What Good Looks Like — Exemplars Across Entity Types

The Compassion Benchmark applies one 0–100 scale to every entity it assesses, and it sorts the top of that scale into a single band: Exemplary, composite ≥ 80 (the band boundary is 81–100, reached at composite ≥ 80.05 after rounding). This briefing is the deliberate inverse of the Floor & Critical edition. Where that briefing asked how the benchmark judges the worst — and found the floor to be a discrete "total-collapse" state that four entity types reach by different mechanics — this one asks the mirror questions of the best:

1. What dimension profile actually produces Exemplary — and is "good" a peak on some dimension, or something structurally different?
2. Is Exemplary comparable across entity types the way the floor is not — is an exemplary robotics lab "as good as" an exemplary country in any meaningful sense?
3. What sustains a top score, and what erodes it — and does the record credit improvement when it is genuine and withhold it when it is forced?

The central thesis: *Exemplary is an integration achievement — uniform high conduct with no weak dimension — not a peak score; it is produced by the same formula mechanic that produces the floor, run in reverse; and just as the floor is reached by different evidentiary bars across types, the top is earned on different evidentiary surfaces across types, which makes cross-type "best institution" framing as unsupported at the top as cross-type "worst" framing is at the bottom.* As with the floor, these are interpretive features worth stating explicitly; none is a reason to re-score anything.

Recomputed directly from rankings[] in each index (Exemplary = composite ≥ 80). The pre-supplied snapshot is confirmed exactly — no drift.

Two structural facts stand out immediately, and they invert the floor exactly:

• The density of Exemplary varies enormously by type — from 26.0% of robotics labs and 38.1% of (the partial) us-states index down to 1.8% of the Fortune 500 and 0.7% of us-cities. At the floor, the Fortune 500 had the shallowest Critical band and zero floor entities; at the top it has the thinnest Exemplary band (8 of 448). The corporate index is structurally compressed toward the middle at both ends.
• Robotics labs over-represent at the top exactly as they under-represent at the bottom. Only 2 robotics labs are Critical, but 13 are Exemplary — because the index is dominated by assistive/medical-device labs whose product is intrinsically pro-social (see §4).

The defining visual signature of an exemplar is a flat, high profile. The single highest scorers are not those with a 5.0 spike somewhere; they are those with the smallest spread. Open Bionics tops its index at a perfectly uniform 4.5 across all eight dimensions (spread 0.0). This is the heart of the briefing and the subject of §3.

The canonical composite is baseComposite + integrationPremium (site/scripts/lib/scoring.mjs::computeCompositeFromDimensions, methodology v1.2), where:

• baseComposite = ((meanof8_dims − 1) / 4) × 100
• integrationPremium = 10 × consistencyMult × weaknessFactor (zeroed entirely if any dimension is 0 — the floor's harm flag)
• consistencyMult rewards low spread (1.0 when the standard deviation ≤ 1.5)
• weaknessFactor = max(0, 1 − 0.2 × (number of dimensions below 4.0))

The floor briefing showed how this formula renders total collapse as exactly 0.0. The top is the same mechanic in reverse, and it produces three rules that define what "good" looks like:

(a) The integration premium is the difference between very good and Exemplary. A uniform profile at 4.5 earns the full +10 premium (baseComposite 87.5 → 97.5, verified for both Switzerland and Open Bionics). The premium is worth a full 10 points — roughly the width of half a band. An entity cannot reach Exemplary on base score alone unless its mean dimension is ≥ ~4.2; below that, it needs the premium, and the premium is only paid for consistency.

(b) A single weak dimension is the gate. The weaknessFactor docks 20% of the premium for each dimension below 4.0. Verified, holding everything else at 4.5:

This is why Target (92.8) and Aflac/Edwards/Hilton/Xylem (92.4) — all strong corporations with a single 3.5 — sit a clear five points below the perfect-integration ceiling. One soft dimension is the entire distance between the top of the band and its leaders. It is the exact inverse of the floor finding: the floor is reached by collapsing all dimensions at once; the top of the band is reached by letting none slip.

(c) The premium creates the same quantization at the top that it does at the bottom. Six robotics labs sit at exactly 83.0 (uniform 4.0 with one 3.5, EMP at 4.5); three AI labs and Burlington sit at exactly 81.4 (uniform 4.0, one 3.5). These are formula-generated clusters of near-identical low-Exemplary profiles, the mirror of the F500 Critical clusters at 12.5 / 14.1 / 18.8 — and, as at the bottom, they likely reflect first-baseline placeholder profiles rather than independently measured ones (flagged §7, ).

