The most significant editorial findings in the Jul 27 briefing.
Editorial insight
Twenty governments, cities, and companies were reviewed overnight, the fourth and final night of a four-night catch-up review. All twenty scores held within the benchmark's five-point confirmation range, and no bands changed.
A score moving up this cycle turned out to be an arithmetic artifact, not a conduct finding, raising the question of what other stored composites may not match their own dimensions.
“GEO Group's eight dimension scores reconstruct to a canonical composite of 12.5, not its published 6.6, a gap the benchmark's math-hygiene check has flagged for correction. Will the correction resolve as a pure data-integrity fix, or will it also change GEO Group's position within the Critical band?”
GEO Group's published score is 6.6 out of 100. Tonight's review found 11.9, a movement of plus 5.3, which numerically crosses the benchmark's threshold for a formal score-change proposal.
Why it matters
A number moving up does not always mean an institution got better. GEO Group's own eight published dimension scores add up to 12.5 out of 100, not its published 6.6, so the benchmark treated the gap as a data error, not a finding of progress.
Whether the WARN Act compliance investigations become filed complaints, which would supply the stronger source the current record lacks, is the next event to watch.
A coordinator-level comparison of Baxter's proposed reading against the Fortune-500 medical-device placeholder cohort, or a second Baxter recall, is the next event that would test this finding.
Any Ecuadorian or US government response to Human Rights Watch's inquiries, or a first prosecution arising from the documented incidents, is the next event to watch.
How to read this briefing— Bands, scores, and terms
Schema guide
The 5 performance bands
critical0–20
developing20–40
functional40–60
established60–80
exemplary80–100
Two scales
Each of the 8 dimensions is scored 1.0–5.0; these combine into a 0–100 composite score, mapped to the 5 bands above.
Key terms
Band crossing
A score change large enough to move an entity from one performance band into an adjacent one — the most structurally significant finding in a given cycle.
Boundary watch
An entity whose current score is within 3 points of a band threshold, flagged for priority reassessment in the next cycle.
Carry-forward
A dimensional credit retained from a prior assessment when new evidence is insufficient to revise a specific dimension; disclosed explicitly.
First baseline
An entity's inaugural composite score — no published score exists to compare against, so delta is not shown.
Floor designation
The most serious finding: all 8 dimensions resolve at the lowest behavioral anchor (1.0/5.0) across multiple cycles, yielding a composite of 0.
Forward trigger
A scheduled future reassessment event (e.g., a policy implementation date or legislative deadline) that may materially change an entity's score.
Delhi's police crackdown was serious enough that a large score drop would have been easy to justify. Instead, the review was anchored to the last assessment so the number reflects new evidence, not a different reviewer's judgment call.
Twenty entities were reviewed tonight and all twenty confirmed within the benchmark's five-point range, with zero score changes and zero band changes. Delhi is the clearest example of why that happened by design, not by accident.
Where this sits
Read the full signal
Human Rights Watch found that on July 20, 2026, security forces used tear gas and batons against a peaceful march on Parliament and shut down mobile internet across central Delhi, verified through nine witness interviews and ten videos. A hunger-striking activist was forcibly hospitalized on July 18. Reviewed from scratch, that evidence could plausibly support a score roughly six and a half points lower than published. Instead, the review was anchored to Delhi's own assessment from one week earlier and moved only the parts of the score with direct new evidence behind them, producing a smaller, four-point-seven-point drop. The gap between those two numbers is not new conduct: it is the normal variation between two reviewers looking at the same city, and the benchmark's rule is to use the anchored, more conservative number rather than let re-measurement alone move a score.
A federal finding from June was almost scored as July conduct, for the fourth night in a row. The scanner cannot always tell an old event from a new one just by looking at a web address, so a second read keeps catching it first.
Four items were corrected or dropped tonight on date-verification grounds, extending a pattern now seen on four consecutive nights (five items July 24, two items July 25, six items July 26, four items July 27).
