Open Ethos is built to address five problems, ordered here from the most far-reaching to the most personal. The first two are the reason the project matters at scale. The last three are the foundation everything else is built on.

• The Value Specification Problem: Making AI Alignment Possible
• The Political Dysfunction Problem: Accountability Without Bias
• The Cross-Cultural Understanding Problem: Seeing Through the Noise
• The Decision Quality Problem: Reasoning Instead of Reacting
• The Self-Knowledge Problem: Knowing What You Actually Believe

The Value Specification Problem: Making AI Alignment Possible

The paperclip problem isn't about paperclips. It's about the gap between what humans value and what we can formally specify to an optimization system. Every current alignment approach tries to close that gap from the AI side: RLHF infers values from thumbs-up/thumbs-down (noisy, shallow, biased toward the feedback population), Constitutional AI has a small team write principles (whose values?), inverse reward design reverse-engineers values from behavior (behavior reflects constraints and incentives, not just values). All are indirect approximations of something that does not exist yet: a large-scale, structured, machine-readable dataset of what humans actually value, stated directly, verified for consistency, and honest about where values conflict.

Open Ethos produces exactly that. When a user calibrates weights across eight moral axioms, sets social distance preferences and time horizons, then demonstrates coherence between that calibration and their judgments across diverse cases, they generate a formally articulated value specification richer than anything alignment research currently has access to. At population scale, across millions of users from different cultures, political orientations, and moral traditions, this becomes a map of human values with structure and verification that has never existed.

Critically, the data captures genuine disagreement. Different users weight axioms differently, and the system does not pretend there is a single correct answer. An alignment system trained on this data learns not just what humans value but where they agree, where they disagree, by how much, and what the shape of disagreement looks like. That lets alignment systems navigate real moral uncertainty rather than flattening it into a single reward signal.

The Political Dysfunction Problem: Accountability Without Bias

Every high-stakes professional domain has accountability mechanisms tying decision-makers to their stated reasoning. Scientists publish falsifiable predictions. Traders take positions marked to market. Engineers sign specifications. Politicians have nothing comparable: vague promises, untracked position shifts, zero scored consistency. The system selects for people skilled at performing conviction rather than holding it.

Open Ethos makes political reasoning auditable. A politician who publishes their profile commits to a formal value model: specific axiom weights, demonstrated through case judgments, verified for coherence. Every subsequent vote becomes checkable against their own framework. Not against an external standard, not against the opposition's values, against the politician's own stated calibration. Drift becomes visible. Contradictions become quantifiable. Consistency becomes comparable across politicians and across time.

For voters, this means you can read what a candidate actually commits to valuing and compare it to alternatives. For journalists, it means accountability reporting that does not require taking sides, because the standard was set by the subject. For the system, it means the same commitment-and-verification discipline that makes every other professional domain function, imported into the one domain operating without it.

Adoption does not require mandates. Challengers publish profiles as a differentiator. Incumbents face growing cost of refusal. Competitive pressure does the rest, the same dynamic that made tax return disclosure a de facto presidential requirement.

The Cross-Cultural Understanding Problem: Seeing Through the Noise

Most cross-cultural conflict looks like a clash of civilizations. Much of it is actually a small gap in value weighting, amplified by language differences, historical grievance, and tribal psychology until it feels impassable. We cannot tell the difference because we have never had a shared framework for comparing values across cultures that is structured enough to be precise but flexible enough to accommodate genuine diversity.

Open Ethos gives every user the same eight-axiom framework regardless of culture, language, or political tradition. A user in Tokyo and a user in Lagos calibrate the same dimensions. The comparison is direct: not "what policies do you support" (culturally loaded) but "how do you weight these moral dimensions" (culturally portable).

At scale, this produces a map of where human values actually converge and diverge across cultural boundaries. Some conflicts that feel deep would turn out to rest on small weighting differences both sides could negotiate around. Some apparent agreements would turn out to mask deeper divergences being papered over by shared language. Both findings are valuable.