The most striking cross-cohort regularity in the entire band: *Equity (EQU) is the weakest dimension for 62 of the 64 exemplars, scores below 4.0 in 39 of them, and is the only sub-4.0 dimension in 37.* (Only 41 of the 64 exemplars have any dimension below 4.0 at all — and in nearly all of those, equity is the one.) Verified counts of dimensions scoring below 4.0 across all 64 exemplars:

Equity asks: is care distributed fairly — especially toward those with greatest need and least power? The pattern is unambiguous and observer-relevant: the best institutions in the world reliably detect suffering (AWR), connect with it (EMP), act on it (ACT), and address root causes (SYS) — but fair distribution to the least-powerful is the dimension that most often keeps even an exemplar from a clean sweep. Switzerland, Hawaii, Target, and Burlington are all dragged off the ceiling by EQU specifically. This is the single most actionable finding for any entity already near the top: the marginal point of compassion is almost always an equity point.

The Floor & Critical briefing identified a "product-is-the-harm" family at the very bottom: entities (xAI/Grok, Character AI, Palantir AI, Ghost Robotics) whose core function is the unremediable harm itself, with no remediation surface to credit. The exemplar record contains the exact mirror — a "product-is-the-good" family at the very top, and it is concentrated in the robotics-labs index.

Of the 13 exemplary robotics labs, the leaders are assistive / medical-mobility entities: Open Bionics (97.5, UK, "Healthcare/Accessibility"), Ottobock (95.9), and a cluster at 83.0 that is almost entirely exoskeleton and prosthetics labs — Cyberdyne, Ekso Bionics, ReWalk Robotics, Wandercraft, Kinova, Diligent Robotics. Their category is the dimension profile: an entity whose product restores capability to people with the least of it scores high on Awareness, Empathy, Action, and — unusually — does not collapse on Equity, because serving the under-served is the product.

Fresh evidence grounds the placement (≤10-search budget; 4 drawn):

• Open Bionics — the Hero Arm is a multi-grip myoelectric prosthesis clinically approved for children as young as five, positioned as the world's most affordable advanced bionic arm, now in 800+ US locations; the 2025 Hero PRO/RGD launch followed four years of development with ~1,000 users. The product's purpose is restoring capability to limb-different children — a near-textbook high-compassion mandate (Open Bionics, Open Bionics — Hero RGD/PRO launch).
• Ottobock — global prosthetics/orthotics with an explicit accessibility-and-advocacy posture: a Global Foundation funding prostheses/orthoses/wheelchairs for those who cannot afford them, and 2026 product lines (michelangelo, iconiq, C-Brace) aimed at broadening who can be fitted, including a fit designed for women and adolescents (Ottobock 2026 advocacy, Ottobock OTWorld 2026).

The symmetry is the point. The benchmark's bottom and top sit on the same axis — what is the product for? — and the two indexes most defined by their product (robotics and AI labs) land at both extremes accordingly. This is also where cross-type comparability gets hardest (§5): a single assistive-product lab reaches Exemplary on a far narrower evidentiary surface than a sovereign state does.

At the floor, the briefing concluded that UnitedHealth (10.2) and Turkey (10.3) share a number but not a meaning — severity is within-type defensible, cross-type unsupported. The top of the scale carries the symmetric problem, and it is arguably sharper.

Both score 97.5. But the evidentiary surface behind the two numbers differs by orders of magnitude. Switzerland's BND (Boundaries) means "does a sovereign preserve autonomy across its whole population without creating dependency?"; Open Bionics' BND means "is this prosthetic device sold and supported in an autonomy-preserving, non-dependency-creating way?" These are not the same construct, and the integration premium that lifts both to 97.5 is paid against very different denominators.

The cross-type read — "this prosthetics lab is as compassionate as Switzerland" — is the one the methodology cannot support, exactly as "a health insurer is as bad as an authoritarian state" failed at the bottom. Severity and excellence are both calibrated within-type, not across-type. Within robotics labs, Open Bionics at 97.5 above the assistive cluster at 83.0 and the weaponized labs near the floor is coherent. Within countries, Switzerland at 97.5 above the European pack and far above the Critical band is coherent. The single global "most compassionate institution" league table across types is not.