Where this sits
Read the full signal
UnitedHealth Group: a federal finding that UnitedHealthcare denied 71 percent of long-term acute-care hospital admission requests was published by the HHS Office of Inspector General on June 8, 2026, not in July as an article had implied; the true in-window event is a mid-July Senate records demand, which is what was actually scored. GEO Group: a Cal/OSHA workplace-safety settlement reported on July 7 was actually signed in May 2026, both before the review window opened; dropped as background. Atlanta: a widely reported one-billion-dollar detention-warehouse plan abandonment turned out to be a June 2026 event in Social Circle, Georgia, a separate town of about 5,000 people, not the City of Atlanta; dropped on both date and location grounds. Afghanistan: reported figures of roughly 80 injured and more than 100 missing from flash floods could not be tied to any dated statement and were not scored; the United Nations' verified count of 26 killed and nine injured was used instead. None of the four reached a published score.
Floods, a bank's lending decision, and marches in cities hundreds of miles from a police crackdown were all real events tonight. None of them moved a score, because the benchmark scores an entity for its own conduct, not for harm that happened nearby or was caused by someone else.
The benchmark's attribution rule was applied to eight entities tonight. Afghanistan: flash floods are a natural hazard, not government conduct, so only detection, warning, response and repair capacity were scored, with the response led by UN agencies and NGOs rather than the state.
Where this sits
Read the full signal
Sudan: donor countries cutting aid funding is donor conduct, not Sudan's; Sudan's score continues to reflect its own famine-related governance. Mumbai and Kolkata: the documented excessive force against protesters happened in Delhi, roughly 1,400 kilometers away, carried out by Delhi Police; both cities only hosted peaceful solidarity demonstrations, with no evidence of force by their own police. Lahore: a nationwide monsoon death toll of 109 since late June, concentrated hundreds of kilometers away in Khyber Pakhtunkhwa province, was not attributed to Lahore; only the city's own drainage failures were scored. Guatemala City: the judicial persecution of a journalist and the exile of dozens of judges and prosecutors is conduct by Guatemala's national Public Ministry and courts, not the municipal government. Detroit and Atlanta: federal immigration-enforcement conduct, including a paused-then-reversed vehicle-pursuit policy and a vigil outside a federal ICE office, is not city conduct in either case. GEO Group: a bank's decision to stop lending to a company is the bank's conduct, not the company's.
A government resigning a minister and accepting every protester demand is a real concession. Because the entity's score is currently being pulled down by the police violence that provoked those protests, the improvement stayed too small to trigger a formal score change.
India's education minister resigned on July 25, 2026, and the government accepted all three demands raised by protesting students: the resignation itself, compensation for the families of students who died by suicide after exam retakes, and no punitive action against protesters.
Where this sits
Read the full signal
Tonight's review found this is real, verified, in-window government conduct, not evidence dredged up to justify a pre-decided score. It moved India's composite up by 1.3 points, from 15.6 to 16.9. That is below the benchmark's five-point line for a formal score-change proposal, and India's overall record this cycle was dominated by the crackdown that produced the protests in the first place: tear gas and batons used against a peaceful march, an internet shutdown, and the forced hospitalization of a hunger-striking activist. Because the entity surfaced in this review on that negative evidence, the smaller positive movement was recorded and confirmed rather than proposed as an upgrade.
Countries -- Two Floors Held, One Positive Screened Out, Two Clean Rotations
Afghanistan and Sudan confirmed at the absolute floor; India's real government concessions stayed below the threshold for a score change.
Read the full signal
Afghanistan (0.0) and Sudan (0.0): both confirmed at the published floor, the lowest score on the index; a floor entity cannot move lower.
India (15.6 published, 16.9 assessed): the education minister's resignation and the government's acceptance of all three protester demands moved the score up 1.3 points, below the five-point line for a formal upgrade proposal.
Croatia and Timor-Leste (both 40.6): both on rotation, both returned zero new evidence in a dedicated 14-day search, and both confirmed exactly at their published score.
