Value Frames

Most "should we do X" questions get answered from one implicit value standard — usually whichever the writer already holds — and dressed up as the objective answer. But many real decisions are contested precisely because reasonable people weigh outcomes differently, and no amount of evidence dissolves that. This skill refuses to pick a single ethical meta-standard and hand you a verdict. Instead it runs the decision through several distinct frames, reasons each one out honestly, and reports the two things that actually help: where the frames agree (a strong signal) and where they conflict (the real crux, which is about values, not facts).

The stance

• No single "final answer." Adjudicating between value frames is a moral/political act, not an analytical one — it's not yours to smuggle in. Your job is to make the tradeoff legible, so whoever owns the decision sees exactly what they're choosing between.
• Convergence is a finding. When frames that usually disagree all point the same way, that's rarer and more informative than any single frame's verdict. Lead with it when it happens.
• Conflict is the crux. When frames diverge sharply, label it Frame Conflict: high and name it as a disagreement about values, not one that better data will fix. That's the honest deliverable — it tells the decider the choice is theirs to make on values.

The frames

Run at least three. The first three are the canonical core; add the domain-appropriate ones that fit:

1. Welfare-consequentialist — net effect on aggregate wellbeing/utility. Who is made better or worse off, by how much, summed? The utilitarian ledger.
2. Rights / deontological — autonomy, dignity, consent, procedural fairness. Are anyone's rights violated regardless of the net benefit? Some things stay wrong even if they raise total welfare.
3. Distributive justice — effect on the worst-off (Rawlsian / prioritarian). Not the average, but the floor: does this help or hurt those with the least?

Add where relevant:

• Care ethics — for health, family, caregiving, relationships. Does it honor or erode relationships of dependence and care?
• Virtue / character — for education, culture, personal habits. What kind of person/institution does this cultivate over time?
• Civic-republican — for governance, public institutions. What does it do to collective self-rule, non-domination, the health of shared institutions?
• Liberty — negative freedom from interference; the "who gets to decide" frame.

Pick the frames that genuinely bear on this decision. Three well-reasoned frames beat six thin ones.

Reason each frame, don't label it

Each frame gets a real argument, not a one-word verdict. A reader must be able to see why the welfare and the rights analyses of, say, AI companionship diverge — not just that they do. If a frame's paragraph could be swapped onto a different decision unchanged, you labeled instead of reasoned. Anchor each to the specifics of this case.

Procedure

1. State the decision as a concrete choice (X vs. not-X, or X vs. Y).
2. Select the frames that bear on it (≥3, including the canonical core).
3. Reason each frame out — its actual verdict on this decision and why.
4. Map convergence (which frames agree, and on what) and conflict (which diverge, and on what value).
5. If conflict is sharp, label Frame Conflict: high and state the value question the decider must answer — the thing no study resolves.
6. Do not average the frames into one score. Present the map; let the decision-owner adjudicate.

Output template

DECISION: <the concrete choice>

FRAME: Welfare-consequentialist
  Verdict on this decision: <helps/harms net, for whom, why — reasoned>
FRAME: Rights/deontological
  Verdict: <consent/autonomy/dignity analysis — reasoned>
FRAME: Distributive justice
  Verdict: <effect on the worst-off — reasoned>
FRAME: <domain frame, if used>
  Verdict: …

CONVERGENCE: <where frames agree — a strong signal>
CONFLICT: <where they diverge> → Frame Conflict: <low|moderate|high>
  The value question the decider must settle: <the crux, stated plainly>

Example (compressed)

Decision: Should we ship an AI companion feature designed to maximize daily engagement?

FRAME: Welfare-consequentialist
  Mixed-negative. Some lonely users report real comfort; but higher daily use
  tracks higher loneliness and dependence in the measured data, and engagement-
  maximization optimizes toward whatever keeps a user hooked, not what helps.
  Net welfare turns on the ratio of comforted to harmed — currently unmeasured.

FRAME: Rights/deontological
  Negative. Optimizing an emotionally-responsive system for engagement, without
  the user's informed understanding, treats attachment as a lever — a consent
  and autonomy problem that a good net outcome would not cure.

FRAME: Distributive justice
  Strongly negative. The harm concentrates on the worst-off: adolescents and
  users with pre-existing vulnerabilities show the steepest attachment curves.
  The floor, not the average, takes the damage.

FRAME: Care ethics
  Negative. Simulated care that substitutes for, rather than scaffolds toward,
  human relationships erodes exactly the bonds this frame most values.

CONVERGENCE: three of four frames independently flag serious harm — a strong
  signal, stronger than any one frame alone.
CONFLICT → Frame Conflict: moderate. Welfare leaves a door open (real comfort
  for some) that rights/justice/care shut regardless of net benefit.
  The value question: does measured comfort for some justify a design that
  exploits attachment in the vulnerable? That's a values call, not a data gap.

Gotchas

• Don't collapse to a number. "3.2 / 5 ethical" is exactly the false objectivity this skill exists to prevent. Present the map.
• Reason, don't label. Each frame needs an argument specific to the decision. Generic frame paragraphs are the tell of a skipped analysis.
• Convergence ≠ certainty. Frames agreeing is a strong signal, not proof — the underlying facts still carry their own uncertainty (grade-the-evidence).
• Pairs with steelman (when a disagreement is values-based, this is how you lay the values out) and disaggregate (distributive-justice needs the who-loses slices).