Regret Minimization

SKILL.md

---
name: regret-minimization
description: "Activate when: user says 'what would my 80-year-old self think', 'if I never try this will I regret it', 'I keep hesitating but the math is clear', 'regret minimization', or faces a major hard-to-reverse life decision (career pivot, founding a company, relocation, leaving a relationship) where the units are joy/meaning/identity rather than money.
  Do NOT activate when: the decision is routine or easily reversible (which restaurant, which tool to buy); or when money/EV is genuinely the right unit (portfolio allocation, pricing math)."
---

# Regret Minimization

## Overview

Most frameworks optimize *expected value* — pick the highest probability-weighted payoff. That math breaks down for large, asymmetric, life-defining choices where the true unit is not money: career pivots, founding decisions, relocations, relationship exits. There, the decisive question is not "what's the expected payoff?" but **"which regret will I be unable to live with at 80?"**

Associated with **Jeff Bezos** (D.E. Shaw → Amazon, 1994); rooted in Stoic *Premeditatio Malorum* (Seneca, *Epistulae Morales* 91) and formalized in Regret Theory (Loomes & Sugden, *The Economic Journal*, 1982).

**Compose with neighbors:** use [first-principles](../first-principles/SKILL.md) to clarify what is at stake; [inversion](../inversion/SKILL.md) to surface failure modes; [second-order-thinking](../second-order-thinking/SKILL.md) to verify downstream consequences; then use regret minimization to choose when EV analyses come out close and the true cost is psychological.

## When to Use

Apply when: decision is **major, hard-to-reverse, asymmetric** (career pivot, founding, relocation, relationship, children); EV math feels insufficient because **units are joy/meaning/identity**; hesitation is emotional, not analytical; user says "what would my 80-year-old self think?", "if I never try this will I regret it?", "I keep hesitating but the math is clear."

**When NOT to use:** routine reversible decisions; EV is genuinely the right unit (portfolio allocation, pricing); framework being re-run weekly (procrastination); both regrets are unlivable (redesign the choice instead).

## Coaching Novices (Adaptive Front Door)

- **Engine mode:** user has a concrete major decision → run The Process directly.
- **Coach mode:** user is unfamiliar or has no concrete case → guide step by step.

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.
1. One-line what-it-is: for big life decisions where math is close or doesn't fit, project to **age 80** and ask *which path will haunt you more*. Commit to the option whose regret you can live with.
2. Check fit against When to Use / When NOT to use — if it doesn't fit, say so and point elsewhere.
3. Elicit their real decision. Ask for one if missing; never run the audit on a hypothetical.
> **[WAIT — do not advance until user responds]**
4. Walk The Process one step per turn: describe 80-year-old self → name each path's regret → test asymmetry. Never project for them.
> **[WAIT — do not advance until user responds]**
5. Close by naming the regret *they* identified as harder to live with, plus the commit date.
> **[WAIT — do not advance until user responds]**

## The Process

Run the **Regret Audit**. Project, name the regret on each path, test asymmetry, commit.

1. **Define the fork specifically.** Not "should I change my life" — "leave job to start X by date vs stay 2 years and revisit." Vague forks produce vague regrets.
2. **Project to 80.** Describe yourself looking back: what life, what values, who's around. The age-80 self is your *evaluator function*.
3. **For each option, imagine the path lived realistically.** What did you do, learn, not do? Avoid heroic versions.
4. **Name the regret on each path — split failed-but-tried from never-tried.** Failed regret is usually lighter; test this for *your* case.
5. **Test asymmetry — which regret is harder to live with?** Load-bearing step. Most people know the answer once they ask the right question.
6. **Name the unit being optimized:** joy, meaning, security, identity, contribution, autonomy. Step 5's answer should match the unit you actually weight most.
7. **Commit and pre-bind.** State the date the decision is irreversible — after which you stop relitigating.

