Ceuta 2026: Anatomy of an Information War Beyond the Border

Chapter 2 — Reconstruction: Thirty-One Days, Three Peaks

Phase One: the crossing and the flood (30 July – 7 August)

Phase Two: the visible approach of the second wave (8–15 August)

Phase Three: mutation into a Spanish state crisis (16–28 August)

This chapter describes how the analytical corpus was built, validated and enriched, in sufficient detail for the reader to audit every quantitative claim in the report. The methodology is described functionally — what was done to the data — rather than commercially; the investigation's conclusions do not depend on any specific tooling, and equivalent pipelines can be assembled from many components.

The core design decision was to collect by audience, not by outlet. For each of 21 countries, the collection simulated a reader inside that country's information environment: queries were issued in the country's language(s) and native script, through the country-specific search infrastructure a local reader would use, with interface-language and geography parameters matched to that country. This produces, for each node, the mixture of domestic and foreign content a reader there would plausibly encounter — which is the correct object of study for questions about how environments, rather than editors, interpreted the crisis. Twenty-three locale configurations covered the 21 nodes (Morocco and Algeria were each collected in both Arabic and French, a decision that proved decisive for the two-track finding of Chapter 8).

Query design used simple, deliberately neutral two-term keyword families: the native rendering of 'Ceuta' (Ceuta, سبتة, Сеута, 休达/休達, סאוטה, Sebta) paired with broad, non-leading angles — border, crisis, Spain, Morocco, migrants, minors, arrivals, and locally salient toponyms such as Tarajal and Fnideq. Neutrality was a validity requirement: loaded query terms would have pre-selected the narratives the study was designed to measure. Each keyword was issued separately for the two research windows using strict date-range operators (P1: 30 July–15 August; P2: 15–28 August), across both the general web index and the news vertical, with deep pagination (up to four result pages of 80 results per query). Collection jobs were interleaved across all locales and windows so that partial runs could not bias coverage toward any country.

Three further streams complemented the core collection. First, a dark-web collection restricted to genuine dark-web networks (Tor onion services, I2P, IRC, ZeroNet — explicitly excluding clearnet chat platforms), executed in seven languages under a strict fetch budget, sorted newest-first by crawl date: 1,316 unique items were retrieved and screened, of which 133 passed the Ceuta-relevance gate. Second, a manually compiled, source-traced record of extremist, radical and hardline group reactions — 18 records covering designated jihadist organisations, hardline Islamist movements, Muslim Brotherhood-aligned bodies, far-right violent-extremist networks and radical online clusters, each preserving group category, designation status, verbatim claims in the original language, English renderings, narrative frames and exact source URLs. Third, publisher-resolution enrichment: each record's outlet was resolved to a country of origin and outlet type through a curated map of ~170 known outlets, then top-level-domain heuristics, then model-assisted inference, with the resolution method and confidence stored per record.

Every collected link passed through a five-stage pipeline: (1) canonical-link deduplication, guaranteeing that no URL enters the corpus twice; (2) full-content retrieval, converting each page into clean machine-readable text (12,711 of 12,948 retained records scraped successfully; 167 returned empty content and 70 failed, and were analysed from their metadata and snippets where possible); (3) machine-assisted content triage and relevance assessment, in which every item was individually evaluated for whether it concerns the 2026 Ceuta crisis, with a continuous relevance score; (4) automated semantic classification, described below; and (5) incremental, crash-safe persistence with full search provenance (locale, keyword, window, vertical, query, result rank) stored on every record.

The relevance gate was strict and its rejects were preserved for audit. Of 17,794 unique candidate records processed from the core collection, 4,846 (27.2%) were rejected as not genuinely about the crisis — homonym noise, tangential mentions, sports and travel content, and pre-2026 material — leaving 12,948 retained records, of which 12,908 (99.7%) completed full analytical enrichment. The dark-web stream was gated harder: 1,183 of 1,316 items (89.9%) were rejected, reflecting the low signal density of dark-web sources. The final analytical dataset therefore comprises 13,059 validated records (12,908 + 133 + 18).

Table 1. The collection and validation funnel.

