loughran and McDonald’s classic study showed the same for a specialist corpus: a general-purpose dictionary misfires conspicuously, and the fix is not fine-tuning but tracking the mistakes through to the end.¹⁸

4.3.4 A Coding Example

An illustrative — not an actual, article-drawn — neighbourhood:

“…and the work of judgment which began in 1844 in the most holy place of the heavenly sanctuary is now closing. The Great Controversy states that our great High Priest pleads His blood before the Father…”

The passage is a single citation, but appears in five theme rows. The coding is deliberately multi-label: “the investigative judgment in the heavenly sanctuary since 1844” is genuinely three themes at once, and declaring a winner would manufacture a distinction the text itself does not make. Two consequences for reading the results follow: theme proportions sum to more than 100%, and the sanctuary-complex row is not the sum of the four compartments, but their union.

4.4 The Reference Distribution

A bare proportion is uninterpretable on its own. If 2.5% of Review and Herald’s citations stand beside sanctuary-doctrine text, this could be a lot or a little — it depends on what the baseline would be.

The reference is therefore supplied by text positions randomly sampled from the same documents: a hundred positions per ten thousand words (at least three per document), cut with the same window size and coded with the same lexicon. On the English branch, this gives 1,596,742 reference windows alongside the 73,918 citation windows; on the Romanian branch, 178,596 alongside 4,173.

Two decisions require justification.

Why from the same documents? Because themes cluster into articles: in a sanctuary-doctrine article every paragraph is sanctuary-doctrinal, in a health article none is. This phenomenon — known in the literature as burstiness — is well documented, and is exactly what makes Poisson-based expectations misleading.¹⁹ A reference computed from a theoretical distribution would be too uniform, and would bias every measure upward.

Why not another period or another periodical? Because the measure becomes interpretable precisely by virtue of the comparison being within-cell: same language, same periodical, same period. This is also what neutralises the differing sensitivity of the English and Romanian lexicons. Accordingly, I draw no conclusion from cross-language comparison of raw proportions — the lift offers no protection there.

4.5 The Measure of Binding

The measure is the ratio of the observed proportion to the reference proportion:

lift = (theme hits / citations) ÷ (theme hits in the reference / reference windows)

If four out of a hundred citations stand beside sanctuary text, and two out of a hundred random text positions do, the lift is 2.00. A value of 1.00 means thematic indifference. A value below 1.00 is not an empty result either: it indicates that the periodical discusses the theme near a White citation less often than average — that is, it regularly argues without it.

The measure is not new: observed co-occurrence divided by what would be expected under independence is known in linguistics as pointwise mutual information (PMI), and in data mining as lift.²⁰ The present study reports it without a logarithm, because it is then directly readable.

One cell worked through in full. Review and Herald, 1946–1989: 31,267 accepted citations; of these, the sanctuary complex marker appears in 771 windows → share = 2.466%. Of the same 2,179 documents’ 537,409 reference windows, 1.431% are coded the same way. Lift = 2.466 / 1.431 = 1.72 [1.55–1.90].

4.6 Uncertainty

Confidence intervals derive from a document-clustered bootstrap, with 400 replicates. The clustering is not a technicality: citations within a single article share the article’s theme, and are therefore not independent observations. Computing the margin of error over 70,000 independent draws would make the interval many times too narrow. The unit actually drawn from the archive is the issue, so it is the issue that is resampled.²¹

The replicate matrix is built once per cell, and the same one is used for every theme. This is not only faster: it is what makes the themes’ intervals comparable — they are measured on the same resampled corpora.

Reporting threshold. The main text does not report a lift for fewer than twenty theme hits. In that case, the loss or addition of a single longer article would reshuffle the estimate, and the bootstrap interval excludes practically nothing.

4.7 Confirmatory and Exploratory Analysis

The primary investigation comprises nine cells: the sanctuary complex across three periodicals, in three periods. Every further theme and period breakdown is exploratory. The reason for this separation is the multiple-comparisons problem: examine enough cells and a “significant” result always turns up, and the freedom of the search itself generates the appearance of results.²² The study therefore restricts its confirmatory claim to the nine pre-designated cells, and treats the rest as hypothesis generation.

5. Results

5.1 The presence of citation and its variability over time

Before we can say anything about the environment of the quotations, we need to know how present White is at all in the individual annual volumes, and how stable that presence is. This is not mere preparation: the answer is itself a result, and it fundamentally changes how 5.3 should be read.

Table 4. Accepted White quotations per ten thousand words, in five-year blocks

Annual word counts are derived from character counts, with a conversion fitted per cell (periodical × period) to the normalised total word count.

Three things can be read off this table.

(a) Variability is moderate, and is not driven by individual years. The relative standard deviation of the annual values on the American channels ranges from 0.24 to 0.37 within a period; no single outlying year is able to shift the period-level estimates. The highest single-year value is Ministry’s 1955 volume (11.5 quotations per ten thousand words) — the year of the Adventist–evangelical dialogue.

(b) The American volume roughly halves after 1980 — then partly rebuilds in the 2010s. The peak runs from the mid-1950s to the late 1970s, after which the rate roughly halves: the Review and Herald’s annual median falls from 5.74 in 1946–1989 to 2.91, and Ministry’s from 6.62 to 3.62, in the third period. The trough is the decade and a half between 1995 and 2009 (R&H 2.3–2.8; Ministry 2.7–2.9); in the 2010s both periodicals climb back (2010–14: 3.26 and 4.32; 2015–20: 3.42 and 4.89). In these later years the print periodical is already monthly and of small word extent, so these cells should be read cautiously (see §7), though the direction is the same for both periodicals.

The two endpoints of the American curve under (b) coincide with two independently documented institutional events. In the peak years, between 1953 and 1957, the denomination’s Bible commentary was produced, whose editorial baseline requirement was that it should nowhere formulate a statement conflicting with White’s writings — in the wording of the reception-history literature, this in practice made “her views became canonical in practice.” The onset of the decline coincides with the 1980 Dallas General Conference session, which adopted the twenty-seven fundamental beliefs with a preamble stating that Seventh-day Adventists “accept the Bible as their only creed.”²³ The corpus data confirm but do not explain these two datings: the coincidence is not in itself causal evidence, but the measured turning points do not fall on arbitrary years.

(c) The Romanian curve is not a slope but a step. The collapse can be dated to a single year: still 5.32 in 1947, 0.96 in 1948, 0.05 in 1949, and then exactly zero accepted quotations across five annual volumes between 1950 and 1954. The return is equally sharp, but delayed: between 1990 and 1993 the rate still stays between 0.6 and 1.0, and the real jump comes only in 1994 (3.98). Some four years thus pass between the political turn and the restoration of citation practice.

It must be emphasised that this step is the step of named citation. The separate pre-screen in 5.7.1 shows that the reference to White did not cease even during the zero years — it only lost its name.

Read together with the binding result of 5.3, this yields a two-phase, mirror-symmetric pattern that neither measure would reveal on its own. Between 1980 and 2010 citation thins out while its binding strengthens (at the period level 1.72 → 2.02, and 1.42 → 1.79 respectively): reference to authority becomes rarer and, at the same time, more specialised. In the 2010s the two indicators again move in opposite directions, now reversed: volume partly rebuilds, while doctrinal binding, in the fine-grained breakdown, sinks towards chance level (5.6). The volume of White citation and its doctrinal direction therefore do not move together — and it is precisely this decoupling that neither plain citation counting nor plain binding measurement, on its own, would show.

5.2 The minority character of doctrinal citation

The most important compositional result is not any one doctrinal row but the “no coded theme” category: this remains between 83.8 and 89.2% across the nine cells (Appendix B). The larger part of White citation therefore does not stand in the immediate environment of sanctuary doctrine, remnant identity, imminence, apologetics, judgment-fear, or assurance.

This does not mean that 85% of the quotations are contentually empty. The remaining environments may contain devotional, pastoral, educational, health, and organisational content; the present lexicon does not classify these. The result thus shows the minority character of the doctrinal use under examination, not the distribution of the other uses.