### Output: the Regret Audit

```
Decision: Option A vs Option B (concrete, time-bounded)
Age-80 self: <one paragraph — what life, what values looking back>
Path A lived (realistic): <what I did / learned / didn't do>
Path B lived (realistic): <what I did / learned / didn't do>
Regrets — A failed: "..." | A never tried: "..."
Regrets — B failed: "..." | B never tried: "..."
Asymmetry: which regret is harder to live with at 80, and why?
Unit optimized: <joy / meaning / security / identity / autonomy>
Commitment: I commit to <option> by <date>. Stop relitigating after that.
Re-open trigger: only if <specific observable evidence> before <date>.
```

*→ Method in Action: [Jeff Bezos, D.E. Shaw → Amazon (1994)](examples/jeff-bezos-de-shaw-amazon-1994.md)*

## Regret Packs

The Audit runs identically across domains; what differs is the typical load-bearing regret and the unit optimized.

- **Career pivots / founding:** load-bearing = *never having tried*; discount "I'll have less money than peers" (vivid at 35, faint at 80).
- **Family formation:** structurally asymmetric; discount "slower career progression for a decade."
- **Geographic relocation:** load-bearing = *staying somewhere that quietly drained you*; discount "I'll miss old friends."
- **Sunk investments (PhD, long relationship, failing startup):** load-bearing = *staying past the point you knew*; discount sunk-cost framing.

## Applying It Well

- **Project, don't fantasize.** Imagine the ordinary version of each path — successful-A vs worst-B corrupts the audit.
- **Failed-but-tried regret is usually lighter than never-tried — but verify for *your* case.** Often true, not universally true.
- **The unit matters more than the math.** Most failed audits asked the right question against the wrong unit.
- **Pre-commit or it didn't happen.** Re-running weekly is postponement, not decision-making.

*→ Primary sources: [references/sources.md](references/sources.md)*

## Common Rationalizations

**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**

| Fake move | Reality |
|---|---|
| [D] Using regret minimization for **trivial decisions** | Theater, not analysis. The framework is for large, asymmetric, hard-to-reverse decisions. |
| [D] **Asymmetric application** | Running regret minimization for the option you want; running EV for the option you don't. Both options must use the same audit in the same units. |
| [D] Treating "regret" as **emotional discomfort** | Real regret haunts you at 80 — not passing disappointment. Mild annoyance = wrong tool. |
| [D] Conflating **never-tried** with **tried-and-failed regret** | Different magnitudes. The whole framework rests on telling them apart. |
| [D] Mistaking it for **EV optimization** | Different objective functions that often disagree. Picking regret because EV gives an unwelcome answer is rationalization. |
| [D] Importing **someone else's 80-year-old self** | Your future self is not Jeff Bezos's. Build your *own* age-80 evaluator first. |
| [D] Failing to **pre-commit** | Getting the answer then re-litigating weekly. The point is to commit and stop the inner argument. |
| [D] **Fantasy projection** | Imagining the successful version of A and the worst version of B. Both must be imagined at realistic outcomes. |
| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |

## Red Flags

- Decision is reversible or low-stakes; options not specified with concrete actions and dates
- One option imagined heroically, the other realistically
- "Regret" weighed is present-day annoyance, not haunting-at-80 regret
- Audit re-run multiple times with no commitment date; framework used to justify a decision already made
- No unit named — "asymmetry" comparison is across categories
- Age-80 self borrowed from someone famous, not the user's own values

## Verification

- [ ] Specific, time-bounded fork defined; age-80 self described with values explicit
- [ ] Each path imagined *realistically* (including the boring middle)
- [ ] Both failed-but-tried and never-tried regrets named for each option
- [ ] Asymmetry test answered; unit optimized named; commitment date set

---

*Part of **deciqAI Knowledge Skills** — open-source thinking skills that make rigor executable for AI agents. Built by deciqAI · https://deciqai.com · Contributions welcome — see the template at the repo root.*