Chile's Score Falls 20 Points — Part of the Evidence Behind It Was Three Years Old, and the Benchmark Caught It

One strike, on 10 August, hit an out-of-session school in the village of Merewa where fighters from the Tigray People's Liberation Front were reportedly sheltering. Whether civilians were killed is disputed between AFP and the BBC, and the benchmark does not score the disputed casualty count in either direction — it scores the decision to strike a school in a populated village. A search summary earlier this cycle had credited Amnesty International with describing these drone strikes. Amnesty's article does not mention drone strikes at all, and is dated 3 August, twenty-four days before the evidence window opened; its forced-conscription findings are attributed by Amnesty to Tigrayan regional authorities, not the federal government. The correction moved the conduct finding off Amnesty and onto Critical Threats, where the evidence actually is. The assessed decline is 2.8 points, from a published 4.7 to 1.9 of 100. Ethiopia's published score was already only 4.7 points above zero, and the benchmark only records a change of 5 points or more. A real, sourced deterioration inside a country's own military conduct could not be logged. Four other entities — Sudan, Israel, the Democratic Republic of the Congo and Haiti — sit in the same position tonight, each with less room to fall than the benchmark's own filing rule requires.

It is a repeat: the CFPB had already ordered Regions in 2015 to refund $49 million and pay a $7.5 million penalty for the same category of overdraft failure, seven years earlier. Regions' published score of 60.9 of 100 was never the product of an individual review. It was a placeholder shared, digit for digit, with three other Fortune 500 entities — Nationwide, Principal Financial and Raymond James Financial — all never assessed and all sitting at consecutive ranks 44 through 48. Those three, carrying no adverse evidence of their own, assess at 25.0, 25.6 and 26.9 once reviewed for the first time — that range is what an honest first look produces from absence of disclosure alone. Regions assesses at 22.5, two to four points lower, and that smaller gap is the real conduct component of tonight's finding. The other three were confirmed unchanged with no proposal filed, because a shared placeholder score is a calibration question, not a one-off conduct finding. Regions was filed anyway, because letting an Established-band label stand against a documented $141 million consumer-harm order would leave a label the record itself contradicts.

It is the third documented protest-policing incident across three consecutive review cycles: abductions and brutality during a 25 June protest, now under investigation by Kenya's Independent Policing Oversight Authority; six activists arrested outside police headquarters on 17 August while demanding the police chief's resignation; and the 28 August tear-gassing. The benchmark's own prior review, on 18 August, committed in writing to escalate to a formal finding if a third incident landed — it has landed. IPOA has been investigating protest-policing brutality since June, and protest policing continued unchanged through two further incidents in the meantime, which is scored as a failure to correct course on evidence Kenya's own oversight body was already examining. Counted in Kenya's favor: IPOA itself is a funded, active oversight body, and the Kenya Revenue Authority publicly defended and explained the contested customs measure. The triggering 28 August incident rests on a single news source, which reports no injury or arrest count; the finding's real weight comes from the three-incident pattern, independently corroborated across four sources.

None of that is scored: a boycott, a labor federation's endorsement and another union's vote are actions taken by other organizations, not by Starbucks itself. Union-asserted counts of unfair-labor-practice violations — reported variously as 550, more than 700, and more than 400 — come from a party to the dispute, contradict each other by more than 150, and could not be verified against a primary National Labor Relations Board record, so none of them was used. What remains, once third-party conduct is excluded, is thinner: another month without a contract, and a company statement restating its bargaining position rather than acknowledging the delay. Counted in Starbucks' favor: a 2026 federal appeals court decision vacated an earlier unfair-labor-practice ruling against the company, and Starbucks continues to provide part-time health coverage and tuition support. The finding is filed on the magnitude of the movement, not as a confident downgrade — the evidence for a 5.3-point fall is real but thin, and a reasonable reading of the same facts could land on a smaller number.

The group itself later clarified the number is "a raw count of approximately 284 million data records, or lines, rather than a count of unique individuals," and told reporters it was "unsure of how many individuals are ultimately affected." McKesson has confirmed no patient count of its own. No patient figure is used anywhere in this review. What is verified: McKesson discovered the intrusion on 25 August and filed publicly with the U.S. Securities and Exchange Commission on 28 August — three days, faster than the law required, since the filing was made under a voluntary disclosure rule rather than the mandatory one for material incidents. In the same filing, McKesson stated it "has not determined that the incident is material." The filing says nothing about notifying the people whose records were affected; it directs stakeholders to the company's website. Fast, voluntary disclosure to investors and no evidenced outreach to patients roughly offset each other, and McKesson's score moves down just 0.7 points, from 31.3 to 30.6 of 100, remaining in the Developing band.

This cycle, the same false claim resurfaced, carried by wire pickups (Manila Times, Cambodianess) that had republished the original story on 26 August, the same day it was retracted, and had not caught up to the correction by the time this review ran. Direct verification against the retraction record confirmed the story remains false, and it was dropped again before it could be published a second time. Peru's score is unchanged from the 26 August review, at 36.2 of 100. A correctly dated news wire can still carry a story that has already been proven wrong — the second time in two consecutive cycles this exact failure pattern has been caught before publication.