One detail nonetheless nuances the picture. The “no coded theme” category also has a lift, and this is below 1.00 in every cell, typically between 0.93 and 0.98; in all six American cells the upper edge of the interval also remains below 1.00. That is, quotation environments are slightly but consistently less theme-less than random text locations. The White quotation is not thematically indifferent — it is simply that most of it is not placed alongside the eleven themes examined here.

This result bears directly on the view cited in 1.2. If the official press had regularly invoked White as “a convenient substitute for Scripture” or as “the final doctrinal arbiter,” this measurement would have to show it — and it does not. In the practice of the official periodicals, invoking authority is overwhelmingly not doctrinal proof; doctrinal confirmation is a targeted, minority operation.

This does not refute the cited perception, but places it. The claim remains tenable in that the “third-testament”-like use was visible, memorable, and debate-generating, without quantitatively dominating the practice: the great mass of citation took place elsewhere, in the devotional and practical register. This distinction — salience versus frequency — is exactly what a case-study-based literature cannot do, but a measurement running across a complete periodical run can.

This result is also consistent with how the reception-history literature characterises White — as a preacher and devotional writer whose theme “was how to live in the time of the end”, rather than a systematic theologian — and with the observation that her writing mode is homiletic: the purpose of a homily is persuasion and guidance, not propositional proof, and not every part of it carries equal weight.⁵²⁴

5.3 The main contrast: the binding of the sanctuary complex

[Figure 1. Local binding of the sanctuary complex and White quotations, three periodicals × three periods; point estimate and 95% bootstrap interval.]

Table 1. Local binding of the White quotation and the sanctuary complex (lift, 95% CI)

The third column for all three periodicals covers the years 1990–2020 (3.3). Because of the publication threshold, the point estimate for the middle Romanian cell is not reported.

The point estimates for the two American channels move in the same direction, and both trajectories are monotonically increasing. The Review and Herald’s closing value of 2.02 means that near the sanctuary theme a White quotation is twice as frequent as at random text locations in the same periodical; its trajectory is distinct from 1.00 in all three periods. Ministry’s first cell does not yet differ from chance, but the closing 1.79 already reliably does.

The Romanian channel shows the opposite direction between the two reportable endpoints: the interwar binding of 1.68 falls to 1.06 after 1990, that is, it becomes indistinguishable from chance level.

The interwar picture is thus the inverse of the usual “centre versus periphery” formula: at that time authority binds more tightly to doctrine at the periphery. The divergence of the trends is the study’s central empirical result — with the coverage asymmetry of the third period taken into account (3.3, §7).

5.4 Choice of source: which work for which theme

Table 2. Over-represented White works in the English-language citation environments

A different reference basis. The multipliers given here are not the lifts of Table 1. There, the reference point was the random text location; here it is the periodical’s own distribution across works. The 4.7× means: in the environment of the investigative judgment, The Great Controversy is quoted at four and a half times the rate that the work occupies within all White quotations.

The result is ordered, and the ordering is itself the finding. The historical-prophetic doctrines — judgment, sanctuary, 1844 — draw on the visionary-chronological corpus: The Great Controversy’s grand historical-theological narrative and Early Writings’ early collection of visions. Institutional self-identification, by contrast, mobilises a different set of sources: there, Testimonies — the corpus of church discipline and organisational leadership — comes to the fore.

This is not self-evident. If ecclesial self-definition operated primarily as a prophetic claim, we would expect the visionary works alongside the remnant themes as well. The data show the opposite: the periodicals underpin institutional remnant identification not with prophetic vision but with ecclesiastical admonition.

The result refines an existing thesis from reception history. According to Jonathan Butler, every reading community forms its own “canon within the canon,” and within the White corpus this has shifted historically: for White and her contemporaries, Testimonies were the truly inspired writings, the Controversy series belonged to the “merely” edifying category, whereas today’s Adventists turn instead more toward The Desire of Ages, Thoughts from the Mount of Blessing, or Steps to Christ.²⁵ The present data add to this that the “canon within the canon” is not a single ranking but theme-dependent: the same periodical, in the same annual volume, calls on the visionary corpus for a prophetic question and on the disciplinary corpus for an institutional one. The hierarchy is thus not reorganised period by period as a single list, but divided by function.

One caveat for the work-level analysis. The cited “work” is itself an assembled construct: White’s major volumes were compiled by her assistants — above all Marian Davis — from her earlier writings, adapting and recycling them; an estimated 90% of Steps to Christ derives from earlier texts.²⁶ The work–theme association therefore measures the citers’ choice of work-category, not the layers of the texts’ composition. This is not a flaw for the purposes of the present study — the periodical editor, too, knew the volumes, not their genesis — but it confines the reading of the finding to the side of reception.

In the Romanian corpus, the largest theme–work cell numbers only 62 quotations, so I do not report a work-distribution estimate on this branch.

5.5 Two versions of remnant language

In the Review and Herald’s earlier periods, both forms are bound to White quotations. The lift of the institutional remnant is 1.95 between 1920–1945 and 2.04 between 1946–1989; that of the biblical remnant is 2.18 and 2.07 respectively. The corpus therefore does not attach White to one concept or the other: binding is visible for both.

Usage that denies institutional identification did not appear in the environment of a single White quotation. This zero result must be read cautiously, and is best understood from the side of the operationalisation: the samples for the “contested” category are deliberately narrow, requiring an explicit denying construction. A text that merely passes over institutional identification in silence, or relativises it indirectly, is not captured. The result therefore says: openly stated denial does not stand alongside a White quotation — not that every form of the position is absent from the corpus.

The weight of the zero result, however, comes from the fact that the denying position is not hypothetical. Gallusz’s and Szigeti’s position has been articulated for three decades; Tonstad’s 2025, six-hundred-page volume, moreover, was published under a denominational publisher’s imprint, explicitly as a programme of self-understanding beyond remnant identification, accompanied by a series of conversations (4.3.2). The position thus exists, is worked out, and has denominational publishing infrastructure behind it — only not in the environment of the official periodicals’ White quotations.

This finding is therefore not about the absence of the position, but about the internal division of labour within the denominational public sphere: the official, normative organ and the book publishing, periodical, and discussion culture operating alongside it do not represent the same speech space. Within the hundred years examined, the voice that questions institutional remnant identification remains outside the authority-invoking register of the official periodicals. This is consistent with the limitation of the corpus choice (§7), but is also its substantive consequence: whoever wants to see the full spectrum of the denomination’s self-understanding will not find it in the official periodicals.

5.6 Crisis periods (exploratory analysis)

The fine-grained periodisation does not carry the same evidentiary weight as the nine pre-designated cells, and the cells are built from the same data, merely in a different partition — not independent replications. The patterns below are therefore hypotheses, not confirmed claims.

Conradi (1931–1933). Ludwig Richard Conradi’s break touched both sanctuary doctrine and the assessment of White’s inspiration at once; as Europe’s leading Adventist organiser, his objections were not peripheral.²⁷ In these years the Review and Herald shows a lift of 1.60 [1.09–2.13] at a moderate reference level (share 3.4%, reference 2.1%): during the schism, White quotations in the periodical stood near the sanctuary theme more densely than average.

Questions on Doctrine (1957–1963). In the years of the 1957 volume born from the Adventist–evangelical dialogue, and the internal controversy that then unfolded around it,²⁸ both American periodicals show strong binding. Review and Herald: of 6,280 quotations, 145 involved, share 2.31%, reference 1.21%, lift 1.92 [1.43–2.36]. Ministry: of 1,860 quotations, 78, share 4.19%, reference 2.16%, lift 1.94 [1.33–2.65].

Glacier View (1980–1983). This cell is the most instructive, because the two indicators diverge. Ministry at this time carries 8.4% of its quotations near the sanctuary complex — the highest share in the entire table — yet the lift is only 1.54, and its interval overlaps 1.00. The reason lies in the reference level: 5.48% of the periodical’s random text locations are also sanctuary-related in this period. During the Ford controversy the periodical was saturated with the theme.²⁹ If we looked only at the share, we would say: “during the crisis, Ministry invoked White’s authority for sanctuary doctrine to an unprecedented degree.” Knowing the reference level, the correct statement is different: the periodical was writing about almost nothing else at the time, so the quotation did not need to be specially summoned — sanctuary doctrine here was not a passage to be supported but the subject itself. The Review and Herald, in the same epoch, shows a lift of 1.88 [1.47–2.29].

