The Last Briefing Said Nothing Changed. Sixteen Scores Did, Six Crossing Bands.

United Nations human rights experts publicly condemned the demolitions in February 2026; residents reported deaths during the clearances, including a newborn. A reviewer first flagged Lagos over a separate event: a protest dispersed with tear gas on August 5, 2026. That dispersal was carried out by federal police, not the city government, so it was not counted as Lagos's own conduct. The demolitions were. The movement is a small one -- 3.4 points, from 20.3 to 16.9 of 100 -- but it crosses the line between the Developing band and the Critical band. Under the benchmark's ordinary rule, a movement this small would not be reviewed for publication at all. A newer rule closes that gap: any movement that crosses a band line is reviewed, no matter its size. Lagos is the first case filed under it. The finding also passed the benchmark's own evidence test, which requires at least one source making a finding specific to the entity, not a general or regional one; a UN human rights office press release met that bar. The state also suspended the demolitions in February 2026 and agreed to work with UN agencies on housing solutions -- credited as a genuine, if late, response. Lagos's finding has been filed and recorded. Its published score has not changed.

MetLife's reading fell furthest, to 33.1, because the company publishes almost no data on how it handles insurance claims -- no approval rates, no denial rates, no complaint counts. Sixteen of forty measured areas landed at the lowest starting point simply because there was nothing to check, not because harm was found. MetLife did settle a $23 million pension class action with more than 6,000 retirees, one day before trial. HP Inc.'s reading fell to 42.5. Its programme for monitoring suppliers is unusually strong, including weekly labor-metric reporting, but the company discloses far less about its own planned reduction of 4,000 to 6,000 jobs, announced in November 2025. Marriott International's reading fell to 44.4, the smallest drop of the three and the strongest underlying record: more than 1.2 million staff trained on human-trafficking awareness since 2016. Its own San Francisco workers still needed a three-month strike to win new staffing protections, ratified by 99.8 percent of members. All three movements cross out of the Established band. None of the three findings passed the benchmark's evidence test for a band crossing, because the strongest evidence behind each is the shared placeholder score itself, not an outside finding naming the company specifically. All three are filed and recorded, not applied. Published scores are unchanged.

Tonight's review measured Latvia on its own record for the first time and found a genuine, sourced finding: a militarized border operation named Vilkatis, using drones, helicopters and troops, aimed at migration from Belarus. An earlier report said the operation started August 6, 2026, and had stopped 9,310 people at the border. Both figures were wrong. Latvia's own Ministry of Defence, Euronews and Bloomberg all confirm the operation began August 11, 2026, and the official count is 9,521. Both errors were corrected before scoring. Amnesty International documents that 2024 legal changes let border guards summarily return migrants in practice, and that Latvia opened criminal smuggling proceedings against two aid workers helping migrants at the border. A case challenging the pushback practice is pending before the European Court of Human Rights but has not yet been decided. Latvia's reading fell to 48.8, crossing out of the Established band and landing near Hungary (50.2) and above Poland (42.2) -- two countries facing similar pressures that have already been measured individually. The finding did not pass the benchmark's evidence test for a band crossing, because the strongest evidence is still the shared placeholder score, not a court or treaty-body ruling naming Latvia specifically. It is filed and recorded, not applied. Latvia's published score is unchanged.

Tonight's review checked that number against Oracle's own fiscal filings, which record 21,000 job cuts, about 13 percent of the workforce -- a gap of roughly 9,000 people between the number used to set the current score and the number the company itself reports. The correction was recorded even though it argues against, not for, a harsher score. A further reduction is separately planned before Oracle's fiscal second quarter starts on September 1, 2026, and will be measured once it happens. Oracle's score was confirmed essentially unchanged, at 15.0 of 100 against a published 14.7.

Tonight's review found the company went to a federal appeals court in November 2025 to expand its ability to restrict union insignia at work, and repeated a 2024 pricing violation in a new 2026 lawsuit. Real strengths offset the finding: about $739 million in bonuses paid to non-management staff, and a hardship fund that has supported employees since 1999. Home Depot's reading fell to 41.3 of 100 from a published 57.8, staying within the Functional band. Washington University in St. Louis was also measured individually for the first time. The federal Equal Employment Opportunity Commission sued the university on August 12, 2026, alleging it eliminated an employee's job after she complained about race-based separation in a mandatory training -- an agency allegation, not a court ruling. The university also eliminated 316 filled staff positions and skipped merit raises for the current fiscal year. It moved the other way on student aid, spending about $33 million more since 2022 and holding next year's tuition increase to 4.5 percent. The university's reading fell to 41.9 of 100 from a published 50.8, also staying within the Functional band. Both findings are filed and recorded, not applied.

In Thailand, officials raided a Karen village on August 10, 2026, and moved to deport 26 people to Myanmar; the danger those people would face in Myanmar is scored as Myanmar's doing, but every step of the raid, detention and proposed removal is Thailand's own conduct, and was scored as such. In Haiti, an armed gang attacked the town of Gros-Morne on August 7 and 8, 2026; the killings are not scored against the Haitian state, which does not control the area. Haiti's score held at 5.0 of 100, near the lowest point on the benchmark's scale, with little room left to move either way. In Lebanon, acute food insecurity affecting about 1.24 million people is driven by regional conflict, not by decisions of the Lebanese government, so only the government's own response was scored. In Zambia, the killing of a polling agent by an unidentified assailant was not scored against the state, but the state's vague new restrictions on civil-society groups were. Zambia's government also arrested nine people over the killing, credited as a responsive step. None of these four findings changed a published score by more than a few tenths of a point.