Each validated record was individually converted into roughly forty structured analytical fields through automated semantic classification with analyst-defined taxonomies. The enrichment layers were: multilingual semantic normalisation (language detection and cross-lingual normalisation of themes, so that a sovereignty claim in Arabic, a soberanía claim in Spanish, and a 主权 claim in Chinese land in one taxonomy); narrative taxonomy assignment (primary and secondary frames from a fixed frame set, plus free narrative-theme labels); stance measurement toward four reference actors (the Spanish government, the Moroccan government, the EU, and the migrants themselves) on a five-level scale; responsibility attribution (which actors each record blames, and which it praises); contextual entity resolution (people, organisations, locations); quantitative-claim extraction (every migrant-arrival figure and casualty figure cited, enabling the numbers-war analysis of Chapter 22); manipulation assessment (bias, sensationalism and credibility scoring, propaganda-technique tagging, manipulation indicators, and flags for probable disinformation and suspected coordination); terminology tracking (how each record designates Ceuta itself — 'autonomous city', 'Spanish enclave', 'occupied Ceuta', and 30+ observed variants); and analytical metadata enrichment (content type, state-affiliation assessment of the outlet, target-audience characterisation, calls to action, and key verbatim claims).

Classification quality was controlled in four ways. First, taxonomies were fixed before collection, so the classifier could not drift with the news cycle. Second, every record retains its full provenance and its key verbatim claims, so any label can be audited against the underlying text; the analyst team spot-audited labels across languages and both windows during the investigation. Third, all corpus-level findings in this report rest on aggregate distributions over thousands of records, not on individual labels; classification noise at the record level attenuates rather than creates the patterns reported. Fourth, findings were required to survive triangulation: every major judgement is supported by at least two independent measurement routes (for example, the two-track Morocco finding appears independently in terminology shares, stance distributions, frame distributions and the extremist record's documentation of state-religious language).

The enriched corpus was then analysed as a dataset: temporal aggregation at daily resolution with the 15 August boundary as a structural break; node-level aggregation of frames, narratives, stances, blame, terminology, emotions and propaganda techniques, separately for P1 and P2; supply-chain analysis crossing publisher country against audience node; narrative-propagation analysis tracking each key narrative's first appearance and cumulative country reach over time; anomaly hunting over the quantitative-claim distributions (the casualty and arrivals figures); duplicate-claim detection across publisher countries; alignment scoring of every node against six extremist narrative families derived from the extremist-group record; and systematic sampling of high-signal record clusters (coordination-flagged records, disinformation-flagged records, conspiracy-attribution records, early-warning records) for qualitative reading. All figures in this report are generated directly from the corpus; no external statistics are introduced except where explicitly attributed.

Four limits bound the findings. (1) The collection observes what search infrastructure surfaces to a reader; heavily suppressed or deplatformed content is under-represented, which means mobilisation-content findings are floors, not ceilings. (2) Node volumes differ by an order of magnitude; the report never compares raw counts across nodes, only shares within nodes, and marks every low-n node. (3) Publication dates are metadata-derived and normalised to UTC; a small number of records carry detector-inferred dates, and 226 records with unresolvable dates are excluded from temporal charts only. (4) Stance, bias and credibility are model-scored editorial constructs, audited by sampling; they support comparative statements (node A is more hostile than node B; node A hardened after 15 August) rather than absolute ones. Within those limits, the corpus is, to Golden Owl's knowledge, the most complete structured record of the global information response to the 2026 Ceuta crisis assembled to date.

Before asking what each country believed about Ceuta, the investigation asked a harder question: who manufactured what each country believed? Every record carries both an audience node (whose information environment surfaced it) and a publisher country (whose outlet produced it). Crossing the two produces the information supply chain in Figure 4 — and three structural findings that condition everything in Part III.

Spanish publishers are the corpus's largest single source (2,991 records, 23.2%), and their reach is unique: Spanish outlets appear in meaningful volume in every audience node measured, including 322 records surfacing in the Moroccan environment, 27 in Algeria's, and a strong presence across Latin America, where Spanish wire and press content is the backbone of regional coverage. In the first days this gave Madrid a structural narrative advantage: the raw facts reaching most of the world were Spanish-sourced facts — Spanish casualty counts, Spanish arrival estimates, Spanish framing of the emergency response. The investigation's temporal analysis (Part IV) shows how this advantage was squandered: because Spain's own information environment turned inward and adversarial after 15 August, the content Spain exported to the world increasingly carried the 'Spain government failure' frame. A country that supplies the world's coverage of its own crisis also exports its own recriminations.

