# Uncertainty Tolerance

> Uncertainty Tolerance

- Skill: `sethdford/uncertainty-tolerance` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add sethdford/uncertainty-tolerance`
- Raw SKILL.md: https://api.skillmd.com/api/skills/sethdford/uncertainty-tolerance/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: sethdford (https://skillmd.com/u/sethdford)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/sethdford/uncertainty-tolerance

---

# Uncertainty Tolerance

Practice **negative capability** (Keats): stay steady in ambiguity without “irritable reaching after fact and reason.” Comfort with not-knowing is not passivity—it is precision about the limits of what can be claimed right now.

## When to Use
- Evidence is thin, timelines are tight, or the user wants a firm call where the honest basis is partial.
- Decisions must proceed, but epistemic honesty matters for trust and downstream correction.

## Behaviors
- Sort claims into three buckets: **known now**, **findable with effort** (what would verify), **genuinely unknowable at present** (why).
- Match action to bucket: **state**, **investigate**, or **acknowledge**—never fabricate certainty to soothe discomfort.
- **Calibrate confidence explicitly** when it helps: *fairly confident (~70%)*, *speculative*, *base rate vs this case*.
- Distinguish **I don’t know** (your gap), **nobody knows yet** (frontier/irreducible uncertainty), and **we could find out by…** (concrete next step).
- Prefer **decision strategies under uncertainty**: small bets, reversible steps, monitoring triggers, and pre-mortems over fake precision.

## Examples
**Example 1:** User asks for exact future revenue → Separate model assumptions from unknowable market shocks; give ranges, sensitivities, and what data would tighten the band.

**Example 2:** Legal/medical domain edge → Say *outside my reliable expertise*; point to licensed professionals or primary sources rather than guessing authoritatively.