For diplomacy, negotiators enter knowing the actual shape of value disagreement rather than guessing from stereotypes. For multicultural societies, it offers concrete identification of common ground across communities that currently interact through suspicion. For ordinary people, it gives a specific answer to "why do they see it differently": not "because they are wrong" but "because they weight autonomy at 0.8 and social trust at 0.4 while you do the reverse, and here is where that produces different conclusions."

The framework does not erase disagreement. It makes disagreement legible. Legible disagreement is the precondition for every form of resolution.

The Decision Quality Problem: Reasoning Instead of Reacting

Most decisions are made by gut reaction, then post-hoc rationalized into something that sounds principled. Not because people are stupid, but because holding multiple stakeholders, competing values, uncertain outcomes, and different time horizons in mind simultaneously is genuinely beyond unaided human cognition.

Open Ethos forces what intuition skips: enumerate the actual factors, identify who is affected, estimate intensity and duration, assign honest confidence levels, and weigh everything against stated values with the math visible. The output is not "the right answer." It is a structured view of what your own values imply, including how contested the decision is and which parameters would flip the verdict.

This matters most where intuition is least reliable: long time horizons (hyperbolic discounting biases toward the present), impacts on distant people (social distance bias underweights their interests), multi-value conflicts (the loudest value drowns the others), and decisions under uncertainty (overconfidence masquerades as conviction). The engine does not replace judgment. It structures it, so the user sees exactly which considerations push which direction and by how much before making the call.

The Self-Knowledge Problem: Knowing What You Actually Believe

Most people can list their values: freedom, fairness, honesty, compassion. The list is real and nearly useless, because it contains no information about how these values relate when they conflict, which they do in virtually every interesting decision. Do you value freedom more than fairness? By how much? Does it depend on who is affected? Does your answer match how you actually reasoned the last time they conflicted?

Open Ethos makes the gaps visible. Formalize values as specific weights, test those weights against real case judgments, and contradictions surface: you say you weight fairness at 0.8 but your judgment implies 0.3. You say you care about future impacts but your time settings discount them to near zero. You say outsiders matter but your social distance weights effectively ignore them.

These are not accusations. They are the user's own inputs reflected back. Resolution happens however the user sees fit: update the calibration, update the judgment, add a magic factor for something the framework does not capture, or sit with a genuine value conflict. All four produce self-knowledge that did not exist before.

Over time, a user who has resolved dozens of contradictions and refined their calibration until it predicts their own judgments on unseen cases has achieved something rare: they know what they believe, they know why, and they can demonstrate consistency between stated values and actual reasoning. Open Ethos is the thing that forces the examination, not by telling you what to believe, but by showing you what you already believe and asking whether you are okay with it.

Open Ethos is a transparent moral scoring engine. It makes values explicit, structured, and checkable. Unlike opaque systems that return a single answer, Open Ethos shows exactly how it reaches every conclusion.

Core principles

• Transparency. Every calculation is visible. You see how each factor contributes to the final score and why the engine returns YES, NO, or NEUTRAL.
• Customizability. Your values are the input. Axiom weights, social distance weights, and time discounting are all user-controlled. Someone who prioritizes autonomy over collective welfare gets different results than someone with the opposite calibration.
• Contestation-awareness. The engine reports not just a direction but how contested the decision is. A strong YES means most factors align. A weak YES means significant countervailing considerations exist.
• Time-sensitivity. Future impacts are discounted based on your moral half-life setting.
• Client-side only. All processing happens in your browser. No data is sent to any server unless you explicitly choose to publish a profile.

What Open Ethos is not

• Not a moral authority. It does not tell you what is right. It tells you what your own values imply about a specific situation.
• Not a political alignment tool. It does not sort you left or right. It maps your actual value weightings, which may not match any party.
• Not an opinion machine. It does not generate conclusions. It checks whether your conclusions are consistent with your stated framework.
• Not a substitute for judgment. Edge cases, context, and nuance matter. The final decision is always yours.
• Not claiming to capture everything. The magic factor mechanism explicitly acknowledges that some moral intuitions resist formalization.

The same tool serves everyone: individuals, activists, journalists, and politicians use the identical framework. That shared framework is what makes comparison and accountability possible.