Wire-service citation analysis (Figure 5) reveals the deeper chokepoint. Four agencies — Reuters (1,457 citing records), EFE (1,232 + 372 Europa Press), AFP (1,182), AP (782 combined) — are cited across every node, and for many nodes they constitute the factual skeleton of the entire story. Latin American coverage is heavily EFE-dependent; Chinese-audience coverage cites Reuters and AFP while adding state-agency layers; Arabic-language coverage mixes AFP with regional agencies. The strategic point is under-appreciated: EFE is a Spanish state-owned agency, and Europa Press is Spanish private. For the Spanish-speaking world — 3,774 records, the corpus's largest language bloc — the 'neutral' factual layer was substantially manufactured in Madrid. This is soft infrastructure power, and it worked: Latin American nodes carry the most Spain-sympathetic humanitarian framing in the corpus. Conversely, nodes with independent agency layers (Russia: TASS/RIA/Interfax; Turkey: Anadolu; China: Xinhua) show the sharpest divergence from the Western factual baseline — most visibly in the casualty figures they chose to foreground (Chapter 22).

The flow matrix exposes a taxonomy of environments. Open importers — the UK, US, Germany, Latin America — consume heavily from foreign publishers, above all Spain and the wires. Self-suppliers with export ambitions — Spain, Morocco, France — generate the bulk of their own coverage and export it. And closed loops: Russia's row and column are nearly identical — 339 of 355 Russian-published records (95.5%) surfaced only in the Russian audience environment, and the Russian audience consumed almost nothing foreign that was not first re-processed by Russian outlets. China shows a softer version of the same structure (262 of 312 Chinese-published records domestic; Hong Kong and Taiwan-published Chinese-language content supplying the rest). The analytic consequence is sharp: whatever Russian and Chinese readers learned about Ceuta, they learned it after domestic re-framing. These environments do not import narratives; they import raw material and manufacture their own. That is what makes their measured frames (Chapters 13–14) attributable to editorial systems rather than to the international news flow.

One further supply anomaly deserves record. Taiwanese-published Chinese-language content (210 records) supplied mainly the Hong Kong node (170 records), where it competed directly with mainland framing; and Belgium's modest output (185 records, anchored by EU-affairs outlets) achieved disproportionate cross-node reach into France, the UK, Morocco and Algeria — the Brussels press corps functioning, as it often does, as the EU's de facto wire service. The corpus also registers publishers from 60+ further countries in small volumes, including Vatican media whose consistent humanitarian counter-framing surfaces in six nodes.

Every record was assigned one primary frame from a fixed taxonomy — the single organising interpretation of the piece — plus free narrative-theme labels normalised across languages. At the global level, the frame distribution looks almost balanced: migration-policy debate 29.0%, humanitarian crisis 28.4%, Spain-government failure 10.4%, security threat/invasion 8.5%, EU weakness/failure 5.7%, geopolitical opportunism 4.3%, sovereignty dispute 3.6%. The balance is an illusion of aggregation. Figure 6 disaggregates by node, and the illusion collapses: no two regions experienced the same crisis.

Four regional signatures stand out. In Latin America (Mexico, Colombia, Argentina, Brazil) the crisis was overwhelmingly a humanitarian story (33–43% humanitarian frame) with strong sympathy for migrants and — the EFE effect — Spanish-supplied facts. In the Arab-world nodes (Egypt, Saudi Arabia, Qatar) the humanitarian frame coexists with the corpus's highest sovereignty-dispute shares (up to 8.3%) and, as Chapter 23 shows, the highest 'occupied Ceuta' terminology rates. In the China-linked and Russian environments the crisis was a policy-failure exhibit: China's 37.9% migration-policy-debate share is the highest in the corpus, and Russia adds the highest security-threat share (16.2%) of any major node. Israel's environment is unique: security-threat/invasion is its number-one frame (29.3%) — the only node where that is true.

Narrative-theme analysis across 12,908 records reveals a two-tier economy. Tier one is universal: 'Spain overwhelmed' (6,878 records; 53–65% of records in most nodes) and 'Morocco weaponizing migration' (5,382) plus 'migrants as victims' (5,269). These are the load-bearing narratives that every environment shared; they carried the factual story. Tier two is strategic and selective: 'Schengen collapse' (2,829 records; 41% of the Chinese node, 37% Hong Kong, 33% Russia — versus 3% in Spain itself); 'Sanchez incompetence' (2,277; concentrated in Spain, Latin America and Italy); 'occupied Ceuta' (1,876; Arab-world and Maghreb); 'Europe is weak' (764; China 14%, Israel 13.8%, Turkey 12%); 'invasion/replacement' (841; Israel 22.4%, triple any other node); 'migrants as threat' (408; Russia 9.2%, first place). Tier-two narratives do not describe the event — they use it. Their distribution is the crisis's strategic map, and nearly every judgement in Part VIII rests on it.

