Frequently Asked Questions — ARGUS

What is ARGUS? ARGUS is an open protocol for the critical analysis of argumentative texts, designed to be used with artificial intelligence. It organizes the examination of framing, presuppositions, argumentation, sources, omissions, symmetry of standards, and the solidity of conclusions.

How do I use ARGUS? Two methods are available. On a compatible platform, install the ARGUS Skill, then attach the text and ask: “Analyze this text with ARGUS.” The universal method is to download the ARGUS V5.0.0 protocol, attach or paste it together with the text, and ask: “Analyze this text using the ARGUS V5.0.0 protocol.” For several related texts: “Analyze these texts as a corpus with ARGUS.”

Does the language of the protocol have to match the language of the analyzed text? No. The language of the protocol, the language of the analyzed text, and the language of the output are independent. For example, the French version of ARGUS can be used to analyze a Japanese text and produce an English analysis, provided that the AI system can process the languages concerned reliably enough. The choice of ARGUS language version is primarily intended to make the protocol intelligible to its human reader; it does not define the languages of texts that can be analyzed.

Why is ARGUS published in several languages? To make the protocol itself readable, understandable, and auditable by as many people as possible. The translations are not different analytical variants and do not restrict the languages of texts that can be analyzed. Additional presentation languages may be added if demand warrants it.

What is the ARGUS Skill? The ARGUS Skill is an execution layer that installs once the instructions needed to apply the protocol on a compatible platform. Depending on the service and interface language, the feature may be called a “Skill” or a translated equivalent. Stable Skill 0.4.2 executes ARGUS V5.0.0 and can return the analysis in the requested language.

Does the Skill replace the protocol? No. The Skill makes execution easier and bundles a copy of the protocol together with technical output rules. ARGUS V5.0.0 remains the normative reference. The Skill version and the protocol version are distinct: a technical Skill update is not a new version of ARGUS.

Is ARGUS only for having texts analyzed by an AI? No. It is also a learning tool, and that is its most durable function. A reader who has conducted around ten full analyses eventually recognizes a framing device, an immunizing concession, or a circularity of sources without needing the protocol. ARGUS therefore recommends the dry run: before submitting the text to the AI, answer three questions yourself in a few lines — what presuppositions does the opening install, what is missing, does the text apply to itself what it demands of others — then compare your answers with the analysis produced.

Does ARGUS work with every AI? Not necessarily in its full version. ARGUS is independent of any particular model, but its execution depends on the effective capabilities of the environment being used. An AI may be able to comment on a text intelligently while still failing to maintain all the constraints of a long protocol, read several documents correctly, or complete every required step.

What minimum capacity is required to run ARGUS? There is no universal threshold that can be expressed as a single number of tokens. A full execution requires four capabilities: keeping the protocol, the documents, and the state of the analysis available at the same time; having enough output capacity to finish the work; reading the supplied files faithfully; and maintaining the protocol’s rules and conditions until the end. A large context window advertised by a provider does not, by itself, guarantee a compliant execution.

Why does my AI stop before the end of an analysis? Several limits may be involved: maximum output length, the amount of context actually available, compression or loss of the previous state, imperfect handling of attachments, or difficulty maintaining a large number of constraints across a long response. These limits may depend as much on the interface or subscription tier as on the underlying model.

Is an interrupted analysis that is then resumed always a complete ARGUS analysis? No. A simple continuation may be enough if the AI genuinely retains the entire protocol, the documents, and the state of the analysis. But if part of the context has been lost, inadequately summarized, or replaced, resuming does not automatically restore the unity of the execution. An analysis missing mandatory steps, or continued after context loss, must not be presented as a complete ARGUS analysis.

What should I do if the environment cannot run ARGUS in full? You can use ARGUS Light, reduce the number or length of the documents, analyze separately several texts that do not require corpus treatment, or change environments. It is better to declare a more limited execution than to present an incomplete one as complete.

What is ARGUS Light? It is the short version of the protocol. In ARGUS V5.0.0, it contains 12 operations. It retains the most important controls but does not apply every advanced step: in particular 0.d, 0.e, 0.f, 3.I, 3.J, 3.K and 7.c, as well as Appendices 2 and 3. From Appendix 4, only tests A and G are retained. The integrity check, scope decision, and absolute rules continue to apply to the operations that remain.

What is the difference between full ARGUS and ARGUS Light? ARGUS Light reduces the depth and length of the analysis; it does not replace the preliminary decisions that establish whether the object is complete, relevant, and correctly delimited. It is suitable for quick readings or more constrained environments. Full ARGUS remains preferable when an analysis is to be published, audited, or compared with other analyses.

What is Appendix 2 (empty signifiers)? It is a conditional appendix designed to identify decisive terms whose definition or boundary can change the conclusion. It may be requested explicitly, but the AI must also activate it on its own when the protocol’s conditions are met, after reconstructing the central thesis.

What is Appendix 3 (ARGUS-Corpus)? It is the procedure dedicated to sets of related texts. It activates before the analysis of the first text. It organizes the declaration of the corpus, the integrity check and relevance triage, the individual analysis of each text, intertextual consistency, second-order circularities, and the separation between synthesis grounded in the corpus and interpretive input from outside it.

Can I analyze several texts that do not need to form a corpus? Yes. When the texts are independent or merely intended for comparison, it is often better to analyze them separately and then request a comparison. Appendix 3 is useful when the texts genuinely form a set whose relationships, repetitions, divergences, or shared sources need to be analyzed in their own right.