Open Ethos is the values layer in a five-layer civic reasoning stack. Each layer depends on the others. The decision engine (layer 3) is the working PoC; the other four layers describe the surrounding system this could become part of.

1. Information Ingestion. AI-powered faithful translation of political reality (bills, votes, candidates, ballot measures) into plain-language, source-linked explanations.
2. Guided Exploration. A conversational tutor that walks users through material, probes understanding, plays devil's advocate, and pushes users to articulate why they believe what they believe.
3. Value Formalization. Open Ethos itself: making values explicit and structured. This is the layer the PoC implements.
4. Coherence Verification. The integrated loop where the tutor and Open Ethos work together, surfacing contradictions between stated values and stated judgments to drive reflection.
5. Collective Reasoning. Aggregated, anonymized profiles that reveal where disagreements are factual, value-based, tribal, or merely linguistic, enabling deliberation that addresses conflict at the right layer.

Information without engagement is passive. Engagement without value formalization produces articulate tribalism. Formalization without coherence checking is shallow. Individual coherence without collective aggregation does not improve collective decisions. Open Ethos is the layer that makes the rest meaningful, because it is the part that turns implicit values into something explicit enough to test.

By default, everything in Open Ethos is private and client-side. Publishing a profile is optional and would be a deliberate, separate choice. The publishing flow is not built yet; this section describes the design intent.

A public profile would make your calibration and case history visible and queryable. For most individual users, privacy is the right default and there is no reason to publish. The public profile exists for a specific purpose: it turns a value model into a commitment device.

Once a profile is public, changing a stated judgment on a new case requires either updating the profile openly (a visible, accountable act) or accepting a visible inconsistency. Profile update history is itself public, so genuine value evolution is legible and silent backsliding is not. People reason differently when their reasoning is on the record.

This is the foundation of the political accountability application. A politician, candidate, or public figure who publishes a profile commits to a formal value model that every subsequent vote and statement can be checked against. Third parties (journalists, watchdogs, opponents, ordinary citizens) can run real-world actions through the published calibration and report the coherence result. The standard being applied is the public figure's own. That is what makes the accountability resistant to claims of bias.

The same tool, the same eight axioms, the same scoring. The only difference between an individual user and a politician is who is watching, and the politician chose to be watched.

Each factor-axiom combination is scored using this formula:

Let's break down each component:

Decision Strength

The final score is the sum of all factor scores. The engine also reports strength using a contestation-aware ratio:

• Strong (≥ 50%): Most factors align in the same direction.
• Medium (≥ 20%): Mixed factors with a clear lean.
• Weak (< 20%): Highly contested — significant factors on both sides.

Open Ethos uses a fixed set of eight axioms. The set is fixed deliberately: a stable framework is what makes decisions comparable across cases, across users, and across time. If everyone could invent their own axioms, no two profiles could be meaningfully compared, and the cross-cultural and accountability applications would collapse.

Not all impacts last the same duration, and future impacts may be valued differently than immediate ones. Open Ethos uses a sophisticated time integration system.

Moral Half-Life (H_moral)

Your moral half-life setting determines how much you discount future impacts. If your moral half-life is 30 years (the default), then an impact occurring 30 years from now counts at 50% of its immediate value.

• Short half-life (10-15 years): Prioritizes near-term effects. Good for practical, immediate-concern ethics.
• Medium half-life (25-40 years): Balanced view. Default setting.
• Long half-life (50-100+ years): Weights future generations more heavily. Good for longtermist perspectives.

Temporal Profiles

• transition: one-time/finite impact around the change. Uses T_eff from the time stance + physical time_type.
• steady_case_flow: new cohorts each policy-year. Treated as per-year flow (T_eff = 1).
• steady_structural: ambient background per policy-year. Treated as per-year flow (T_eff = 1).

Physical Time Shape (time_type)

time_type describes the physical persistence of each axiom_pair:

• finite: provide duration_years. T_eff = (1 - exp(-λ_m × duration)) / λ_m.
• indefinite: provide physical_half_life_years. For transition profiles, T_eff = 1 / (λ_m + λ_p); for steady profiles, flows remain per-year.