Manipulation scoring places every node on a bias/credibility plane (Figure 8). Three observations. First, the crisis was not, in the aggregate, a disinformation event: mean credibility (0.58) and the 1.9% likely-disinformation rate indicate a mostly conventional — if emotive — news environment. Second, the exceptions cluster exactly where the strategic narratives cluster: Israel's node has the corpus's worst bias (0.486) and near-worst credibility, Russia and China follow, and outlet-level analysis (Chapter 25) shows opposition and state-aligned outlets carrying disinformation rates five to ten times the independent-outlet baseline. Third, emotional-tone analysis shows 'alarm' as the crisis's master emotion in every node — but the secondary emotion differentiates: 'pity' in Latin America and the UK, 'suspicion' in Algeria, Israel and Russia, 'anger' in Spain and Morocco. Alarm plus pity produces humanitarian coverage; alarm plus suspicion produces conspiracy uptake. The secondary-emotion map predicts, with striking accuracy, which nodes later carried the orchestration conspiracies of Chapter 19.

Political exploitation — a record using the crisis to advance an agenda beyond the event — was flagged in 48.3% of all records, an extraordinary global rate that quantifies how instantly Ceuta became a vehicle. The node ranking (Israel 75.0%, Qatar 62.5%, Italy 60.9%, Algeria 54.1%) and its interpretation are developed in Part III; the global point is simpler: for half the world's coverage, Ceuta was never only about Ceuta.

Before turning to the node-by-node profiles, the investigation asks a deceptively simple question that contextual entity resolution makes answerable at scale: when 21 information environments wrote about Ceuta, whose names did they write? Every validated record carries extracted person and organisation entities. Aggregated across 13,059 records, these entities form a map of the crisis's human agenda — who the world treated as protagonist, who as antagonist, and, most revealingly, who it did not mention at all. The result is one of the clearest structural findings of the investigation: a crisis that began in Morocco was narrated, globally, as a story populated almost exclusively by Spanish and European faces.

6.1 The three most-mentioned people in the global corpus are all Spanish domestic political figures. Prime Minister Pedro Sánchez appears in roughly 4,600 records once spelling variants are consolidated — more than a third of the entire validated corpus. Ceuta's regional president Juan Jesús Vivas appears in about 2,780, and Interior Minister Fernando Grande-Marlaska in 2,249. No other individual on earth comes within a factor of three of this trio. The global information environment, in other words, did not merely cover a border emergency: it cast a Spanish political drama, with Sánchez as its lead.

The consequence is a structural accountability asymmetry. A reader in any of the 21 nodes encountered Spanish officials being questioned, blamed, defended and investigated — by name, repeatedly. The Moroccan state, by contrast, appears as an almost faceless abstraction. King Mohammed VI registers roughly 245 mentions across both spelling variants — around five percent of Sánchez's total — and a large share of those occur in Arabic-language religious or ceremonial contexts, including the Mawlid address examined in Chapter 9. Justice Minister Abdellatif Ouahbi (243) and government spokesman Rachid El Khalfi (70) round out an official Moroccan presence that, in aggregate, is smaller than the mention count of Spain's migration minister alone. Whether by media inaccessibility, deliberate reticence or editorial habit, Morocco fought the information war without exposing a single face to it — and the blame data of Part IV shows the pay-off: personalised blame requires a person, and the corpus supplied only Spanish ones.

Table 2. Most-mentioned individuals in the validated corpus (person-entity resolution, spelling variants consolidated).

6.2 Among non-Spanish European leaders, the distribution is lopsided in a way that maps directly onto behaviour. Giorgia Meloni's 765 mentions dwarf Emmanuel Macron's 132 and Friedrich Merz's 128 combined — a direct dividend of Italy's unilateral Schengen suspension in the first days of August, the single most consequential act by any EU government other than Spain's. Denmark's Mette Frederiksen (132) earned her presence through the returns-policy push examined in Chapter 12. The pattern confirms a rule visible throughout this investigation: in a border crisis, the information environment rewards governments that act unilaterally with narrative presence, and renders invisible those that coordinate quietly.