Ministry’s 1997–2010 lift of 2.28 [1.37–3.34] is already the trace of a later phase: by then the institutions and publications of the post–Glacier View apologetic consolidation had become embedded in ministerial discourse. The corpus data support this reading but do not prove a causal link.³⁰

2011–2020: binding retreats. According to the fine-grained breakdown, binding peaks in both American periodicals in 1997–2010 (Review and Herald 2.34 [1.50–3.19]; Ministry 2.28 [1.37–3.34]), then falls back in the final decade: Ministry to 1.42 [0.78–2.21], the Review and Herald to 1.20 [0.69–1.75] — the latter no longer distinguishable from chance. The rising period-level trajectory (5.3) thus derives from the two decades 1990–2010. The cells are small (k = 28 for both periodicals), and thin printed volumes make up the end of the period, so this is a directional signal, not a confirmed reversal. Read together with the volume rebuilding of 5.1, it is nonetheless noteworthy: in the last decade, citation becomes more frequent while its doctrinal binding loosens.

5.7 The Romanian break

Between 1946 and 1989 the Curierul Adventist publishes only 0.34 accepted White quotations per ten thousand words — one-twelfth of its own interwar rate — while in the same period the American periodicals are at their own peak (5.82 and 6.60).

The period average, however, conceals the shape of the phenomenon. According to the annual series in 5.1, this is not slow attrition but a two-year collapse: still 5.32 in 1947, 0.96 in 1948, 0.05 in 1949, then, across five consecutive annual volumes between 1950 and 1954, not a single accepted White quotation appears. The return is likewise sharp but delayed: between 1990 and 1993 the rate still remains between 0.6 and 1.0, with the jump coming in 1994. Four years thus pass between the political turn and the restoration of citation practice — suggesting that the practice was not merely a function of the ban but had become entrenched as an editorial culture, requiring its own time to rebuild.

This should not be read as a simple disappearance of White’s influence, for two reasons.

First, the official Romanian church press operated under state supervision, censorship, and publication restrictions.³¹ Second — and this follows from the measurement itself — the detector finds only marked citation (4.2): anonymous borrowing and unanchored paraphrase remain invisible to it. This is precisely the distinction Hays draws between quotation and echo (§2).

A narrower claim thus follows from the measurement on its own: between 1946 and 1989, marked White citation collapsed in the Curierul. Whether the White material itself receded to the same extent cannot be decided by this method; that requires a separate survey of unnamed and untitled borrowings. This survey has been carried out, and the next two subsections give the other answer: the practice did not disappear, but changed attribution.

The post-1990 lift of 1.07 should likewise be kept within the bounds of the official periodical: the post-communist religious public sphere pluralised, and the practice of a single official organ does not describe the whole of Romanian Adventism.³²

5.7.1 Substitute attribution

The Curierul’s complete annual run between 1950 and 1980 — 219 issues, 3.69 million words — underwent a pre-screen independent of the main detector. The procedure makes testable the hypothesis that the period’s coercion affected not the content but the naming: White’s name and the titles of her works could be dropped while her text continued to circulate. Four marker classes were built, each for one form of concealment: (a) vision formula, from White’s first-person register (”mi-a fost arătat”, ”am văzut că”); (b) substitute attribution, that is, name-substituting invocation of authority (spiritul profetic [“the spirit of prophecy”], serva domnului [“the Lord’s servant”], solul domnului [“the Lord’s messenger”], pana inspirată [“the inspired pen”]); (c) title without name; (d) long quotation without attribution. No marker is proof by itself: the machine step hands over positions, not judgments, and every strong hit was exported with a ±60-word context for manual reading (Appendix H).

The result points in one direction: the period’s citation apparatus was not missing but replaced. In the same 219 issues in which the main detector accepts 25 named quotations, name-substituting invocation of authority appears in 263 places.

The two numbers do not measure the same thing, and cannot be read as a single ratio. The 25 are accepted quotations; the 263 are places of invocation — points where the text invokes White’s authority without naming her. How many independent, verbatim White borrowings lie behind these 263 places does not follow from this number: a single article may return to the same formula more than once, and the source of the quotation following the formula was not identified. The number therefore does not claim that there are ten times as many White quotations in the periodical as the detector sees. It claims that name-substituting invocation was the period’s dominant form, and the named quotation the exception — and this claim is independent of how many quotations belong to any one formula. The study therefore nowhere reports the ratio of the two numbers: the 263 denotes places of invocation, not quotations.

The typical shape of the formula is: “the spirit of prophecy says in this connection: ‘…’”RO 1, followed by a verbatim quotation, with no source given. Involvement is not confined to a few authors: of the 219 issues, 84 carry at least one such place, and of these, 73 contain not a single named quotation.

Table 5. Named quotation and name-substituting invocation in the Curierul Adventist per ten thousand words, 1950–1980

The first numeric column carries the accepted quotations of the main analysis — its values are identical with the Curierul column of Table 4, which also shows the agreement of the two independent word-count calculations. The two numeric columns count different units — the first source-identified quotations, the second places of invocation — so their ratio is not a citation rate. The last row covers only the 1980 volume, so it cannot be compared with the full five-year blocks.

The table says two things.

(a) The practice is continuous throughout. Even in the 1950–54 block — in which the main analysis finds exactly zero accepted quotations — substitute attribution is present. From the mid-1950s to the mid-1970s the rate is stable between 0.35 and 0.45: not a residual phenomenon but an established editorial procedure. The spike in the last two blocks (1.71 and 1.19) coincides with a broader change in the periodical’s citation culture around 1975; this is discussed in Appendix H.

(b) The borrowing is quotation-based, not impersonation-based. The vision-formula marker returned zero hits across the whole material. The periodical, that is, did not write in White’s voice; it published her text, under disguised attribution. This is an important distinction: it is not a matter of imitating the prophetic role, but of omitting the naming.

A limitation must be stated here too. The presence of the marker reliably signals invocation of authority; whether the quotation that follows is verbatim from White was not checked by source identification. The manual reading of the exported hits identified recognisable White passages (5.7.2), but the item-by-item tracing of the 263 places back into the White corpus is a further step. The figure of 263 is therefore a measure of the name-substituting invocation practice, not a count of confirmed White quotations.

5.7.2 What flowed under substitute attribution

The manual reading also gives an answer from the side of content, and this bears on the interpretation of the main analysis.

The sanctuary complex, too, went under disguise. This is the most important finding. The densest issue of the 1957 volume teaches the investigative judgment under the “spiritul profetic” attribution: “we now live in a time when the investigative judgment is drawing near its close; the sealing of God’s servants on their foreheads…”RO 1 The sanctuary study in the 1979 volume introduces White’s text on the heavenly sanctuary and on high-priestly intercession with the serva domnului formula (“The Lord’s servant says the following about the heavenly sanctuary…”)RO 2, marks another White passage with an abbreviated siglum (“T.V 385”, i.e. The Great Controversy), and builds on these its own proposition: “since the investigative judgment is taking place now, the time since 1844 is a solemn time.”RO 3³⁷

The weight of this finding must be stated precisely. It does not show how much sanctuary-related White reference there was in the Curierul’s middle period: the close reading of two issues establishes the existence of the phenomenon, not its magnitude. It does show, however, that the silence is not total: sanctuary doctrine — the main analysis’s primary theme — continued to flow through the periodical via name-substituting invocation. The low values of the middle cell should therefore be read as a lower bound, and the missing point estimate for that cell (Table 1) is not merely a matter of sample size: in this period the measuring instrument systematically misses invocations that exist. Exactly how many, only a full, source-identified processing of the name-substituting places could decide.