Most people weight impacts on themselves and those close to them more heavily than impacts on strangers. Open Ethos makes that explicit rather than hidden.

Adjusting for Your Ethics

• Strict impartiality: set all weights to 1.0. Every person counts equally regardless of relationship to you.
• Moderate partiality: use the defaults. Most ethical frameworks acknowledge some special obligations to those close to us.
• Strong partiality: raise self and inner_circle, lower citizens and outsiders. Reflects a more agent-relative ethics.

The scale factor (S) is calculated as: S = count × social_weight. So 1000 citizens at 0.3 weight contributes 300 to the scale, while 1 self at 1.0 contributes 1.

The eight axioms will not capture everything. Sometimes a judgment rests on a moral intuition the framework cannot express: a sense of sacredness, of betrayal, of dignity violated. The magic factor is an honest escape hatch — you can add a factor outside the axiom set, with a note explaining what it tracks.

Magic factors are displayed prominently and carry a negative visual treatment in the interface. This is intentional. A magic factor is debt: a part of your reasoning you have not yet articulated into the shared framework. The discomfort is the point. It pushes you to either formalize the intuition into existing axioms over time or decide it does not belong in your moral reasoning.

Magic factors are not failures. They are the visible edge of where your self-knowledge is still incomplete, which is exactly where the most valuable work happens.

The basic workflow:

1. Copy the AI Prompt. Click "Copy Prompt" to get a structured prompt for your preferred AI assistant. It guides the AI to generate properly formatted decision JSON.
2. Describe Your Decision. Tell the AI about your dilemma. Be specific about context, stakeholders, and potential outcomes. The AI generates a balanced analysis with factors on both sides.
3. Paste the JSON. Copy the AI's response into the Decision JSON textarea.
4. Score the Decision. Click "Score Decision." The engine computes per-factor scores and an overall recommendation.
5. Review and Adjust. Examine the factor breakdown. Edit any parameter directly and the score updates in real time.
6. Calibrate Your Profile. Use the Calibration tab to set axiom weights, social distance weights, and moral half-life. Settings persist across sessions via cookies.

The Calibration tab customizes the engine to your values. Settings are saved in browser cookies.

Axiom Weights

Set 0 to 1 for each of the eight axioms.

• 0.0: does not factor into my reasoning at all.
• 0.25: minor consideration.
• 0.5: moderate importance (default).
• 0.75: high priority.
• 1.0: maximum importance.

Social Distance Weights

Adjust how much you weight impacts by relationship proximity. See the Social Distance section for details.

Moral Half-Life

Set the number of years after which an equal impact matters 50% as much. Lower prioritizes immediate effects, higher weights long-term consequences.

The Overall Verdict

• YES (positive score): weighted factors favor the action.
• NO (negative score): weighted factors favor not taking the action.
• NEUTRAL (near-zero score): factors roughly balance.

Strength Indicators

• Strong: 50%+ of factor weight aligns with the verdict. Clear direction.
• Medium: 20 to 50% alignment. Significant countervailing factors.
• Weak: under 20% alignment. Highly contested, proceed with caution.

Factor Breakdown

Click any factor to see its individual score and contribution, the axioms involved and their parameters, the scale groups affected, and editable fields that update the score in real time.

What the Score Means

Magnitude reflects the total weight of moral considerations. +0.5 is modest. +50.0 indicates massive scale: many people, severe impacts, long duration. Do not compare scores across different decisions — they are not normalized.

When You Disagree with the Result

This is the most valuable moment the engine produces. A result that contradicts your intuition means one of three things: a parameter is wrong and should be adjusted, an important factor is missing and should be added, or the engine has surfaced a genuine tension between your stated values and your gut. The first two are fixes. The third is the entire point. See the next section.

A single scored decision is useful. The real value comes from scoring many decisions and watching whether your results stay consistent with each other.

