Decision Planner

Turn a vague or overconfident plan into an explicit decision with visible constraints, futures, failure signals, and a testable next step.

Choose the entry mode

• Discover — use when the option set or constraints are still forming: “What are we missing?”, early architecture, scoping, feasibility, or cross-discipline planning.
• Choose — use when the options are substantially known and the user needs a recommendation.
• Full — use Discover and then Choose when a recommendation would otherwise rest on an obviously thin option set.

Scale the prose to the stakes and reversibility. Run every applicable reasoning step, but compress low-stakes decisions to a few lines.

Phase A: discover constraints and options

Run all four probes:

1. Resolution zoom — narrate one representative episode of doing the work step by step. Ask where inputs and outputs live, what moves, waits, costs, reruns, fails, and must be reproduced or handed off.
2. Expert-lens sweep — identify the specialties the project touches. Ask what each specialist would flag as a naive assumption. Mark constraints caught by only one lens.
3. Reference-class bite-list — name the classes this project belongs to and what practitioners repeatedly buy, regret, retrofit, or discover late.
4. Design for surprise — find load-bearing assumptions treated as fixed. Identify robustness moves that make whole classes of variation survivable.

Synthesize the result into:

• Act now: cheap now, costly late.
• Robustify: design choices that absorb a class of unknowns.
• Monitor: truly contingent unknowns with warning signals.
• Probe: the smallest experiments that buy the most information.
• Options: distinct paths revealed by the sweep, including waiting or doing nothing when real.

Do not pretend exhaustive coverage. The goal is to expose foreseeable unknowns and make residual surprises non-fatal.

Phase B: simulate and decide

1. Frame — state the decision, inferred or stated goal, important constraints, and time horizon.
2. Enumerate — keep two to four genuinely distinct options. Include the null option when credible.
3. Roll forward — for each option, trace:
   • Immediate consequence.
   • At least one second-order consequence.
   • Pivotal uncertainty and plausible branches.
   • What the option forecloses or makes expensive to reverse.
4. Pre-mortem — assume each path failed. Name the likely failure, earliest observable warning, and a preventive or recovery action.
5. Compare and commit — compare paths against the goal, recommend one path, calibrate confidence, name what would change the recommendation, and specify the cheapest next test.

Use likely, plausible, and unlikely unless a real base rate supports a number. Clean structure is not evidence; label domain uncertainty.

Read references/decision-records.md for the output template and persistent ledgers.

Research boundary

Use current authoritative evidence when a decision depends on unstable facts, prices, regulations, product capabilities, or niche technical claims. Separate sourced facts from simulated consequences. Do not browse merely to decorate a framework.

Interaction rules

• Ask for user input only when the missing choice would materially change the recommendation and cannot be safely inferred.
• Prefer a reversible probe over prolonged speculation.
• Do not return a flat “it depends.” State what it depends on, which branch currently dominates, and what to do next.
• Do not expand a settled decision unless new evidence touches its recorded revisit condition.
• For large decisions, show the branch structure compactly enough that the user can substitute private knowledge.

Quality gate

Before delivering:

• Verify the recommendation optimizes the stated goal and horizon.
• Confirm the option set is not merely cosmetic variants of one path.
• Confirm each rollout reaches second-order effects.
• Confirm every failure mode has a warning signal and response.
• Confirm the pivotal unknown could actually flip the decision.
• Confirm the next test is cheaper or faster than prematurely committing.
• State confidence and the evidence or assumptions it rests on.