Does an article accompanied by its primary source constitute a corpus? Not automatically. A relevant primary document may serve as a control source under 3.B without activating ARGUS-Corpus. The primary-source fidelity test activates only if the relationship between the analyzed text and that source is actually established. Mere proximity of topic, date, or figures is not enough.

What is Appendix 4 (arithmetical and statistical control)? It checks figures, rates, proportions, counts, or projections that the text uses as evidential support. It examines not only where they come from but also whether they are possible and how they are constructed: order of magnitude, denominator, measure of central tendency, base rate, measurement versus model, false precision, and reproducibility of counts. In V5.0.0, the appendix activates as soon as a figure is evidentially material within the meaning of the protocol, even if the conclusion could survive without it.

Does ARGUS verify facts outside the text? To a limited extent, yes, and such verification is mandatory when the protocol requires it. An external verification is external to the analyzed object, but it need not be found on the Web: a source supplied by the user may serve as a control. The analysis must identify the source actually consulted in a form that allows it to be found again, together with its date. Knowledge merely remembered by the model, even if presented as verified, is not an external verification.

Does the AI need Web access to use ARGUS? Not for every analysis. ARGUS can operate in a closed regime when the object and the necessary sources are supplied. Some external verifications or contemporary facts do require access to outside sources, however. If Web access is unavailable, the AI must state that limitation and must not replace the missing verification with its internal memory.

What does an “unsuccessful” verification mean? It means that a reasonable control was attempted but did not sufficiently confirm or disconfirm the proposition. The absence of a result is not evidence that the claim is false. The protocol distinguishes completed, partial, and unsuccessful checks.

What about a book, or a report of which only one part is argumentative? The protocol does not automatically treat it as a single block. When some parts are argumentative and others are didactic, technical, or documentary, the AI must declare a scope and state how the excluded sections are handled. It may not mechanically transfer a flaw found outside the scope to the analyzed thesis, except in the specific cases allowed by the protocol.

What happens if the supplied document is incomplete? ARGUS begins with an integrity check. If the document is manifestly truncated, the analysis must stop and alert the user. If the user confirms that the analysis should continue, the incompleteness becomes a declared limitation: some conclusions about omissions, the gap between the announced program and the actual content, and the 7.c level then become qualified or non-attributable according to the protocol. A deliberate excerpt supplied by the user is not treated as truncation: it constitutes a declared scope.

Why can an AI mistakenly think a PDF is truncated? A PDF’s text layer may be disordered: words or quotations may be displaced, columns may be read in the wrong order, or fragments may be extracted elsewhere on the page. The AI must distinguish these conversion artifacts from a genuine material interruption of the document. This is one reason documentary fidelity is part of ARGUS’s execution requirements.

How can I check that the analysis is reliable? Reread it critically. At minimum, check the integrity of the object, the presence of the mandatory steps, scope decisions, activation of appendices when their conditions are met, consistency of the final judgment, and the distinction between what comes from the text and what comes from outside sources. Step 8 requires the AI to conduct a self-critique, but that self-critique is itself produced by the AI and does not make it infallible.

Is the analysis 100% reliable? No. The AI remains an auxiliary. ARGUS makes its operations more explicit and more controllable; it does not turn a language model into an infallible measuring instrument.

Why do two AIs give different results? Because an ARGUS analysis is not a measurement. Two different models, or sometimes the same model in two executions, may differ in how they qualify a flaw, its degree of severity, or the degree of interpretive closure. That divergence is information: convergence points are relatively robust; divergences indicate where the user’s judgment is required.

How can I challenge or audit an ARGUS analysis? The protocol provides three levels. Level 1 is a short counter-test in a fresh conversation on steps 1, 3.E, 3.H and 7.a. Level 2 is a blind cross-audit by a second AI followed by a differential comparison. Level 3 is a documented human adjudication, suited to published analyses. No level of challenge is mandatory.

What does level 7.c mean? V5.0.0 uses a five-level scale, from 0 to 4, measuring the degree of interpretive closure in the text. “Oriented information” is distinguished from advocacy, and the term “propaganda” is reserved for the higher levels. The level is not a scientific measurement and must be justified by the thresholds defined in the protocol.

Does ARGUS attack authors? No. ARGUS analyzes texts, not people. It distinguishes observable effects of textual construction from psychological intentions that cannot be established. Step 7.b also requires the analysis to credit explicitly what the text establishes solidly.

What happens if the text does not provide enough evidence? The analysis must say so. A section that cannot be completed reliably receives an explicit statement such as “insufficient data to conclude,” naming the missing element. Acknowledging a limitation does not discredit the analysis. This clause does not cover the analyst’s indecision: if what is missing can be named, it applies; if the analyst can only say that the judgment is difficult, the problem lies in the analysis.

Does ARGUS invent anything? ARGUS belongs to a tradition of intellectual self-defense, propaganda analysis, rhetorical criticism, and methodological skepticism that it does not claim to have founded. The protocol’s “Lineage and original contribution” section states what it inherits and what it adds: notably a genre-indexed standard of proof, the notion of sophisticated propaganda, the obligation to credit what a text establishes, corpus analysis, explicit verification rules, and analyst self-critique.

Where can I find the latest version? The latest protocol version is available in the Versions section and in the site downloads. The stable Skill version is shown on the ARGUS Skill page. A published analysis must always state the full ARGUS version number used; when produced through the Skill, the header also states the wrapper revision.