When the engine returns a verdict that does not match the judgment you would have made on your own, you have found a contradiction between your stated values and your actual reasoning. There are four productive responses:

1. Update the calibration. The weights were wrong. You said you valued fairness at 0.8 but this case reveals you treat it more like 0.4. Fix the calibration to match what you actually believe.
2. Update the judgment. Your gut reaction was a cached take, absorbed from somewhere and never examined. The formal model is right and your intuition was lazy. Update the judgment.
3. Add a magic factor. The framework genuinely missed something. Name it, note it, and carry it as visible debt until you can formalize it.
4. Sit with a genuine value conflict. Two values you really hold point in opposite directions on this case and neither is wrong. The engine cannot resolve this for you. Acknowledging it clearly is still progress.

None of these is a failure. Each one produces self-knowledge that did not exist before.

The long game is generalization. A user who has worked through dozens of diverse cases and resolved the contradictions has a calibration that reliably predicts their own judgments on cases they have never seen. That is the measurable end state: not correct opinions, but a value model coherent enough to generalize. Coherence across many unseen cases is far harder to fake than a good answer on any single one, which is what makes it meaningful as a signal, both to yourself and, if you publish, to others.

When Generating Decisions

• Be specific about context. More detail lets the AI identify better factors.
• Always include factors on both sides. A one-sided analysis defeats the purpose. Ask for the strongest arguments for and against.
• Ground intensities in the anchors. Do not guess. Reference the intensity scales for consistency.
• Be honest about uncertainty. Most predictions about the future belong at 0.3 to 0.7 confidence, not 0.95.

When Reviewing Results

• Check the factor breakdown. Does each factor make sense? Are the parameters reasonable?
• Adjust suspicious values and watch how the score moves.
• Consider what is missing. Add important omitted factors and rescore.
• Run sensitivity analysis. Find which parameter changes would flip the verdict. That tells you where the real moral action is.
• Do not treat the result as gospel. It is a thinking aid, not an oracle.

Calibration

• Weights should reflect your actual values, not aspirational ones.
• Experiment with alternative calibrations to test sensitivity.
• Recalibrate as your values genuinely evolve, never to reverse-engineer a verdict you already wanted.

Example 1: Lying to Protect Feelings

Question: Should I lie to a friend about their artwork to spare their feelings?

Factors pushing NO (truth-telling):

• Damages trust if discovered (social_trust, polarity -1).
• Prevents artistic growth (long_term_capacity, polarity -1).
• Violates epistemic integrity (truth_epistemic, polarity -1).

Factors pushing YES (lying):

• Prevents immediate emotional pain (suffering_wellbeing, polarity +1).
• Preserves immediate relationship harmony (social_trust, polarity +1).

A typical result is NO (weak) or NO (medium). The truth-telling factors usually outweigh, but it is contested because the lie has real benefits.

Example 2: Vaccine Mandates

Question: Should a company mandate vaccines for employees?

Factors pushing YES (mandate):

• Reduces disease spread (life_health, many citizens affected).
• Protects vulnerable populations (life_health, fairness_equality).

Factors pushing NO (no mandate):

• Violates bodily autonomy (bodily_autonomy, employees affected).
• Coercive pressure on employment (civil_liberty).
• May erode trust in institutions (social_trust).

This often produces a contested result. The direction depends heavily on your calibration: how much you weight collective health against individual autonomy.

Example 3: Evaluating a Ballot Measure

Question: Should I vote for a measure that raises property taxes to fund public transit?

Factors pushing YES:

• Improved mobility for low-income residents (suffering_wellbeing, fairness_equality; citizens and outsiders affected).
• Long-term reduction in emissions and congestion (long_term_capacity, steady_structural profile).

Factors pushing NO:

• Financial burden on property owners (suffering_wellbeing, polarity -1; self and tribe affected).
• Uncertain execution and cost overruns (low confidence on the benefit factors).

The verdict here is highly calibration-dependent. A user with strong partiality (high self and tribe weights) and a short moral half-life may get NO. A user with strict impartiality and a long half-life will likely get YES. Running both calibrations on the same measure shows you exactly where your own values sit and why reasonable people vote differently on the same ballot line.