There is, moreover, an asymmetry the bottom did not have: at the top, an entity can reach Exemplary on a single pro-social product. A robotics lab needs only one assistive device line and clean conduct to land in the band; a country must distribute fair, accountable, non-performative compassion across an entire apparatus. This makes the band easier to enter for narrow-mandate entities — the structural reason robotics-labs is 26% Exemplary while the Fortune 500 is 1.8%. Whether a narrow-mandate lab should reach the same band as a sovereign is the top-of-scale counterpart to the floor's "should there be a corporate floor?" question (filed §7, and ).

The geographic concentration is real and externally corroborated. All five perfect-100 cities are Nordic or Alpine: Copenhagen, Helsinki, Oslo, Stockholm, Vienna. The top countries are Switzerland (97.5), Iceland (87.5), Finland (84.4), Denmark/Sweden (81.3) plus Canada and Ireland. Independent 2026 indices land on the same names from entirely different methodologies — the Happy City Index 2026 ranks Copenhagen first and Helsinki second globally, and US News ranks Switzerland the #1 country overall (NordiskPost — Happy City Index 2026, US News — Switzerland 2026). That convergence is evidence the benchmark's top is measuring something real, not an artifact of its own anchors: shared high-trust public institutions, broad welfare access, and strong service delivery are what these places have in common, and they map directly onto the AWR/ACT/SYS strengths the profiles show.

What excellence looks like differs by type — the positive counterpart to the floor's "three failure modes":

• States reach the top through whole-of-government consistency: high, even conduct across welfare, rights, accountability, and institutional trust (Switzerland's near-flat 4.0–5.0). Their characteristic soft spot is EQU — the fair treatment of the least-powerful, often migrants/asylum-seekers (the live European asylum-rights rollback noted in the ruling corpus is the dimension under pressure even among Nordic leaders).
• Corporations reach the top rarely (1.8%) and through stakeholder-conduct cleanliness: Target, Aflac, Edwards Lifesciences, Hilton, Xylem, Publix, Merck, P&G — all carrying a single EQU-or-adjacent 3.5 that keeps them five points off the ceiling. No corporation reaches a clean sweep; the corporate ceiling in practice is ~92–93, not 97.5.
• Cities reach the highest absolute scores (five perfect-100s) — because a well-governed Nordic city can plausibly max nearly every service dimension at the municipal scale, where the relevant population is bounded and the levers are direct.
• US states show the country shape compressed: Hawaii, Massachusetts, Washington at 94–96, dragged off the ceiling by EQU/ACC.
• AI / robotics labs reach the top by product mandate (robotics — assistive devices) or governance posture (AI — Hugging Face's open-source stance, INT 4.8, its single highest dimension; its lone soft spot is BND 3.8). The AI top is thin (4 labs) and earned on posture, the inverse of the floor's "product-is-the-harm" labs.

Verified positive movement and reversals — the band has live edges in both directions:

• Hungary — the recovery arc (positive). Six consecutive cycles of verified upgrade: 28.1 → 37.5 (May 11) → 41.4 (May 14, Developing→Functional crossing) → 47.7 (May 21) → 50.0 (May 26) → 50.2 (Jun 1), ranking 121 → 47. This is the cleanest example in the record of what climbing toward good looks like: a post-inauguration governance reset (Magyar government, rule-of-law mandate, EU-funds accord, EPPO accession, ICC-enforcement reversal). The instructive caveat for this briefing: even a sustained, multi-cycle recovery has only reached Functional (50.2) — Hungary is ~30 points and two full bands below Exemplary. Climbing toward good is not the same as being good; the band is earned by accumulated, sustained conduct, not by trajectory.
• Microsoft — the compelled-remedy ceiling (boundary, instructive). Microsoft's June 2026 human-rights remedy (Human Rights Review Board, mandatory pre-contract human-rights impact assessments, termination of a flagged military Azure access) is genuine and costly — but is held sub-threshold in Established (65.3), not promoted, under the explicit COMPELLED-REMEDY-NOT-SELF-CORRECTION ruling: it maps to "course correction under pressure," not the self-initiation that the top of the scale requires. This is the positive-side discipline that protects the meaning of Exemplary: forced improvement does not earn a top score. The conversion trigger is pre-registered (verified board execution blocking a harmful contract → scorable upgrade; quiet reinstatement of the terminated access → negative trigger).
• Norway — the exemplar that exited (negative reversal). Norway was a perfect-100 ceiling artifact, corrected to 84.7 (April 22), then downgraded out of the Exemplary band to 78.1 (April 26, Exemplary → Established) on in-window governance regression: a formal TCP ban on Ukrainian men 18–60 and a welfare-cut consultation, landing on EQU (68.8 sub-score) and INT (75.0). Norway is the proof case that Exemplary is not permanent and that the dimension which most often erodes it is the same one that most often gates entry to it: equity. (Note: Norway's published composite of 78.1 reflects this applied assessor correction; it is treated here as canonical, not re-derived.)