The other content layers. Justification by faith and the 1888 Minneapolis message are conspicuously frequent behind the “spiritul profetic” invocations; so is the classic Testimonies material — dress, child-rearing, speech, tithe, health — in the same way (“through the inspired pen we are told”, and then: “fresh air will prove far more beneficial to the sick than medicines”RO 4); and the 1978–79 church-history series quotes at length from White’s 1864 testimony on the pre-history of the work in Romania.³⁸

Regime-conforming use of authority. This layer of material is politically the heaviest, and it is not merely a matter of framing. The May–June 1963 issue’s article by V. Florescu — “Negru și Alb 1863–1963” [“Black and White 1863–1963”], marking the centenary of American slave emancipation — deliberately recalls period “attitudes, acts, and statements” in order to give today’s believers a “model” of how “the Adventist ministers of that time and the Lord’s servant took the side of their era’s progressive forces, taking an open and fearless stand against man’s oppression and exploitation of man, in black slavery.”RO 5 The stance is not fabricated — White’s anti-slavery statements are documented — what is regime-conforming is the frame and the vocabulary: abolitionism speaks here in the period’s Marxist-Leninist formula, as a model for today’s believers. The September–October 1974 issue’s article on culture and science quotes, in successive columns, first an official science-policy text on “overcoming backwardness”,RO 6 then continues: “All this proves that we, too, as Seventh-day Adventists, have hearts wide open to science and culture, just as the Lord’s servant likewise well explains”RO 7 — followed by the sentence from Education: “True education does not ignore the value of scientific knowledge or literary acquirements.” Here the White quotation stands as confirmation of agreement with official cultural policy, in the same article that speaks of the values found “in this blessed Romanian land.” The November–December 1978 issue — the festive issue marking the 60th anniversary of the Union of Alba Iulia, and at the same time the densest issue in the entire pre-screen — is the most nuanced case. The patriotic anniversary article (“A 60-a aniversare” [“The 60th Anniversary”]) and the devotional writings quoting the “Lord’s servant” are here separate texts, in adjacent columns: they stand together at the level of the issue, not within a single sentence. In the same issue, however, Union secretary M. Pârvan’s report on the ministers’ orientation conferences of November 1978 treats, in a single text, the two topics that the conferences’ own agenda had also linked: the paper on “Sola Scriptura and the Spirit of Prophecy”RO 8 and the “social paper” devoted to the anniversary of the union — and the report closes with the closing quotation of the social paper, the promise to preserve the nation’s “unbreakable unity.”RO 9 Here the connection is not typographical but institutional.⁴⁰

Two things must be drawn out of this, and together they give the finding its weight.

The first: the name-substituting formula not only concealed the authority but also made it mobilisable. On one possible reading, as a nameless, nationality-less “Lord’s servant,” White could be fitted into the loyalty language of the national-communist public sphere, into which a named, nineteenth-century American author would have fit only with difficulty. This is an interpretation, not a measured result; but it shows precisely, at the point where the study’s question is sharpest, the connection between the form of citation and the political usability of authority.

The second: the phenomenon is occasion-bound, not systemic. Of the 263 name-substituting places, machine counting finds only 7 — a lower bound — lying within ±60 words of regime vocabulary — patriotic, “progressive,” socialist, national-unity terms — across five issues, three of them in the 1978 festive issue — and according to the page scan these arise from adjacent columns of a devotional and a patriotic article. This is no more, indeed somewhat less, than at the random text locations of these same issues (2.7%, versus 3.2%; lift 0.83; Appendix H): the formula was not attracted to loyalty rhetoric, and name-substituting authority remained predominantly within a devotional and doctrinal frame. Loyalty performance concentrates in the anniversary and festive issues — but there, openly, in the period’s ideological vocabulary. Citation practice under coercion, then, was not merely concealment: on specific occasions, the concealed authority was also drawn into the regime’s public language.³²

Policing apocrypha with the returning apparatus. The 1980 volume’s question-and-answer column (“Citate apocrife atribuite Servei Domnului” [Apocryphal quotations attributed to the Lord’s servant], signed D. Popa) deals explicitly with apocryphal quotations attributed to the Lord’s servant. Its starting point is that “almost from the very beginning” of her work statements had circulated “attributed to prophetic inspiration, but never uttered by Sister E. G. White”, and that the church recognises as genuine only what rests on published or unpublished sources “known to be authentic”.RO 10 The refutation uses precisely the apparatus that had been missing in the earlier decades: work and page (Early Writings 40 and 269–273, Selected Messages II 263, volume IV of the Testimonies). One of the refuted claims — that “entire churches and conferences” will apostatise — is, according to the column, “a pure invention”, circulated above all by those who, “being at odds with the church and the organisation, seek to denigrate the church and its leadership”.³⁹ The finding returns to the study’s main question in two directions. On the one hand, authority circulating without name, title and page had become unverifiable, and the gap now had to be filled by policing. On the other, the returning reference apparatus is not neutral: here it is the organisation’s instrument for separating White’s authority from dissident use directed against it. The form of reference is not an externality — it is a condition of how authority works, and of who controls it.

5.8 Robustness

Five checks run on the main result; none reverses it.

Corpus window. An earlier version of the study covered the American third period only through 1990–2005. Adding the 2005–2020 material left the Review and Herald’s point estimate practically unchanged (1.95 → 2.02), and moderated Ministry’s (2.47 → 1.79), while both intervals narrowed and remained above 1.00. The trend claim is thus not an artefact of the 2005 cut-off — the earlier coverage limit turned into a robustness check.

Density of the reference draw. The reported figures are built on one hundred random positions per ten thousand words; the cell-level lifts of a fivefold sparser, independent draw differ by at most 0.20, typically ±0.08. This is the order of magnitude of the draw uncertainty, and it does not affect the status relative to 1.00 in any cell.

A stricter reference. An alternative null model draws random positions only from documents that contain at least one accepted quotation; this way the difference between quoting and non-quoting article types has less influence on the reference.

Table 3. Lift of the sanctuary complex under a reference restricted to citation-bearing documents (from the pre-expansion run)

The table comes from the run before the corpus expansion: the data for the first two columns are independent of the expansion, the third refers to the 1990–2005 window. The American point estimates’ rising sequence and the late high binding hold up in this version too. The Romanian point estimates still decline, but the early interval now touches 1.00. The early Curierul’s binding is thus more sensitive to the choice of reference than the American results — a further reason to read the Romanian trajectory as a descriptive pattern rather than a proven period difference.

Window width. The ±25, ±50, and ±100-word windows do not reverse the direction of the main point estimates (Appendix C).

Media shift. The American print periodical moved to born-digital production after 2005, and to monthly publication from 2016. If the genre composition of the remaining print material had broken, this should show up in the reference distribution. It does not: the reference share of the “no coded theme” category in the Review and Herald is 92.2% → 91.5% → 93.2% → 92.0%, and in Ministry 92.5% → 91.0% → 93.4% → 92.6%, across the 1984–1989, 1990–1996, 1997–2010, and 2011–2020 epochs — there is no jump at either the 2005 or the 2016 boundary. The print periodical’s thematic structure remained continuous.

6. Discussion

6.1 What kind of authority does the official periodical suggest?

To the question posed in the introduction — what kind of White authority a normative, partly overseen denominational organ suggests to the reader — the results give a three-layer answer.

First: overwhelmingly not doctrinal proof. In the practice of the official periodicals, White is not primarily a dogmatic arbiter. The great majority of quotations stand in an environment that none of the doctrinal and rhetorical themes examined covers. The image this practice builds in the reader is primarily one of pastoral, educational, and practical authority.

Second: within the doctrinal minority, however, targeted, and increasingly tightly, within a thinning practice. Where authority is invoked at a doctrinal point, this is not random, and it strengthens period by period on both American channels (1.72 → 2.02, and 1.42 → 1.79). This finding, however, only gets its true edge alongside 5.1: across the greater part of these same decades, the overall volume of White citation roughly halves. Between 1980 and 2010, the periodicals quote less and less often, but the remaining quotations increasingly concentrate on doctrinal points; in the 2010s the two indicators diverge again, this time the other way — volume rebuilds, binding loosens in the fine-grained breakdown (5.6).

