Methodology

Content is annotated at the item level (article, post, video) along seven independent layers — faceted annotation instead of a single flat label. The verifiable (veracity) is separated from the inferred (intent). Every taxonomy value has a citable academic source; this page is generated directly from the canonical JSON Schema (v0.2) and codebook (v0.3), so it cannot drift away from the research canon.

The canonical wording of the schema and codebook is Slovak; the English descriptions and value definitions below are translations kept in sync with the canon by an automated drift guard.

L0 — Provenance and metadata

Filled in by the collection pipeline, not by the annotator.

field	type	values
item_id	text	free text / structured
url	text	free text / structured
archive_url	text	free text / structured
platform	single choice	facebook x telegram youtube web ine
media_type	single choice	text image video audio multimodal
source_id pseudonymised (hashed) for ordinary users	text	free text / structured
published_at	text	free text / structured
collected_at	text	free text / structured
language_variety	single choice	sk cz sk-dialect mixed ine
reach_snapshot	object	free text / structured

L1 — Veracity / type of information disorder

Wardle & Derakhshan (2017): Information Disorder. Council of Europe.

field	type	values
disorder_type disinformation requires falsity (L1) AND an indicator of intent/harm (L5/L6); without it, misinformation.	single choice	disinformation misinformation malinformation not_applicable
veracity If a claim cannot be verified from the text alone, use 'unverifiable' with low confidence. Never infer falsity from disagreement.	single choice	true mostly_true misleading false fabricated unverifiable opinion_not_factual
content_type Wardle's 7 types + not_applicable (ADR-028: the counterpart of disorder_type=not_applicable).	single choice	satire_parody false_connection misleading_content false_context imposter_content manipulated_content fabricated_content not_applicable
check_worthy Does it contain a verifiable factual claim? (filters out pure opinion)	yes/no	free text / structured

L2 — Narrative and framing

EUvsDisinfo (EEAS) — 6 macro-narratives · Card et al. (2015): Media Frames Corpus — 14 frames.

field	type	values
meta_narrative The six EUvsDisinfo macro-templates. Empty if none fits.	multi-label	elites_vs_people — Elites vs. the people — corrupt “elites” detached from “ordinary people”; secret elites; electoral fraud threatened_values — Threatened values — the decadent West; traditional values under attack from liberalism/LGBTI/feminism; Russia as the guardian of decency lost_sovereignty — Lost sovereignty — countries are no longer “truly sovereign”; ruled by the USA/Brussels; the “colour revolutions” trope imminent_collapse — Imminent collapse — the West/EU/Ukraine will soon collapse economically/politically/morally hahaganda — Hahaganda — ridicule used to discredit both the target and democracy itself nazism_accusations — Nazism accusations — labelling opponents (especially Ukraine) as “Nazis/fascists”
frame Card et al. 2015 — the 14 MFC frames.	multi-label	Economic Capacity_and_resources Morality Fairness_and_equality Legality_Constitutionality Policy_prescription Crime_and_punishment Security_and_defense Health_and_safety Quality_of_life Cultural_identity Public_opinion Political External_regulation_and_reputation
primary_topic Slovak-localised seed set + emergent codes (other).	multi-label	war_ukraine — War in Ukraine — the West/Ukraine is to blame; “the war began in 2014 in Donbas”; Ukraine = a fascist state; “Russophobia” — nazism_accusations, lost_sovereignty anti_nato — Against NATO — NATO is aggressive and drags us into war; a US instrument; enlargement = betrayal of Russia — lost_sovereignty, imminent_collapse anti_eu — Against the EU — the “Brussels diktat”; nonsensical regulations; the EU run from Washington — lost_sovereignty, elites_vs_people sanctions_energy_economy — Sanctions / energy — sanctions hurt Europe more than Russia; expensive energy — imminent_collapse sovereignty_foreign_policy — Sovereignty / foreign policy — integration made Slovakia a “servant of foreign interests” — lost_sovereignty traditional_values — Traditional values — Russia as the “big brother” and protector of tradition — threatened_values gender_lgbti — Gender / LGBTI — “gender ideology”; a threat to the family — threatened_values migration — Migration — migrants as a threat to identity and values — threatened_values health_antivax_covid — Health / antivax / COVID — the “infodemic”; bio-laboratories; vaccine conspiracies — elites_vs_people elections_delegitimization — Delegitimisation of elections — electoral fraud; questioning the results — elites_vs_people, hahaganda distrust_institutions — Distrust of institutions — distrust of the state/police (e.g. after the 2024 assassination attempt on Fico) — elites_vs_people anti_media — Against the media — scepticism towards independent media; “the mainstream lies” — elites_vs_people, hahaganda anti_americanism — Anti-Americanism — the USA as the real hegemon/aggressor — lost_sovereignty globalist_elite_conspiracy — Globalist conspiracy — Soros; the “Great Reset”; secret elites — elites_vs_people other
sub_narrative	text	free text / structured
narrative_summary One sentence in the annotator's own words.	text	free text / structured
euvsdisinfo_xwalk	text	free text / structured