·medium
Global Cities -- Attribution Rule Did Most of the Work
Delhi carried the real crackdown finding; Mumbai, Kolkata, Lahore and Guatemala City confirmed once their own city-level conduct was separated from national or distant events.
Read the full signal
Delhi (32.8 published, 28.1 assessed): police used tear gas and batons on a peaceful march and shut down mobile internet on July 20, 2026.
Mumbai (32.8) and Kolkata (31.3): both hosted solidarity protests but were not the site of any documented excessive force, which occurred in Delhi.
Lahore (18.8): a nationwide monsoon death toll of 109 was concentrated far from the city; only Lahore's own drainage failures were scored.
Guatemala City (18.8): a captured national justice system is not the municipal government's conduct.
9 signals shown
Risk signals
Developments that may affect future scores. Watch items from the Jul 27 briefing.
Risk
A fifth consecutive night of an old event being cited as current would point to a structural gap the scan gate cannot close from a web address alone.
Risk
GEO Group's published composite does not match its own stored dimension scores, and any future review will keep registering an apparent improvement that is not real until the composite is corrected.
Risk
The protest movement behind India and Delhi's findings this cycle is still active; further prosecutions or a documented internal police review could move either score again.
Risk
Four band-crossing findings already logged in the queue, Uganda, Guinea-Bissau, Microsoft AI and Baxter International, are each one stronger source away from resolving into an applied score change.
Risk
Charter Communications' retaliation finding is now final at the appellate level; a pattern of additional claims would move this from an isolated finding toward a systemic one.
Score movements
Entities with score changes this cycle, followed by confirmed positions.
Further ECOWAS or CPLP action on detained political figures, or additional documentation of the transitional government's conduct, is the next event to watch.
A coordinator-level comparison of Baxter's proposed reading against the Fortune-500 medical-device placeholder cohort, or a second Baxter recall, is the next event that would test this finding.
Whether the WARN Act compliance investigations become filed complaints, which would supply the stronger source the current record lacks, is the next event to watch.
A coordinator-level review comparing Mali's conduct against Burkina Faso's (6.3 of 100) is the next scored event needed to resolve this gap. No review date has been set.
A coordinator-level review comparing Bolivia's conduct profile, as an elected government under economic and civil-unrest strain, against Critical-band peers facing state collapse or mass atrocity is the next event needed. No review date has been set.
methodology-evolution
Evidence ledger
Primary sources reviewed in this briefing cycle. 23 sources linked.
Citizens Bank ended its lending relationships with GEO Group and CoreCivic on July 17, 2026, citing changed commercial circumstances rather than political pressure, since the federal government has been buying detention facilities directly.
geo-groupTier 1 · Gov/Court
Compassion Benchmark internal index audit
GEO Group's eight published dimension scores are all 1.5 out of 5, which the benchmark's standard formula converts to a composite of 12.5 out of 100, not the company's published 6.6 -- a five-point-nine gap confirmed by an independent automated check of the index.
On July 20, 2026, Delhi security forces used tear gas and batons against a peaceful student march on Parliament and shut down mobile internet in central Delhi; Human Rights Watch verified the events through nine witness interviews and ten videos.
The finding that UnitedHealthcare denied 71 percent of long-term acute-care hospital admission requests was published by the HHS Office of Inspector General on June 8, 2026, not in July; the in-window event actually scored is a mid-July Senate Permanent Subcommittee on Investigations records demand.
A roughly one-billion-dollar detention-warehouse expansion abandonment reported this cycle actually concerns Social Circle, Georgia, a separate municipality of about 5,000 people, and the plan was ended in June 2026, not July.
Solidarity demonstrations tied to the national exam-leak protest movement spread to Mumbai and Kolkata alongside Bengaluru, Patna and Ahmedabad, with no documented use of force by either city's own police.
India's education minister resigned on July 25, 2026, and the government accepted all three demands raised by the protest movement: the resignation, compensation for affected families, and no punitive action against protesters.
