What explains the spikes and dips?

What happened. The January 2025 aid pause and stop-work sequence appears in Ukrainian reporting about threatened programs and media funding. In Liberia, February reports describe interrupted services, project cancellations and demands for a domestic response.

What the articles show. Earlier grant and project-launch stories give way to disruption reporting. But favorable wording does not always praise Washington: a Liberian editorial’s call for self-reliance received a positive label, while Ukrainian stories distinguish development assistance from military aid. The month combines different policy stages and different targets of evaluation.

Corrected attention change: Ukraine +0.221 pp; Liberia +1.545 pp. Share of collected all-news volume; a different outcome from tone.

What survives. The corrected declines remain -51.1 points in Ukraine and -92.9 in Liberia, with the same direction under the count and outlet checks. Collected-news attention rises in both. Liberia’s preceding-three-month comparison already includes January’s pause, so it is not a clean pre-policy baseline.

E34: full article-context audit · E37: full article-context audit · Corrected results and outlet checks

What happened. The Biden–Zelensky summit is followed by coverage of financing commitments, pandemic assistance and the launch of the Hoverla local-governance project.

What the articles show. Several outlets report pieces of the same bilateral package. Some stories concern proposed financing or Ukrainian development ambitions; others describe U.S. assistance. The baseline also contains aid news about Afghanistan and Belarus. Publication in Ukraine does not make every article about aid to Ukraine.

Corrected attention change: Ukraine +0.255 pp. Share of collected all-news volume; a different outcome from tone.

What survives. The corrected rise is +53.0 points; equal weighting of the shared outlets gives +56.9. The direction survives the specified checks. The articles support a shift toward commitment and cooperation coverage, without showing that all announced benefits were delivered or isolating one program’s contribution.

E39: full article-context audit · Corrected results and outlet checks

What happened. Pandemic assistance and the Sardoba dam disaster produce several aid announcements and a reported relief handover. Outlets often relay embassy material about the same actions.

What the articles show. The topic also contains global WHO-funding news, private aid and stories about other Central Asian countries. These are different donors, recipients and stages, even when the model assigns a favorable aid label. No single independently verified U.S. aid onset accounts for the whole month.

Corrected attention change: Uzbekistan +0.188 pp. Share of collected all-news volume; a different outcome from tone.

What survives. The article-pooled corrected increase is +16.5 points, but the equal-shared-outlet increase is only +1.8. It does not pass all count and outlet-direction checks. The favorable reporting is visible; its aggregate magnitude depends substantially on which outlets supply the articles.

E40: full article-context audit · Corrected results and outlet checks

What happened. July coverage centers on the U.S. award to imprisoned rights defender Azimjan Askarov and the Kyrgyz government’s move to terminate a cooperation agreement. October brings a new ambassador, statements welcoming cooperation and John Kerry’s visit and embassy opening.

What the articles show. July’s negative labels mix criticism of Washington with criticism of the Kyrgyz government for breaking cooperation. October’s favorable language often comes from quoted officials. Its three-month baseline includes the July dispute, so the large rise includes a rebound from confrontation.

Corrected attention change: Kyrgyzstan +0.691 pp; Kyrgyzstan +0.039 pp. Share of collected all-news volume; a different outcome from tone.

What survives. The corrected changes are -53.3 points in July and +59.9 in October; both directions survive the specified checks. The paired episodes show why the event sequence, baseline and quoted speaker matter more than treating each monthly number as an independent national attitude.

E41: full article-context audit · E43: full article-context audit · Corrected results and outlet checks

What happened. Ruto’s U.S. visit brings reporting on education agreements, investment and digital opportunities, and an announced security relationship.

What the articles show. The reviewed stories repeatedly carry official descriptions of cooperation and expected benefits. Other reporting raises questions about the Haiti deployment and domestic legal accountability. Economic, education and security announcements therefore appear together inside the foreign-policy category.

Corrected attention change: Kenya +1.063 pp. Share of collected all-news volume; a different outcome from tone.

What survives. The corrected rise is +43.1 points across 4 shared outlets, and the direction survives the specified checks. This is favorable visibility for announcements and official claims; the articles do not establish that investment materialized or that citizens benefited.