L3 — Manipulation techniques

SemEval-2023 Task 3 (Piskorski et al., 2023) — 23 persuasion techniques · Da San Martino et al. (2019).

field	type	values
techniques Empty if no technique is present.	multi-label	Name_Calling-Labeling — Name Calling / Labeling — labelling the object with a term the audience hates or adores — Name-calling, Labeling, Demonizing the enemy, Stereotyping Doubt — Doubt — questioning the credibility of a person/institution without evidence — (part of FUD) Guilt_by_Association — Guilt by Association — discrediting through association with a hated person/group — Guilt by association Appeal_to_Hypocrisy — Appeal to Hypocrisy — “you did it too” — attacking the opponent's consistency — (absent in DIMARL) Questioning_the_Reputation — Questioning the Reputation — attacking reputation/credentials rather than the argument — Smears, Ad hominem, False accusations Appeal_to_Authority — Appeal to Authority — the claim is true because an authority says so — Appeal to authority, Testimonial, Third party technique Appeal_to_Popularity — Appeal to Popularity — “everyone does it” / the majority supports it — Bandwagon Appeal_to_Values — Appeal to Values — an appeal to shared values (freedom, homeland…) — Glittering generalities, Virtue words Appeal_to_Fear-Prejudice — Appeal to Fear / Prejudice — evoking fear/prejudice instead of an argument — Appeal to fear, Appeal to prejudice Flag_Waving — Flag Waving — an appeal to patriotism / group identity — Flag-waving Straw_Man — Straw Man — distorting the opponent's position into an easier target — Straw man Red_Herring — Red Herring — diverting attention to an irrelevant topic — Red herring Whataboutism — Whataboutism — a counter-accusation instead of an answer — Whataboutism Causal_Oversimplification — Causal Oversimplification — a single cause for a complex phenomenon — Oversimplification False_Dilemma-No_Choice — False Dilemma / No Choice — only two options, as if no others existed — Black-and-white fallacy Consequential_Oversimplification — Consequential Oversimplification — “if A, the chain B, C, D inevitably follows” (slippery slope) — (absent in DIMARL) Slogans — Slogans — a short, punchy catchphrase instead of an argument — Slogans Appeal_to_Time — Appeal to Time — “now or never” — urgency pressure — (absent in DIMARL) Conversation_Killer — Conversation Killer — a phrase that shuts down the discussion — Thought-terminating cliché Loaded_Language — Loaded Language — emotionally charged words — Loaded language, Dysphemism, Euphemism Repetition — Repetition — repeating the same message over and over — Repetition / Ad nauseam Exaggeration-Minimisation — Exaggeration / Minimisation — inflating or downplaying — Exaggeration, Minimisation Obfuscation-Vagueness-Confusion — Obfuscation – Vagueness / Confusion — deliberate vagueness/confusion — Obfuscation, Intentional vagueness Scapegoating

L4 — Actor and behaviour

DISARM Framework (T-codes).

field	type	values
source_type ADR-028: added mainstream_media (a legitimate/mainstream outlet) and commentator_influencer (an individual commentator/influencer with no political affiliation). Rejected: alternative_media (an evaluative, not a structural category).	single choice	state_affiliated known_disinfo_outlet political_actor influencer anonymous suspected_inauthentic ordinary_user mainstream_media commentator_influencer unknown
origin	single choice	domestic foreign unknown
coordination_signals	single choice	yes no unknown
disarm_ttp DISARM T-codes (actor/environment behaviour). Only when behaviour is indicated.	list	free text / structured
linked_entities Pseudonymised for ordinary users.	list	free text / structured

L5 — Target and effect

Nimmo (2015): 4D — dismiss, distort, distract, dismay · DISARM.

field	type	values
target_entity	single choice	institution EU NATO ukraine government specific_person science_experts minority_group election other
intended_effect Nimmo 4D = DISARM T0075–T0078, + not_applicable (ADR-028). NEVER null: null cannot distinguish 'does not apply' from 'the model did not know'.	single choice	dismiss distort distract dismay not_applicable
communicative_function	multi-label	discredit mobilize demobilize monetize polarize emotional_arousal

L6 — Intent (inferred)

Inferred layer — intent; annotated with low confidence, never the sole basis for classification.

field	type	values
assessed_intent_to_harm	single choice	yes no unknown
intent_confidence	single choice	low medium high
intent_rationale	text	free text / structured

L7 — Annotation process

Confidence, adjudication, audit trail.

field	type	values
annotator_id e.g. 'A1', 'A2' or 'model:claude-...'	text	free text / structured
annotated_at	text	free text / structured
field_confidence Map of field -> confidence for the key fields (at least L1.veracity, L3, L5, L6).	object	free text / structured
notes	text	free text / structured
adjudication_status	single choice	pending resolved
gold_label	yes/no	free text / structured

Reliability measurement

The gold corpus is built by independent annotation (coders cannot see each other's answers; a submitted annotation is immutable). Reliability is measured with Krippendorff's α per layer and per label; inter-annotator agreement forms the empirical ceiling against which the AI classifier is evaluated — not an arbitrary fixed percentage.