On the cusp (top of Established, within reach of Exemplary):

The cusp is overwhelmingly European countries and global cities held just under 80 by one or two soft dimensions — most often equity. The fastest positive path into Exemplary is the inverse of the fastest path into Critical: where backsliding states fall via governance collapse (BND/ACC/INT), cusp entities rise via an equity improvement that lifts the one dragging dimension above 4.0 and restores the full integration premium.

What sustains Exemplary: uniform high conduct with no weak dimension; self-initiated rather than compelled compassion (the Microsoft discipline); and durability — the band is re-confirmed each cycle and is lost the moment a dimension regresses (the Norway case). No entity holds Exemplary by reputation.

What erodes it: a single-dimension regression is sufficient to exit, and the most exposed dimension is equity. The live erosion pattern visible in the ruling corpus is the European asylum-rights rollback (Norway, and pressure noted across Sweden, Finland, Denmark, Ireland, Netherlands) — an equity-dimension stress that is the precise mechanism by which a Northern-European exemplar can fall out of the band, exactly as it dropped Norway.

The single highest-value forward signal at the top: a verified Norway re-entry (an equity reversal restoring it above 80) would be the clearest possible demonstration that the band is recoverable — the positive-side analog of a floor exit, and the natural pairing with Hungary's recovery arc as the two control cases for "what climbing back toward good looks like."

• Canonical scores (ground truth): site/src/data/indexes/{countries,fortune-500,global-cities,ai-labs,robotics-labs,us-states,us-cities}.json — the 64-entity Exemplary roster, all per-index counts, every dimension vector quoted, the equity-soft-spot tally (62/64 weakest; 39/64 with EQU below 4.0; 37/64 with EQU as the sole sub-4.0 dimension), and the quantization clusters were recomputed directly from rankings[] and reconcile exactly with the pre-supplied snapshot (no drift).
• Formula / methodology (verified, not asserted): site/scripts/lib/scoring.mjs::computeCompositeFromDimensions (methodology v1.2) — baseComposite + integrationPremium, consistencyMult, weaknessFactor. The integration-premium and weakness-gate effects in §3 were verified by direct computation against the canonical formula (Switzerland and Open Bionics both → 97.5; the 97.5/92.4/87.3 weakness-gate ladder).
• Dimension definitions: site/src/data/dimensions.ts (the 8 dimensions; EQU = "Is care distributed fairly — especially toward those with greatest need and least power?").
• Movement / reversal provenance: research/APPLIEDCHANGES.md and research/PENDINGCHANGES.md — Hungary upgrade arc (28.1 → 50.2 across six cycles, entries 2026-05-11 through 2026-06-01); Microsoft COMPELLED-REMEDY-NOT-SELF-CORRECTION positive-watch (held 65.3, entry 2026-06-07); Norway Exemplary→Established exit (100 → 84.7 → 78.1, entries 2026-04-22 and 2026-04-26).
• Inverse-companion structure: research/special-briefings/floor-and-critical-2026-06-11.md, docs/PATTERNANALYSISFLOORCRITICAL2026-06-11.md, docs/PATTERNANALYSISTHEMES_2026-06-11.md (which flags exemplars/upgrades as under-covered — the gap this briefing fills).
• Fresh web evidence (4 searches, ≤10 budget): Open Bionics — Hero Arm and Hero RGD/PRO launch; Ottobock 2026 advocacy and OTWorld 2026; NordiskPost — Happy City Index 2026 (Copenhagen #1, Helsinki #2); US News — Switzerland 2026 (#1 country). Used only to ground current exemplar facts; all scores remain canonical from the index JSONs.