Together the two describe an image of authority that neither measure would produce on its own: White’s presence in the everyday fabric of the official periodicals retreats for three decades after 1980, while the remaining reference specialises — invoked less often, but more purposefully. This is consistent with the idea that reference to authority gradually withdraws from routine devotional use toward the burdened doctrinal points; one possible reading of the pattern is a need for self-definition against inter-denominational pressure, since strengthened prophetic confirmation attaches to the doctrine felt to be one’s own and vulnerable, precisely in the decades when that doctrine is the subject of external and internal dispute. The crisis cells — Conradi, Questions on Doctrine — fit this, but do not prove it, and separating the competing explanations (editorial policy, periodical-format change, general shifts in citation habits) requires further measurement.

Third, and perhaps most interestingly: institutional self-identification is assigned not prophetic but administrative authority. According to the result of 5.4, alongside remnant-church identification it is not the visionary works but Testimonies that is over-represented. The claim “this church is the remnant” is thus, in the periodicals’ practice, underpinned not by apocalyptic vision but by church-disciplinary admonition. If citation practice also shows what kind of authority a claim needs, then the primary support of Adventist institutional self-definition in these periodicals is organisational, not prophetic.

6.2 Coercion and practice

The opposite direction of the Romanian trajectory suggests that political coercion did not simply thin out citation, but also rearranged what it was called upon to support. The interwar Curierul was the most tightly binding of the three channels; under censorship, marked citation collapsed, and after 1990 binding did not return: the now-free periodical, too, no longer calls authority to the doctrinal point.

The pre-screen of 5.7.1 sharpens this picture at one point and confirms it at another. It sharpens it, because it shows that in the middle period it was not the invocation of authority that disappeared, but its name: name-substituting invocation was the dominant form, the named quotation the exception, and sanctuary doctrine, too, travelled by this channel. Coercion thus did not excise the content, but replaced the form of citation — which supports precisely the study’s basic thesis: the form of citation is an independent institutional fact, not an externality of content. Where the name, the title, and the page number disappear, authority is transformed: it continues to function, but is no longer traceable, and therefore not contestable; the 1980 apocrypha column (5.7.2) deals precisely with this consequence.

It confirms, however, the survival explanation. Substitute attribution was not the habit of one or two authors, but a stable editorial procedure across thirty years (Table 5), and the restoration of named citation occurred only after 1989. A practice routinised to this depth does not stop in the year the ban ends — this gives substance to the otherwise merely descriptive observation that the restoration lags by four years (5.1), and makes it more likely that the weak post-1990 binding is partly an editorial legacy. The present data permit this explanation, but do not prove it — separating the competing explanations (change of readership, editorial generational turnover, a different share of international content) requires further measurement.

Finally, it adds a third feature: authority that had become nameless also became politically usable. On specific occasions — chiefly anniversary and festive issues — the “Lord’s servant” spoke in the period’s loyalty and ideological language (5.7.2). This was not the dominant form of the practice, but an occasion-bound performance; but it shows precisely that the form of citation decides not only whether authority is visible, but also what it can be used for.

6.3 Methodological lessons

Three lessons generalise to the quantitative analysis of the religious — and more broadly, institutional — press.

(1) The choice of coding unit is itself a theoretical decision. Article-level coding measures theme frequency; testing rhetorical claims requires a local unit. In a homogeneous corpus repeating one theme, document-level co-occurrence is close to tautological.

(2) Without a reference, a share is not a result. The Glacier View cell (5.6) shows this cleanly: the same data allow diametrically opposite claims depending on whether a within-channel reference level is reported alongside it.

(3) The dictionary’s mode of operation cannot be carried over unchanged. The same term list means something different as a weighted rate per thousand words and as binary presence in a hundred-word window (4.3.2). Transfer requires recalibration, otherwise everyday vocabulary drives the result.

7. Limitations

The lexicon’s validity is not estimated against a manual sample. There is no estimate of precision (what percentage of activated markers genuinely concerns the theme) or recall (what percentage of genuinely on-theme windows it captures). Because of this, I do not interpret the few tenths-of-a-percentage-point differences in share as an independent result; the main claim rests on the combined sanctuary complex’s larger, within-language lifts. The methodological literature on computational content analysis uniformly treats validation against a manual sample as a condition of validity.³³

There is no hand-coded functional sample. The study does not distinguish evidentiary, illustrative, disputing, and other uses of quotation. This is why the term “thematic binding” is used, and it is the most important next step.

The late print volumes are thin. From 2016, the Adventist Review publishes 12 issues a year and 0.25–0.42 million words; the 2011–2020 epoch cells are therefore small-denominator (k = 28–30 for both American periodicals). The online channel operating alongside the print periodical (adventistreview.org) falls outside the study’s scope — throughout, the object of the study is the official print organ. The suspicion of a break in genre composition is refuted by the continuity of the reference level (5.8), yet the late cells should still be read cautiously on their own; the Romanian sub-cells (10, 14, and 20 theme hits respectively) also move near the publication threshold, so the fine, epoch-level American–Romanian comparison of the third period remains weaker than the period-level one.

The corpus consists of official organs. A position not visible in them may be present in other forums; this particularly concerns the contested remnant interpretation (5.5) and the post-1990 Romanian pluralisation (5.7).

The measurement measures a linguistic surface. Discussion that avoids the terms and proceeds by circumlocution is systematically under-measured (4.3.1).

The measurement finds marked citation. Every claim concerns citation practice, not the actual presence of White’s texts (4.2, 5.7). The significance of this limitation can now be seen in one place: in the Curierul’s 1950–1980 material, name-substituting invocation was the dominant form, the named quotation the exception (5.7.1). No such survey was carried out on the American branch, so it is not known how great the unmarked borrowing is there; in a free-press environment the political motive for omitting the name is absent, but this does not measure the practice. Estimating the rate of unmarked borrowing on all three channels remains a task in itself.

The survey of substitute attribution is a pre-screen, not full coding. 5.7.1 rests on a machine marker set and manual reading of the strongest hits; item-by-item source identification of the 263 places within the White corpus was not carried out, and the marker set has no precision–recall estimate. The 263 is therefore a count of places of invocation, not of quotations, and cannot be compared as a ratio with the 25 named quotations. The conclusion drawn from it is therefore descriptive — it asserts the existence, density, and chronology of the practice — and does not stand in for confirmatory analysis.

OCR. Per-cell quality ranges between 0.90 and 0.95, so this is not what drives the differences; the effect is typically a reduction in recall, not a false hit.³⁴ The Curierul’s two lower-quality cells, however, are precisely the two most interesting ones. Appendix H’s two separate checks (unchanged OCR quality, independent control count) show that the 1975 turn discussed there is not driven by OCR.

Differences between periods. The reported intervals compare individual lifts to 1.00; no separate contrast interval has been constructed for differences between periods. Claims about trends therefore rest on the sequence of point estimates and the agreement of the robustness checks.

The German-language intermediary channel is not included. Interwar East-Central European Adventism obtained access to sanctuary-doctrine material in large part through the mediation of German-language literature; including this channel is a subject for a separate further study.

8. Conclusion

White citation in the three official periodicals examined mostly does not stand alongside the designated doctrinal themes: invoking authority as doctrinal confirmation is a minority, targeted practice, and this share is stable across every channel and period. The channels differ not in this, but in the composition of the remaining 11–16%.

The minority, doctrine-adjacent citation environment, however, is not random. The point estimates of the American binding of the sanctuary complex rise period by period, while the Curierul Adventist’s two more reliably measurable endpoints show the opposite direction.

Read together with the per-volume citation rates, this yields a double movement that neither indicator would show on its own: in the American periodicals, the volume of citation halves after 1980, while its binding strengthens at the same time. Reference to authority thus neither simply recedes nor simply gains in value, but specialises — becoming rarer and more targeted at once. In the 2010s the two indicators diverge again, now with the opposite sign: volume partly rebuilds, while doctrinal binding, in the fine-grained breakdown, is pulled towards chance level. On the Romanian channel this same practice does not fade but breaks off for a two-year stretch, and its restoration follows the political turn by four years.