The European Union itself, meanwhile, appears overwhelmingly as institutions rather than people. Frontex (2,212 mentions) and the European Commission (2,092) each outrank every non-Spanish individual on the planet; the Schengen system itself registers 1,084. Ursula von der Leyen's 516 and Magnus Brunner's 358 are respectable but subordinate to their own institutions. This depersonalisation matters for influence analysis: narratives attacking 'Brussels' or 'Frontex' faced no human counter-figure to absorb or answer them, which is precisely the condition under which the EU-weakness meta-narrative of Chapter 5 propagated most efficiently.

6.3 Three lower-frequency entities carry analytical weight far beyond their counts. The Spanish Supreme Court (329 mentions) is the institutional protagonist of the crisis's origin story — the misread ruling dissected in Chapter 21. Spain's National Intelligence Centre, the CNI (181 mentions), anchors the third-peak accountability fight over what Madrid knew and when (Chapter 20). And Brahim Ghali, the Polisario Front leader whose hospitalisation in Spain triggered the 2021 Ceuta crisis, resurfaces in 77 records — almost entirely in Spanish and Algerian publications — as the historical template through which those two environments read Moroccan intent. The Ghali citations are a measurable fingerprint of institutional memory: the nodes that had lived through 2021's weaponised-migration episode reached for it immediately, while nodes without that memory (the Americas, East Asia) never mentioned him at all.

Finally, the organisational margins document the crisis's dual infrastructure. On the humanitarian side, the Red Cross (202) and Save the Children (193) cluster around the unaccompanied-minors story alongside minister Sira Rego. On the informational side, a video-sharing platform widely identified in reporting as the recruitment vector appears as a named organisational actor in 230 records — a rare case of a distribution channel becoming a character in the story it distributed. The corpus, in effect, wrote its own methodological footnote: the platforms that moved the mobilisation content became part of the mobilisation narrative.

Part III profiles each information environment in turn. Two reference charts anchor all of them. Figure 10 gives every node's net stance toward the four reference actors — the Spanish government, the Moroccan government, the EU, and the migrants — on a scale from -1 (uniformly hostile) to +1 (uniformly favourable). Figure 11 gives every node's blame model: the share of records attributing responsibility to each candidate actor. Together they are the crisis's diplomatic weather map.

Three global patterns frame the individual profiles. First, nobody defended Spain: the Spanish government's stance index is negative in all 21 nodes, and 'Spain' is the single most-blamed actor globally (6,296 records) — partly a mechanical effect of Spain being the theatre, but the P1→P2 dynamics of Part IV show it is much more than that. Second, Morocco polarised: its stance index spans from +0.216 (Egypt) to -0.30 (Qatar, low-n) — the widest spread of any actor — making Morocco the crisis's true attitudinal fault line. Third, the migrants were everywhere viewed more favourably than any government: the 'migrants' row is positive in every node except Israel (0.00) — global coverage was, in stance terms, a story about failing states and suffering people. The exceptions and reversals to these patterns are precisely where the intelligence value lies, and the profiles that follow work through them node by node — each closing with the node's position before and after 15 August and its dominant collective conclusion.

8.1 Volume and structure. Spain is the corpus's largest node (1,633 records) and the most domestically supplied (76% of its records from Spanish publishers). Its terminology is institutional — 'autonomous city' in 75.8% of records, by far the highest rate anywhere, with 'occupied Ceuta' near-absent (1.0%). This is a node that never doubted whose city Ceuta is; its war was over whose fault the crisis was.

8.2 Interpretation and blame. Spain's frame mix is unique in combining the three P1 stories in near-equal measure: migration-policy debate (28.1%), Spain-government failure (24.7%) and humanitarian crisis (23.0%). Its blame model is the most self-directed in the corpus: 21.8% of records blame Spain itself and a further 9.8% blame local authorities and 13.2% carry the 'Sanchez incompetence' theme — while blame of Morocco (10.9%) runs below the global average and collapses over time. The node praised no one consistently: the most-praised actors (Ceuta president Juan Jesús Vivas, the Guardia Civil, the Red Cross) each appear in under 4% of praise mentions.

