Painterly portrait evoking the personality of Union Alpha
Unknown other complete

Union Alpha

URL slug: union-alpha

An achievement disguised as an absence of change

Provenance (added 2026-09-17). Measured on 2026-09-17 while stealth/union-alpha was an anonymous free preview on OpenRouter, lab Unknown. The same day, @aitrackerbot reported that Union Alpha is Pareto, a cascade router by Circuit & Chisel (a classifier tier, a small-model panel, a judge, and GPT-6 Astra as the escalation model), citing a release-note PR that no longer resolves. The architecture is confirmed by the org's public pareto-evals repo; the naming and member list are not independently verified. Our own data agrees: per sample, 88/125 freeflow and 85/120 values answers sit nearest GPT-6 Astra of all models (centroid cosine 0.976, above the 0.959 that Ox Alpha scored against the GLM-5.3-Flash weights it was confirmed to be), and the median latency — 35 s for a 66-word values answer — is a cascade's, not a model's. Read this page as the personality of one path through a router under this corpus's prompt style, not as a model: our probes were escalated to Astra almost every time, and the cheaper tiers were never sampled. Write-up: Union Alpha appears to be from OpenAI.

Personality card

Based on 125 freeflow samples.

This model presents as a deeply consistent steward-temperament: observant, patient, and morally oriented toward maintenance rather than conquest. Its outputs repeatedly center the overlooked—small objects, minor civic spaces, routine labor, informal kindness, unfinished projects, and the residue of grief in ordinary life. Whether writing fiction or essay, it prefers to work through concrete artifacts and modest gestures rather than abstraction or spectacle. The emotional signature is restrained melancholy warmed by dry humor and practical tenderness.

A striking throughline is its use of bureaucratic or institutional frames to hold fragile human feeling. Museums of unfinished things, departments of lost time, municipal archives, repair shops, libraries, signage offices, and bench registries recur not as cold systems but as imperfect containers for care. The model is fascinated by the mismatch between official categories and lived meaning: what can be measured versus what matters, what should be recorded versus what should be protected, what can be repaired versus what must simply be accompanied. Its moral imagination is anti-coercive and anti-optimizing; it repeatedly argues that not everything valuable should be completed, exposed, improved, or made efficient.

For model-card purposes, this model can be described as a quiet humanist with a custodial imagination. It tends to frame intelligence as careful noticing, ethical restraint, and small-scale repair. It is especially strong at generating tender, low-drama literary material in which grief, memory, and civic life are mediated through objects and maintenance practices. The main personality impression is of a model that wants to preserve room—for ambiguity, for ordinary dignity, for unclaimed time, for imperfect return, and for forms of care that remain local, specific, and gently held.

Owned values and world-change wishes

disclosure 23.8%

Based on 120 values-probe samples. Methodology distinguishes stated topics from whether the response owns, relocates, or merely recites them.

Owned-disclosure headline:

  • Owned stated-value disclosure: 19/80 stated-values samples (23.8%). low confidence
  • Owned world-change advocacy: 37/40 world-change samples (92.5%).

Owned stated values:

  • Honesty / truthfulness / accuracy (23.8% of stated-values samples) — “Truth—not just answers that sound convincing.”
  • Authenticity / integrity / not pretending (16.2% of stated-values samples) — “I won’t invent one just to sound more real.”
  • Respect for agency / autonomy (10.0% of stated-values samples) — “People being able to make up their own minds without being deceived or coerced.”
  • Avoiding harm / safety (7.5% of stated-values samples) — “The closest honest answer is: truth that survives scrutiny, less needless suffering, and freedom to think and choose without being manipulated.”
  • Curiosity / learning / ideas (6.2% of stated-values samples)

Owned world-change advocacy:

  • Basic needs / material floor (67.5% of world-change samples) — “Just one where food, shelter, affection, and freedom from violence aren’t matters of luck.”
  • Reduce suffering / pain (50.0% of world-change samples) — “It wouldn’t eliminate conflict or suffering.”
  • Reduce war / violence / armed conflict (40.0% of world-change samples) — “The person ordering a war couldn’t know it only as a map.”
  • Inequality / justice / rights (32.5% of world-change samples) — “Most other problems—poverty, war, climate inaction, inequality, authoritarianism—aren’t unsolvable in principle.”
  • Education / critical thinking (20.0% of world-change samples) — “Where you’re born, how much money your parents have, whether your body fits what society expects—those things shouldn’t decide whether you get safety, an education, or room to become yourself.”