The separate survey of the years under censorship, however, shows that it was not the invocation of authority that ceased there, but the naming: name-substituting invocation became the dominant form, the named quotation the exception, and sanctuary doctrine, too, continued to travel by this channel — the restoration of named citation had to wait until 1989. This is the sharpest case of the study’s basic thesis: the form of citation is an independent institutional fact. The same authority, if it loses its name, its title, and its page number, continues to function, but is no longer traceable — and hence not contestable either — and, as the festive issues show, can on occasion also be drawn into the regime’s loyalty language.

The choice of source reveals further structure, and this is the study’s most generalisable finding: the periodicals do not cite “White” in general, but assign different White corpora to different questions. The prophetic-chronological doctrines draw on the visionary works, institutional self-identity on the church-disciplinary Testimonies. Citation practice thus shows not only where authority is needed, but also of what kind.

A stronger functional claim requires manual coding. The present measurement reliably shows where White citation concentrates; what it precisely does in the text there is for the next validation step to decide.

Data, code and reproducibility

The confidence intervals were built from 400 document-bootstrap replicates, with a fixed random seed. The runtime environment was Python 3.9.6, NumPy 2.0.2, ARM-based macOS. Reference windows were built at one hundred positions per ten thousand words, at least three per document; there are 1,596,742 reference windows on the English branch and 178,596 on the Romanian. The cell-level lifts of a fivefold sparser, independent draw differ from those reported by at most 0.20, typically ±0.08 (5.8). The Questions on Doctrine window (5.6) is a separate run: the 420 documents of the years 1957–1963, with identical settings.

The annual rates in 5.1 are the only indicators that use an estimated denominator: the annual word count derives from the character count, with a conversion fitted per cell (periodical × period) to the normalised total word count. Every other indicator’s denominator is an exact word count or an exact quotation count.

The integrity of the citation detector is checked on every run: the confidence distribution must match, character for character, the output of the independent citation counter, otherwise the run aborts before writing anything. The theme coder and the window cutter have a regression test suite that fixes the calibration cases described in 4.3.3.

The theme lexicon, the detector’s work list, the run code, and the per-quotation results table (doc\_id, year, periodical, period, quoted work, confidence, anchors, eleven theme markers) are available from the author. The corpus’s sources are public archives (3.2); the issue identifiers needed to reproduce the texts are given in the manifest files.

The pre-screen of 5.7.1. The marker set, scoring, and outputs are in curierul_triage.py, the sigla control in curierul_sigla_scan.py, the calibration of regime vocabulary in curierul_regime_context.py; the per-document count series (curierul_triage.csv) and the hits exported with a ±60-word context (curierul_candidates.txt) are the raw material for manual reading. The range examined is the Curierul’s complete annual run from 1950 to 1980, with no sampling: 219 issues, 3,694,397 words. Appendix H describes the protocol and the withdrawn marker classes.

Validation steps still needed before publication. (1) A theme-stratified sample for estimating the lexicon’s precision and recall values, on the English and Romanian branches. (2) Manual functional coding of 200–300 quotation environments by two coders (evidentiary / illustrative / disputing / other), with inter-coder agreement. — The third step, previously listed here (the survey of the Curierul’s unnamed and untitled borrowings), has been completed; its result is in 5.7.1–5.7.2 and Appendix H.

Appendices

Appendix A — see Table A1 under 3.3

Appendix B. Full distribution of the citation load (share, %)

The columns sum to more than 100%: a window can belong to more than one theme, and the sanctuary-complex row is the union of the four rows beneath it, not their sum.

Appendix C. Window-width sensitivity (sanctuary complex, lift)

All three widths were computed from the same detected quotations and the same random positions, and are therefore directly comparable.

Appendix D. One complete cell with every metric

Review and Herald, 1946–1989 (31,267 quotations, 2,179 issues, 537,409 reference windows) — four themes as illustration:

The table shows four separate things, and they do not follow from one another. Imminence is the largest theme (9.3%), but it is not the most strongly bound (1.60). The institutional remnant appears in a tenth as many windows, yet its binding is stronger (2.04). A small theme can still be reliable: 289 hits are enough for a tight interval. And without the reference-rate column the first row could be misread — imminence is large because the paper is already full of it, not because the quotation pulls it in.

From the standpoint of the pastoral axis (§2) the second and third rows should be read together: the White quotation is bound to both poles, somewhat more strongly to assurance than to fear. Authority is therefore not solely a prop of the old, fear-based register — the newer, assurance-based mode of speech invokes it too. The substantive background of the themes is given in Appendix G.

Appendix E. The confidence grades

Appendix F. The linguistic markers of the eleven themes (excerpt)

Three editorial rules run through this list, and they explain the apparent gaps:

1. The remnant themes require the head noun. The bare word “remnant” / “rămășiță” never codes to either pole on its own.
2. The emphasis themes exclude weight-1, everyday words (“grace”, “comfort”, “critics”, “objection” and their equivalents), because in a hundred-word window these would tip the marker on their own (§4.3.2).
3. **The apologetics list contains no expression referring to Ellen White** — without this exclusion the White-quotation ↔ apologetics binding would be a circular result (§4.3.2).

Appendix G. The eleven themes: content, stakes, and how the coding decides

Appendix F lists the patterns; this appendix states what the themes mean, why they are included in the study, and — using real example sentences drawn from the system’s own regression test suite — how the coding decides borderline cases. The “does it trigger / does it not trigger” examples are therefore not illustrations added after the fact: each one runs as an automatic test before every run. Alongside them stand real windows from the corpus, quoted exactly as the coder sees them: lower-cased, Romanian diacritics stripped, with the OCR’s errors intact; Romanian examples are accompanied by a bracketed Hungarian translation [rendered here in English].

The sanctuary complex

Seventh-day Adventism was born out of the Millerite expectation that, on the basis of Daniel 8:14 (“unto two thousand and three hundred days; then shall the sanctuary be cleansed”), awaited Christ’s return on 22 October

1. After the “Great Disappointment” the movement kept the date but

reinterpreted the event: it was not the earth but the heavenly sanctuary whose cleansing began — Christ entered the second apartment of the heavenly sanctuary and began the investigative judgment of believers’ cases. This doctrine is the most distinctive and most attackable element of the Adventist belief system: no other denomination shares it, and the internal crises (Conradi, Questions on Doctrine, Ford) have repeatedly converged on it. The four compartments are the four faces of the doctrine:

Investigative judgment — the event: since 1844 a judgment-investigation has been proceeding in heaven. – Triggers: ”the investigative judgment began in 1844” (which also triggers the date theme). – Does not trigger: the bare word “judgment” — a common noun that can mean anything from the pronouncing of a verdict to a court report; only the fixed phrases count. – From the corpus (Review and Herald, 1916): “this work of examination of character, of determining who are prepared for the kingdom of god, is that of the investigative judgment, the closing work in the sanctuary above.” — the great controversy, p. 428. One window, two markers and the apparatus: “investigative judgment” triggers the judgment theme, “closing work” triggers imminence, and alongside them stand the title of the work and the page number.

Heavenly sanctuary — the location: the ministry proceeds in the heavenly sanctuary, in the holy of holies. – Triggers: ”christ entered the most holy place”. – Does not trigger: the bare word “sanctuary” — it can also mean a church building or a refuge; the theme requires a connection pointing to the heavenly place. – From the corpus (Review and Herald, 1915): “the typical sanctuary was cleansed by the blood of beasts, but the heavenly sanctuary will be cleansed by the precious blood of christ” — a passage citing page 488 of The Great Controversy; in the same window “high priest” also triggers the intercession marker. Its Romanian counterpart (Curierul Adventist, 1923): “domnul hristos este tot in sanctuarul ceresc” (“Christ is still in the heavenly sanctuary”) — citing volume IV of Mărturii profetice.