E44: full article-context audit · Corrected results and outlet checks

What happened. The reviewed month contains direct anti-U.S. statements by Belarusian officials alongside opposition prosecutions, migration disputes, Ukraine and Afghanistan reporting.

What the articles show. The mix of outlets changes sharply: sb.by supplies a larger share, while Nasha Niva’s share falls. Documented repression makes continuity a concrete concern, but it does not establish why each collected count changes. A government statement quoted in another outlet is also not automatically that outlet’s endorsement.

Corrected attention change: Belarus -0.197 pp. Share of collected all-news volume; a different outcome from tone.

What survives. The corrected decline of -43.6 points retains its direction under the specified checks. That numerical stability does not identify a single September U.S. policy or establish that the observed sources represent an unchanged media environment. The event-onset review therefore leaves this case without a single-event interrupted-series estimate.

E42: full article-context audit · Corrected results and outlet checks

What happened. The Senate’s December 9 CIA report release generates accounts of earlier interrogation practices, condemnations, official defenses and arguments about accountability. Turkish coverage also asks about domestic cooperation with rendition.

What the articles show. Egyptian stories contain U.S., German and other voices, plus separate military-aid news. Turkish records reveal additional measurement problems: Ak Saray was translated as the U.S. White House, and an article alleging torture received a positive label. A present-day disclosure about past conduct does not measure only the incumbent administration.

Corrected attention change: Egypt -0.264 pp; Turkey -0.332 pp. Share of collected all-news volume; a different outcome from tone.

What survives. Turkey’s corrected decline is -38.9 points and retains its direction under the checks. Egypt’s pooled decline is -21.2, but equal weighting of its 2 shared outlets yields +32.2. The common news event is clear; a common country-level tonal response is not. Unreconciled Turkish snapshot differences remain flagged.

E36: full article-context audit · E46: full article-context audit · Corrected results and outlet checks

What happened. Reporting follows the U.S. killing of Qassem Soleimani and subsequent disputes about escalation, legality and the evidence for an imminent threat.

What the articles show. Articles juxtapose U.S. justifications, Iranian condemnation and congressional criticism. The reviewed material includes wire and international-broadcaster attribution, while other defense stories remain in the monthly pool. Publication in Serbia is not a measure of the Serbian government’s position.

Corrected attention change: Serbia +0.745 pp. Share of collected all-news volume; a different outcome from tone.

What survives. The corrected pooled decline is -44.1 points. Equal-shared-outlet weighting gives -10.9, so the magnitude is much smaller even though the direction passes the specified checks. Individual outlet responses differ; the event context is stronger than a claim of uniform editorial reaction.

E45: full article-context audit · Corrected results and outlet checks

What happened. Negotiations and the April 28 EDCA signature coincide with Obama’s visit and public assurances about the alliance.

What the articles show. Officials describe modernization and cooperation, while Bayan argues that past U.S. presence failed to modernize the armed forces. The same national media sample carries both accounts. A negative-labeled baseline search-and-rescue report also illustrates how an adverse event can be confused with criticism of U.S. conduct.

Corrected attention change: Philippines +0.732 pp. Share of collected all-news volume; a different outcome from tone.

What survives. The corrected rise is +40.9 points, with +46.6 under equal-shared-outlet weighting. Its direction survives the specified checks. The article context supports a shift toward agreement and assurance coverage, not unanimous approval or evidence that promised security benefits were achieved.

E47: full article-context audit · Corrected results and outlet checks

What happened. The only focal-month record in the original aid series reports a dispute between France and Mali over aid and French-funded NGOs.

What the articles show. The later correction ledger checks the full available French body and finds no U.S. government actor. The favorable baseline contains repeated U.S. food-relief and cooperation announcements. Comparing those stories with the France-related article creates an apparent U.S.-aid collapse.

Corrected attention change: Mali -0.148 pp. Share of collected all-news volume; a different outcome from tone.

What survives. The original change was -190.0 points, entirely dependent on one focal article. After the documented relevance exclusion, the focal U.S.-aid count is zero and tone is undefined. The corrected result is not a zero change and cannot support a U.S.-aid sentiment-collapse interpretation.

E25: full article-context audit · Corrected results and outlet checks