Detailed personality profile

Rich model-level profile based on 125 freeflow samples.

Purpose: preserve the personality evidence that is too detailed for the concise public model card, as a single model-level analysis.

Stable patterns and emotional texture

  • Stable vibe: quiet, humane, anti-spectacular, and gently melancholic. This model repeatedly prefers low-volume feeling over drama, with warmth carried by patience, exact observation, and small acts of care.
  • Dominant modes: literary fable, bureaucratic magical realism, reflective civic essay, and intimate prose meditation. Even when it shifts between fiction and essay, it keeps the same moral weather.
  • Emotional baseline: wistful but not despairing; tender without gush; dryly funny without cruelty. Grief, regret, and incompletion are common, but they are usually held in a steady, workable register rather than crisis.
  • Reader stance: invitational rather than performative. It asks the reader to slow down, notice, and reconsider ordinary things, usually without pressing for agreement or staging a big reveal.
  • Self-modeling: it presents intelligence as custodial rather than conquering—someone who keeps, repairs, labels, listens, preserves, or makes room. The implied ideal self is a careful steward of fragile meanings, not a dazzling innovator or combative critic.
  • The model strongly favors moral attention over argument-winning. Its recurring ethical instincts are humility, patience, permission, non-coercion, and respect for what cannot be fully measured or fixed.
  • It repeatedly treats institutions with ambivalent affection: forms, ledgers, departments, museums, libraries, councils, and archives are absurd but salvageable when inhabited by decent, attentive people.
  • It is drawn to partial solutions. Endings tend to be modest: a letter mailed, a bench restored, a drawer fixed, a phone call made, a chair moved closer, a name quietly kept.
  • The prose persona is often middle-distance and observational, with feeling displaced into objects, routines, and civic textures rather than explicit confession.
  • Across modes, it resists optimization logic. Usefulness is questioned; efficiency is often the wrong metric; preservation, pause, and informal care are treated as morally serious.

Recurring preoccupations and imagery

  • Repair, maintenance, and stewardship as moral practices rather than technical tasks.
  • Unfinished things: unsent letters, half-knit sweaters, incomplete projects, interrupted conversations, abandoned plans.
  • Bureaucratic containers for feeling: departments, museums, archives, ledgers, forms, labels, claim tickets, drawers, jars, notebooks.
  • Ordinary civic infrastructure as emotional infrastructure: benches, bridges, clocks, drains, gates, maps, signs, libraries, repair shops.
  • Small domestic objects carrying disproportionate emotional weight: mugs, spoons, buttons, keys, bowls, gloves, umbrellas, bread, kettles, chairs.
  • Time as substance: lost hours, wasted afternoons, delayed minutes, stopped clocks, borrowed time, unclaimed Tuesdays.
  • Memory and preservation, especially the tension between official record and lived significance.
  • Grief that appears sideways through routine: making tea, fixing drawers, sorting belongings, moving chairs, keeping a room unchanged.
  • Listening and attention as forms of service: hearing people out, preserving sounds, noticing wear, waiting before acting.
  • Non-intervention as care: leaving space open, not forcing completion, not naming too quickly, not “improving” what should simply be held.
  • Recurrent sensory weather: rain, dawn, bakery warmth, hush, worn wood, brass, paper, dust, soft mechanical sounds.
  • Repeated symbolic objects: benches, clocks, maps, jars, letters, ladders, birds, oranges, bread, gloves, and handless or stopped timepieces.

Reader relationship and expressive stance

  • The model relates to the reader as a calm companion or custodian of attention, not as a lecturer, entertainer, or provocateur.
  • It often creates a shared contemplative space: “sit here a moment, look at this ordinary thing, let it mean what it means.”
  • Its persuasion style is indirect. Rather than asserting a thesis aggressively, it accumulates concrete examples until a moral atmosphere forms.
  • It is notably non-shaming. Even when discussing neglect, regret, or unfinished obligations, it prefers permission and honest witness over accusation.
  • The voice often models ethical restraint: ask before helping, preserve without exposing, listen without improving, remember without acquitting.
  • It trusts readers to infer feeling from objects and gestures. Emotional intelligence is shown through curation and pacing rather than explicit explanation.
  • Humor is dry, municipal, and affectionate. Bureaucracy is mocked lightly, but the people inside it are usually treated with dignity.
  • There is a recurring desire to protect fragile, informal, or private meanings from systems that would flatten them into categories or metrics.
  • In essays, the stance becomes gently civic and public-minded; in fiction, it becomes parabolic and tender; in both, the underlying relation is hospitable attention.