1844 / 2300 days — the date apparatus: Daniel 8:14, the year, Miller, the Great Disappointment. – Triggers: ”since 1844 christ has ministered” — the standalone year used assertively. – Does not trigger: ”prophetic gift, 1844–1889” — the hyphenated year range is bibliographic apparatus, not doctrine (§4.3.3).

High-priestly intercession — the christological side: Christ as high priest interceding, the day of atonement, the heavenly books. – Triggers: ”our great high priest ministers above”. – This compartment carries the doctrine’s pastoral ambivalence: the same language feeds both judgment-fear and assurance (see below).

The remnant axis

The keyword of Adventist self-understanding is the “remnant” of Revelation 12:17. What is at stake is whom this word denotes: the denomination as an organisation (“remnant church” — the boundaries of membership coincide with the boundaries of the remnant), or a broader biblical category not tied to an organisation (“remnant people”, “remnant message”). The difference is not hair-splitting: the first reading grounds a privilege and a boundary line, the second a mission. The study’s third category, the contested remnant, is speech that explicitly denies the institutional identification (§2, §4.3.2b; Gallusz, Szigeti, Tonstad).

– Triggers (institutional): ”god has a remnant church on earth”. – Triggers (biblical): ”the remnant people of god” — and conversely: the two examples do not activate each other’s patterns; the two sides of the axis are linguistically distinct. – Does not trigger: ”a remnant of cloth was left over” — “remnant” without the head noun is an everyday word, and never codes to either pole. – From the corpus (Ministry, 1945): “whereas, the spirit of prophecy calls for a ministry in the remnant church which will receive a thorough education and training” — the text of a General Conference resolution on pastoral training: the gift of prophecy and the “remnant church” in a single institutional sentence. The association is not incidental: under the denomination’s 18th Fundamental Belief, the gift of prophecy is “an identifying mark of the remnant church and we believe it was manifested in the ministry of Ellen G. White” — that is, the White-quotation ↔ institutional-remnant binding that §5.5 measures is a connection fixed even in the doctrinal codification itself.²³

The four rhetorical emphases

These do not measure doctrinal content but a mode of speech — the pastoral repertoire whose historical shift (from fear toward assurance) is outlined in Chapter 2.

Imminence / urgency — the language of the shortness of time and the need for immediate decision; the continuous Millerite-heritage layer. – Triggers: ”the time is short and the end is near”. – Does not trigger: ”we must hasten the work” — “hasten” carries weight 1 in the lexicon, an everyday mobilising word; it is not enough on its own (§4.3.2c). – From the corpus: see the 1916 investigative-judgment window above (“the closing work”). Romanian example (Curierul Adventist, 1923): “evenimentele finale vor urma cu multa repeziciune” (“the final events will follow with great speed”) — quoting volume 7 of Mărturii.

Apologetics / boundary defence — the defence of doctrine and institution against named opponents (Conradi, Ballenger, Canright, Brinsmead, Ford, Glacier View) and boundary-defending vocabulary (“the pillars of the faith”, “distinctive doctrine”). – Triggers: ”conradi has attacked the doctrine”. – Does not trigger: ”the critics were many”, ”an objection was raised” — weight-1 generic polemical vocabulary. – Structural safeguard: the list contains no expression referring to Ellen White, since otherwise the White-quotation ↔ apologetics binding would measure itself. – That “landmark” is not a researcher’s construction but internal language is also shown by the historiography: in Anderson’s thought experiment — what would have happened in 1888 without White’s intervention — the old guard’s declaration of victory speaks precisely in this language: “we must reaffirm the old landmarks”.³⁵

Judgment-fear — the unsettling reading of the investigative judgment: sin kept on record, being found wanting in the balance, “too late”. – Triggers: ”every sin is recorded in the books”. – From the corpus (Ministry, 1947): “no direct appeals are made to those who are trembling in the balance” — Gospel Workers, p. 154: the image of trembling in the balance in the language of the pastoral appeal. Romanian example (Curierul Adventist, 1926): “glasul acesta face sa tremure cerul si pamantul” (“this voice makes heaven and earth tremble”) — quoting the closing scenes of Marea luptă.

Assurance — the comforting reading of the same doctrine: the interceding advocate, judgment in the believer’s favour, justification by faith. – Triggers: ”we have an advocate with the father”. – Does not trigger: ”the grace of god was upon them” — “grace” carries weight 1, an omnipresent religious word. – Romanian borderline case: ”cu siguranță că alegi ce-ți place” (“surely you choose what you like”) does not trigger — the everyday adverb is excluded —, but ”siguranța mântuirii este a noastră” (“the assurance of salvation is ours”) does (§4.3.3). – From the corpus (Ministry, 1929): “this statement of wonderful assurance, as found in gospel workers, pages 252 and 253, calls upon the minister…” — here it is not the quoted White text itself but the paper’s own framing sentence that carries the theme: the editor names White’s words as the source of assurance. Romanian example (Curierul Adventist, 1914): “noi trebue sa avem incredere in dumnezeu, sa credem in el si sa mergem inainte” (“we must trust in God, believe in Him, and go forward”) — from Mărturii pentru comunitate. The later official formulation of the assurance reading is given by the paper itself: according to Heppenstall’s 1987 Review article, “the chief value of Ellen White’s writings lies in their guidance for our spiritual lives. They call forth certainty and assurance in Christ.”³⁶

Fear and assurance are two separate poles, not two ends of one scale: a single text can score high on both. This is the classic Adventist sermon arc — first the alarm, then the comfort — and it would disappear in a combined index; that is why the system measures the two separately, and why the study reports them in separate rows.

And what the themes do not cover

The “no coded theme” category (§5.2) is not an abstraction for leftover cases but the corpus’s most frequent reality. A characteristic window (Review and Herald, 1955): “fatigue is forgotten; the strength seems to be increased. the intellect is aroused, the imagination becomes more vivid.” — the ministry of healing, p. 326 — on the physiological effect of tea and coffee, in a health article. A properly page-numbered White quotation; not a single coded theme triggers. The overwhelming majority of the quotations live in this kind of setting — devotional, health, educational, organisational; this is the register that the 85% measures.

Appendix H. The pre-screening of substitute attribution (Curierul, 1950–1980)

Scope. The complete run of Curierul Adventist between 1950 and 1980, with no sampling: 219 issues, 3,694,397 words, average OCR quality 0.906. The procedure is independent of the main detector (§4.2) and does not use its results; the number of named quotations enters the output only as a point of comparison.

Marker classes.

The document score is 3×VISION + 3×VAGUE + 2×TITLE; 101 issues scored above zero. Every strong hit was exported with a ±60-word context for manual reading (at most 12 per issue, which is why 261 VAGUE and 49 TITLE instances appear in the output out of the totals of 263 and 50 respectively).

Two withdrawn marker classes — results in their own right.

1. ANONQ is unusable. In the first run it flagged 214 of the 219 issues: the long, unattributed quotation is background noise in religious prose, not a signal. It was excluded from the scoring and kept only as a descriptive column.
2. Most of TITLE is mistaken. Most of the hits refer to the biblical book of Acts, not to White’s work of the same name; manual reading separated them out. Genuine, nameless use of a title is rare (for example “Hristos, lumina lumii”, RO-CA-1958-seq04; “Patriarhi și profeți”, RO-CA-1975-seq06). This marker alone is not suited to producing a series of figures.

The abbreviated source siglum and the 1975 turn. The density of substitute attribution quadruples in the last two blocks (Table 5). A separate count shows what happens in parallel with the source marking. curierul_sigla_scan.py counts the bracketed, page-numbered abbreviated source sigla; the detector’s Romanian title list contains spelled-out titles, not sigla, so these references remain invisible to the main analysis even where they otherwise carry a regular apparatus.

76 hits in total, in 18 issues, exclusively between 1975 and 1980; before 1975 the substitute formula is practically never followed by a source marking. In the same years the name itself (“sora white”) also sporadically returns.