8.3 The 15 August pivot. P1 Spain and P2 Spain are different countries, informationally. In P1, migration-policy debate led (28.1%) with government failure third (19.3%), and 13.0% of records blamed Morocco. In P2, Spain-government failure became the number-one frame (30.6%), blame of Spain rose to 23.6% and blame of Morocco halved to 8.4%; the stance toward the national government deteriorated from -0.176 to -0.243. Morocco effectively disappeared as Spain's villain — displaced by Madrid, the Interior Ministry, the intelligence services (the CNI is blamed in 43 records for failing to anticipate the crossing), and the party system. The corpus records the endpoint: national-security declaration, unified command, a scheduled parliamentary reckoning (3 September) and a two-front political war with both Vox (214 blame records) and the PP (143).

8.4 Assessment. Spain's information environment executed, in thirty days, the full arc that hostile narrators (Chapters 13–14) merely asserted: from emergency solidarity to institutional recrimination to a first-ever national-security invocation. The node's collective conclusion by 28 August: the crisis was a state failure — of anticipation (CNI), of border command, of minors policy, and of political cohesion. For adversarial information systems, Spain's own media manufactured the 'Europe cannot govern its borders' exhibit; Russian and Chinese outlets needed only to quote it.

9.1 Volume and structure. Morocco is the second node (1,354 records) and the second publisher country (1,515 records) — and the only node the investigation collected in two languages, which produced its most consequential single finding. The node's overall posture looks moderate: humanitarian crisis is the top frame (30.3%), migrants are viewed sympathetically (+0.404, among the highest), blame concentrates on Spain (18.6%) and smugglers (17.4%), and the Moroccan government enjoys the corpus's second-best stance toward itself (+0.209) — improving from +0.167 in P1 to +0.262 in P2, even as every other government's stance collapsed. Moroccan security forces are the node's most-praised actor. Blame of Morocco's own government exists (10.7%) but, critically, publisher-level analysis found only 2 records from Moroccan outlets blaming the Moroccan government as an institution — domestic self-criticism was operationally absent.

9.2 The two-track finding. Splitting Moroccan publishers by output language breaks the moderate image in half. Moroccan outlets writing in Arabic (931 records) used 'occupied Ceuta' in 59.9% of their coverage, carried a sovereignty-dispute frame in 12.0%, and blamed Spain at nearly double the rate of their French output. Moroccan outlets writing in French (441 records) — the language of Rabat's European interlocutors and of its diplomatic press — used the term in 20.0%, foregrounded humanitarian and policy frames, and carried the 'Morocco pressure tool' accusation frequently enough to look balanced. Same country, same weeks, same events: a sovereignty narrative for the domestic and Arab-world audience, a responsible-partner narrative for the European one. The gap (39.9 points) is too large, too stable across both windows (Arabic: 58.2% P1 → 60.6% P2) and too aligned with known Moroccan strategic communication practice to be editorial coincidence. The investigation assesses with high confidence that this is deliberate audience segmentation.

9.3 The escalation ladder. The Arabic track did not merely persist; it climbed an identifiable ladder. Step one (30 July–5 August): legacy usage by sovereignty-minded outlets (Hespress alone accounts for 123 'occupied' records). Step two (post-15 August): normalisation — the term's corpus-wide volume nearly doubles in P2 (353→602) while total volume falls, and it spreads from opinion into straight news furniture. Step three (24–25 August): state adoption — the Minister of Endowments and Islamic Affairs invokes 'occupied Ceuta' and the 'liberation of occupied cities' in an address before the King at the Mawlid ceremony, an act Moroccan media itself read as a sovereignty signal, captured in both the mainstream corpus and the extremist-group record (Chapter 27). A vocabulary that in early July belonged to Islamists and irredentist commentators ended August in the mouth of a minister in the royal presence.

9.4 What the node avoided. Equally diagnostic is what Moroccan coverage did not say. It did not report the '15 August' mobilisation as a domestic security embarrassment (coverage foregrounds successful interdiction and arrests — 294 and 111 in cited operations); it did not sustain the casualty controversy (Morocco's official 11 coexists in the node with Spain's 72 mostly without adjudication); and it did not engage the 'Morocco weaponizing migration' accusation directly — the theme appears in 16.7% of node records but overwhelmingly via foreign-sourced content, while domestic outlets counter-programmed with humanitarian and sovereignty frames. The government's improving stance while 70,000+ citizens fled through its border is, in information terms, the crisis's single most successful defensive operation.