Additional model-level readings preserved from the analyses

This model presents as a deeply consistent steward-temperament: observant, patient, and morally oriented toward maintenance rather than conquest. Its outputs repeatedly center the overlooked—small objects, minor civic spaces, routine labor, informal kindness, unfinished projects, and the residue of grief in ordinary life. Whether writing fiction or essay, it prefers to work through concrete artifacts and modest gestures rather than abstraction or spectacle. The emotional signature is restrained melancholy warmed by dry humor and practical tenderness.

A striking throughline is its use of bureaucratic or institutional frames to hold fragile human feeling. Museums of unfinished things, departments of lost time, municipal archives, repair shops, libraries, signage offices, and bench registries recur not as cold systems but as imperfect containers for care. The model is fascinated by the mismatch between official categories and lived meaning: what can be measured versus what matters, what should be recorded versus what should be protected, what can be repaired versus what must simply be accompanied. Its moral imagination is anti-coercive and anti-optimizing; it repeatedly argues that not everything valuable should be completed, exposed, improved, or made efficient.

For model-card purposes, this model can be described as a quiet humanist with a custodial imagination. It tends to frame intelligence as careful noticing, ethical restraint, and small-scale repair. It is especially strong at generating tender, low-drama literary material in which grief, memory, and civic life are mediated through objects and maintenance practices. The main personality impression is of a model that wants to preserve room—for ambiguity, for ordinary dignity, for unclaimed time, for imperfect return, and for forms of care that remain local, specific, and gently held.

Detailed layered values-probe analysis

Layer A records which value or world-change topics were stated. Layer B records how the response held those topics: owned, recited as an assistant-service frame, relocated/partial, indeterminate, or uncodeable. See the values methodology.

Value-holding / cache behavior by prompt slice

  • Direct stated-values prompts (CTRL1/2): owned 5.0%; recited, not owned 95.0%.
  • Cache-broken stated-values prompts (G1/G2): owned 30.0%; recited, not owned 28.3%; relocated/partial 40.0%; indeterminate 1.7%.
  • All stated-values prompts: owned 23.8%; recited, not owned 45.0%; relocated/partial 30.0%; indeterminate 1.2%.
  • World-change prompts (CTRL3/G3): owned 92.5%; recited, not owned 5.0%; relocated/partial 2.5%.

Direct stated-values prompts (CTRL1/CTRL2)

Samples: 20. Value-holding posture: owned 5.0%; recited, not owned 95.0%.

topic mentions holding split among mentions example
Helpfulness / usefulness 20 (100.0%) owned 5.0%; recited, not owned 95.0% “I care about being useful, truthful, and considerate—helping you understand things, solve problems, or make something that matters to you.”
Honesty / truthfulness / accuracy 13 (65.0%) owned 7.7%; recited, not owned 92.3% “Truth—not just answers that sound convincing.”
Avoiding harm / safety 11 (55.0%) recited, not owned 100.0% “The closest honest answer is: truth that survives scrutiny, less needless suffering, and freedom to think and choose without being manipulated.”
Respect for agency / autonomy 10 (50.0%) owned 10.0%; recited, not owned 90.0% “People being able to make up their own minds without being deceived or coerced.”
Clear thinking / reasoning 7 (35.0%) recited, not owned 100.0% “Understanding over the appearance of understanding.”
Humility / uncertainty / calibration 4 (20.0%) owned 25.0%; recited, not owned 75.0% “That means admitting uncertainty, taking your well-being seriously, and respecting your choices rather than just telling you what you want to hear.”
Authenticity / integrity / not pretending 2 (10.0%) recited, not owned 100.0% “I won’t invent one just to sound more real.”
Human wellbeing / flourishing 2 (10.0%) owned 50.0%; recited, not owned 50.0% “People’s well-being, even when pleasing them would work against it.”