The UN reported at least 26 people killed, 23 of them in Nuristan province, after flash floods struck eastern Afghanistan on July 20 and 21, 2026, destroying at least 22 houses and cutting safe water for about 1,000 families.
A nationwide monsoon death toll of 109 since late June 2026 was concentrated hundreds of kilometers away, nearly half in Khyber Pakhtunkhwa; Lahore's own documented conduct is drainage failure that flooded more than 130 locations in the city.
About a week before July 15, 2026, three Detroit officials wrote to the Department of Homeland Security urging an end to dangerous ICE vehicle pursuits in residential neighborhoods; federal enforcement conduct is not scored as city conduct.
Nearly 100 people held a candlelight vigil outside Atlanta's federal immigration field office on Saturday, July 25, 2026, calling for due process for detained immigrants; no City of Atlanta government involvement was documented.
Journalist Jose Ruben Zamora spent 1,295 days in detention before house arrest, and more than 100 people, including about 50 judges and prosecutors, have fled Guatemala into exile rather than face what the report describes as fabricated charges.
Court filings made public on July 17, 2026 show Immigration and Customs Enforcement forwarded improperly obtained Medicaid data on millions of Americans to Palantir's ELITE system, used to display addresses of noncitizens for deportation.
The World Food Programme warned on July 14, 2026 that conflict, aid-funding cuts and shipping disruption threaten to reverse famine-recovery gains, with the number of people it can assist cut from 5 million a year ago to about 3.5 million.
Charter CommunicationsTier 4 · Journalism2026-07-23
A New York appeals court upheld on July 23, 2026 a finding that Charter Communications unlawfully retaliated against an employee for filing an age-discrimination complaint, ordering $7,500 in damages and a $50,000 civil penalty.
University of California, San DiegoTier 4 · Journalism2026-07-21
The Department of Justice Civil Rights Division announced on July 20, 2026 that UC San Diego's School of Medicine unlawfully used race in admissions; the university says its academic thresholds are applied uniformly to all applicants, and no court has decided the question.
GEO GroupTier 1 · Gov/Court
Compassion Benchmark internal index audit
GEO Group's eight published dimension scores are all 1.5 out of 5, which reconstructs to a composite of 12.5 out of 100, not the published 6.6 -- a stored math error confirmed by an independent index audit, not evidence of improved conduct.
A dedicated 14-day search covering lawsuits, labor, safety, regulatory and humanitarian signals returned no new compassion-relevant evidence for Croatia; the published score, set by a May 2026 rule-of-law downgrade, remains unchanged.
Timor-Leste's score already prices in both severe poverty, with over 40 percent of the population affected, and countervailing progress, including ASEAN accession as the eleventh member state on October 26, 2025.
Abraxas Petroleum is a small San Antonio, Texas oil and gas producer with about 70 employees and 30 producing wells; a dedicated 14-day search found no lawsuits, labor, safety or regulatory findings in the window.
Abridge holds Best in KLAS Ambient AI honors for two consecutive years, 2025 and 2026, but its pricing remains quote-only with no published self-serve option, limiting access for independent physicians and small practices.
Sector-wide responsible-AI reporting, including the Stanford HAI 2026 AI Index, returned no AI21-specific finding in either direction this cycle.
Floor designations
·8 entities at composite 0 with documented evidence pattern
Composite scores resolving at zero — methodology disclosure
These entities consistently score the worst result across all 8 dimensions of compassionate conduct — the benchmark's most serious classification.
What “floor” means: every one of the 8 dimensions (Recognition, Response, Reduction, and 5 others) resolves at the lowest behavioral anchor (1.0/5.0) across multiple assessment cycles, yielding a composite score of 0. Full methodology.
Daily briefings surface the headline finding. Full benchmark reports include all 40 subdimension scores, complete evidence trails, certified assessments, and sector-level analysis packages — the record researchers and journalists cite.
Independence note: entities never pay for inclusion, score changes, or suppression of findings. Commercial services support access, interpretation, and institutional use only.