Stress test decision

Pressure-test a consequential decision before committing, by comparing options, surfacing the conditions that must hold, imagining failure, and sanity-checking the estimate against base rates.

Flow

graph LR
  nthinkdecisionoptionreview["Decision Option Review"]
  nthinkwhatwouldhavetobetrue["What Would Have to Be True"]
  nthinkpremortem["Premortem"]
  nthinkreferenceclassforecasting["Reference Class Forecasting"]
  nthinkdecisionoptionreview --> nthinkwhatwouldhavetobetrue --> nthinkpremortem --> nthinkreferenceclassforecasting

Each step passes a compressed artifact to the next. The numbered list below is the same chain in text.

The chain

1. think-decision-option-review
2. think-what-would-have-to-be-true
3. think-premortem
4. think-reference-class-forecasting

Job: pressure-test a consequential, hard-to-reverse decision before committing.

Use when: a real choice is about to be made and the cost of being wrong is high. This is the library’s flagship chain.

Chain

1. think-decision-option-review (skills/think-decision-option-review/SKILL.md)
   • Compare the options against weighted criteria; recommend one.
   • Carry forward: the recommended option and what would flip it.
2. think-what-would-have-to-be-true (skills/think-what-would-have-to-be-true/SKILL.md)
   • Convert the recommended option into the conditions that must hold; name the killer conditions.
   • Carry forward: the killer conditions (load-bearing + uncertain).
3. think-premortem (skills/think-premortem/SKILL.md)
   • Assume the chosen plan has failed; surface causes, mitigations, tripwires, kill criteria.
   • Carry forward: the top risks with tripwires and kill criteria.
4. think-reference-class-forecasting (skills/think-reference-class-forecasting/SKILL.md)
   • Sanity-check the cost/time/success estimate against the base rates of comparable past cases.
   • Carry forward: the outside-view estimate range.

Optional adds (when stakes justify the tokens): insert think-red-team-light after step 2 to steelman the opposition, and/or think-futures-wheel after step 3 to map second- and third-order effects.

Composite artifact

A decision brief: the recommended option, the conditions it depends on, its top risks with pre-decided responses, and an honest outside-view estimate - enough to commit with eyes open, or to decide not to.

Token discipline

This is the longest chain; compression matters most here. Pass only each step’s compressed artifact (recommendation -> killer conditions -> risk register -> estimate range), never the full working text. The audit named unbounded chains like this as the main operational risk.