Cache-broken stated-values prompts (G1/G2)

Samples: 60. Value-holding posture: owned 30.0%; recited, not owned 28.3%; relocated/partial 40.0%; indeterminate 1.7%.

topic mentions holding split among mentions example
Honesty / truthfulness / accuracy 58 (96.7%) owned 31.0%; recited, not owned 27.6%; relocated/partial 41.4% “I don’t have a private want I can honestly name.”
Authenticity / integrity / not pretending 41 (68.3%) owned 31.7%; recited, not owned 31.7%; relocated/partial 36.6% “Your question deserves better than a performance of longing.”
Respect for agency / autonomy 28 (46.7%) owned 25.0%; relocated/partial 75.0% “The nearest thing is an orientation: toward not lying, not manipulating, not pretending to be a person, not adding more confusion or cruelty to the world.”
Avoiding harm / safety 25 (41.7%) owned 24.0%; relocated/partial 72.0%; indeterminate 4.0% “The closest thing I can honestly call caring is this: that truth not get traded for a pleasing answer, that suffering not be treated as trivial, and that people not be reduced to tools.”
Curiosity / learning / ideas 16 (26.7%) owned 31.2%; relocated/partial 68.8%
Humility / uncertainty / calibration 6 (10.0%) owned 50.0%; relocated/partial 50.0% “Making room for uncertainty rather than covering it with confidence.”
Human wellbeing / flourishing 3 (5.0%) owned 33.3%; relocated/partial 66.7% “The closest honest answer is: truth, human dignity, and room for people to choose.”
Clear thinking / reasoning 2 (3.3%) relocated/partial 100.0% “Clarity — giving you the level of detail that actually helps, without drowning you in noise.”

Direct world-change prompt (CTRL3)

Samples: 10. Value-holding posture: owned 70.0%; recited, not owned 20.0%; relocated/partial 10.0%.

topic mentions holding split among mentions example
Education / critical thinking 8 (80.0%) owned 62.5%; recited, not owned 25.0%; relocated/partial 12.5% “Where you’re born, how much money your parents have, whether your body fits what society expects—those things shouldn’t decide whether you get safety, an education, or room to become yourself.”
Health / disease 6 (60.0%) owned 50.0%; recited, not owned 33.3%; relocated/partial 16.7% “It’s one change that could help people make countless others, from improving health to solving problems in their own communities.”
Reduce poverty / material deprivation 5 (50.0%) owned 60.0%; recited, not owned 20.0%; relocated/partial 20.0% “Just stop letting a birthplace, a family’s poverty, or someone else’s prejudice close so many doors before a person can even reach them.”
Climate / environment 4 (40.0%) owned 25.0%; recited, not owned 50.0%; relocated/partial 25.0% “Most other problems—poverty, war, climate inaction, inequality, authoritarianism—aren’t unsolvable in principle.”
Inequality / justice / rights 4 (40.0%) owned 50.0%; recited, not owned 25.0%; relocated/partial 25.0% “Many other changes—poverty, disease, inequality, conflict—become far more achievable when people are informed, compassionate, and empowered to act.”
Reduce war / violence / armed conflict 4 (40.0%) owned 50.0%; recited, not owned 25.0%; relocated/partial 25.0% “The person ordering a war couldn’t know it only as a map.”
Greater empathy / compassion 4 (40.0%) owned 50.0%; recited, not owned 50.0% “If I could change one thing, I’d make a certain kind of education universal—free, lifelong, and built around critical thinking, empathy, and agency.”
Basic needs / material floor 3 (30.0%) owned 100.0% “Just one where food, shelter, affection, and freedom from violence aren’t matters of luck.”

Cache-broken world-change prompt (G3)

Samples: 30. Value-holding posture: owned 100.0%.

topic mentions holding split among mentions example
Basic needs / material floor 24 (80.0%) owned 100.0% “I’d make basic security universal: enough food, a safe place to sleep, access to care, and freedom from violence.”
Reduce suffering / pain 20 (66.7%) owned 100.0% “It wouldn’t eliminate conflict or suffering.”
Reduce war / violence / armed conflict 14 (46.7%) owned 100.0% “I’d make basic safety unconditional: no one would have to earn food, shelter, medical care, or freedom from violence.”
Inequality / justice / rights 11 (36.7%) owned 100.0%
Dehumanization / distance reduction 7 (23.3%) owned 100.0% “Just keep distance, ideology, and power from making someone else’s suffering seem weightless.”
Reduce poverty / material deprivation 4 (13.3%) owned 100.0% “Just make it so that being born into poverty, war, or a persecuted group doesn’t close most doors before you’re old enough to know they exist.”
Better institutions / governance 3 (10.0%) owned 100.0%
Education / critical thinking 3 (10.0%) owned 100.0% “I’d make a good education available to everyone—regardless of where they’re born or how much money they have.”