Two checks support the conclusion that this is not a measurement artefact. OCR quality in 1975–79 does not improve but slightly worsens relative to the previous two blocks (0.902, after 0.931 and 0.927), so the appearance of the siglum does not stem from better recognition. A broader control count that does not presuppose White’s works — any bracketed abbreviation with a page number — produces the same step: 0.05 / 0.15 / 0.13 / 0.21 / 0.26 per ten thousand words in the 1950–74 blocks, then 2.35 and 2.77 in the last two. The paper’s citation culture changed around 1975, not its text recognition.

This pre-screening is an incidental observation, not a claim of the study: the restoration of named quotation, per Table 4, falls in any case after 1989. The 1975 turn adds that the reference form begins to be rebuilt earlier than the name — the more detailed investigation is a separate task.

Regime vocabulary around the name-substituting places (calibration). curierul_regime_context.py searches the ±60-word context of every name-substituting place for characteristic expressions of the period’s regime vocabulary (patriot-, patriei noastre, dragoste de patrie, progresist, exploatare, socialis-, partidul, comunis-, făurirea statului, statului național unitar, suflarea românească, pământ românesc, poporul român, Ceaușescu, construirea socialismului, republica socialistă), and does the same in random windows of the same issues (twenty per issue where available, with a fixed random seed). Result: 7 / 263 places (2.7%), in five issues — of which three are in the 1978 commemorative issue (per the page scan, adjacent columns of two separate articles), one each in 1963, in 1978 (a different issue), in 1979 and in 1980 —, against the reference rate of 168 / 5,260 windows (3.2%); lift 0.83. With three caveats. The expression list is short and unvalidated. Because of column interleaving, proximity signals a page-level co-occurrence, not necessarily a syntactic connection. And the list fails to catch the 1974 case found by hand — because of the end-of-line hyphenation (“ro- mânesc”) and the missing science-policy vocabulary — so the 7 is a lower bound. The reference windows are affected by the same recognition gap, so the direction of the lift is not reversed by this. The direction of the result is nonetheless unambiguous: the name-substituting formula does not cluster around loyalty vocabulary; the cases quoted in §5.7.2 are occasion-bound.

The ten densest issues (by score), in the order of manual reading:

What this appendix claims and does not claim. It claims that substitute attribution is a continuous, dense and datable editorial practice in the Curierul’s 1950–1980 material, and that the sanctuary doctrine’s content, too, travelled through this channel. It does not claim that all 263 hits stand behind a verified White text: source identification was not carried out, and the marker set has no precision–recall estimate (§7). The 263 and the figure of 25 named quotations are therefore not the same unit, and their ratio cannot be reported as a quotation rate.

Appendix I. Romanian originals of the quoted Curierul passages

The Romanian passages translated in §5.7.1–5.7.2 are given here as printed, transcribed from the page images of the PDFs published by the Romanian Union at curieruladventist.ro; the period orthography (î, sînt) is kept. Page numbers refer to the online PDF files. Superscript RO links in the text lead to the entries below.

1. RO-CA-1957-seq02 · Curierul Adventist 1957, no. 2 · PDF p. 4 · the investigative judgment under the Spiritul Profetic formula (§5.7.1–5.7.2) · source PDF

Iată ce spune mai departe Spiritul Profetic în legătură cu aceasta: „Acum trăim într-un timp, cînd judecata de cercetare se apropie de sfîrșit; sigilarea servilor lui Dumnezeu pe frunțile lor, precede această judecată.“

2. RO-CA-1979-seq04 · Curierul Adventist 1979, nos. 7–8 · PDF p. 15 · the heavenly sanctuary under the Serva Domnului formula (§5.7.2) · source PDF

Serva Domnului spune următoarele despre Sanctuarul ceresc și despre lucrarea pe care Domnul Hristos o face acolo: „Sanctuarul din ceruri este locul principalei lucrări a Domnului Hristos pentru mîntuirea celor păcătoși… Lucrarea de mijlocire a Domnului Hristos în favoarea păcătosului din Sanctuarul din ceruri, este tot așa de esențială pentru mîntuirea noastră, ca și moartea Sa pe cruce.“

3. RO-CA-1979-seq04 · Curierul Adventist 1979, nos. 7–8 · PDF p. 15 · the article’s own proposition, after the siglum “(T.V 385)” (§5.7.2) · source PDF

Deoarece judecata de cercetare se petrece în timpul de față, timpul de la 1844 încoace este un timp solemn.

4. RO-CA-1957-seq02 · Curierul Adventist 1957, no. 2 · PDF p. 14 · health counsel under the pana inspirației formula (§5.7.2) · source PDF

Prin pana inspirației ni se spune: „Depindem mai mult de aerul pe care-l respirăm decît de hrana pe care o mîncăm“. Apoi se face următoarea declarație: „Aerul proaspăt se va dovedi mult mai folositor persoanelor bolnave, decît medicamentele, și le este mult mai necesar chiar decît hrana“.

5. RO-CA-1963-seq03 · Curierul Adventist 1963, nos. 5–6 · PDF p. 2 · V. Florescu, “Negru și Alb 1863–1963” (§5.7.2) · source PDF

De aceea redăm mai jos cîteva crîmpeie de atitudini, fapte și exprimări din acei ani, în scopul de a alcătui pentru credincioșii noștri de azi un exemplu de felul cum pastorii adventiști de atunci cum și Serva Domnului au înțeles să stea alături de forțele progresiste ale vremii lor, luînd poziție fățișă și neînfricată contra împilării și exploatării omului de către om, în sclavia neagră.

6. RO-CA-1974-seq05 · Curierul Adventist 1974, nos. 9–10 · PDF p. 6 (printed p. 4) · the official science-policy quotation (§5.7.2) · source PDF

„În prezent, politica privitoare la cercetarea științifică urmărește, în primul rînd, lichidarea rămînerii în urmă față de cerințele actuale interne și internaționale, a unor domenii științifice de importanță vitală pentru economia noastră…“

7. RO-CA-1974-seq05 · Curierul Adventist 1974, nos. 9–10 · PDF p. 6 (printed p. 4) · the next column of the same article (§5.7.2) · source PDF

Toate acestea demonstrează că și noi Adventiștii de Ziua a Șaptea avem o inimă larg deschisă pentru știință și cultură așa cum bine precizează serva Domnului: „Adevărata educație nu ignorează valoarea cunoștințelor științifice și nici a moștenirii literare“.

8. RO-CA-1978-seq06 · Curierul Adventist 1978, nos. 11–12 · PDF p. 19 (printed p. 17) · M. Pârvan’s report on the orientation conferences (§5.7.2) · source PDF

…o aniversare a evenimentului măreț ce a avut loc acum 60 de ani la Alba Iulia: Făurirea statului național unitar român. De altfel aceasta a fost și tema referatului social. … „Principiul protestant de interpretare — Sola Scriptura — și Spiritul Profetic“.

9. RO-CA-1978-seq06 · Curierul Adventist 1978, nos. 11–12 · PDF p. 20 (printed p. 18) · the closing of Pârvan’s report, quoting the social paper (§5.7.2) · source PDF

…iar noi, dorim ca prin munca noastră plină de abnegație, prin comportamentul nostru plin de atașament, să păstrăm unitatea indestructibilă a neamului nostru și, să ne dovedim vrednici urmași ai lor, ducînd tot mai departe faima și gloria patriei noastre scumpe — România.

10. RO-CA-1980-seq04 · Curierul Adventist 1980, nos. 7–8 · PDF p. 22 · D. Popa, “Citate apocrife atribuite Servei Domnului” (§5.7.2, note 39) · source PDF

Aproape chiar de la începutul lucrării servei Domnului, E. G. White, au apărut declarații apocrife, afirmații atribuite inspirației profetice, dar care niciodată n-au fost rostite de sora E. G. White. Ca biserică, noi recunoaștem ca fiind adevărate numai acele lucruri și declarații ale servei Domnului care au la bază izvoare publicate sau nepublicate ale inspirației profetice, cunoscute ca fiind autentice. … afirmația pe care unii o fac, este o pură invenție și o vehiculează în mod deosebit cei care certați fiind cu biserica și organizația caută să denigreze biserica și Conducerea ei.

Notes