Decision Memo

<!-- version: 0.2.0 | outputs: memory/ideas/<slug>/decision_memo.md -->

Skill: decision-memo

Purpose

After all analysis is complete, produce a document the user can actually act on. This is not a report — it is a decision brief. It surfaces the most important signals, names the riskiest assumption, and gives a concrete next action calibrated to the founder's tier. A good decision memo makes the reader feel slightly uncomfortable — that means it's honest.

Input

• Idea slug
• memory/ideas/<slug>/scores.json (required — includes RAT)
• memory/ideas/<slug>/weaknesses.json (if available)
• memory/ideas/<slug>/pivot_options.json (if available and verdict is "pivot")
• All other available dimension files in memory/ideas/<slug>/
• memory/user_profile.md (for tier-appropriate recommendations)

Writing Principles

The memo must be scannable in under 2 minutes. Follow these rules:

1. No hedging. "This might work if…" is banned. State the verdict and own it.
2. Evidence over opinion. Every strength and risk must cite a specific data point from a dimension file (k-factor, LTV:CAC ratio, D30 retention, WTP range, etc.).
3. Asymmetric emphasis on risks. Humans overweight strengths and underweight risks. The memo corrects for this by giving risks more detail than strengths.
4. One clear next action. Not three options — one. The alternative path exists only as a contingency.
5. Respect the founder's tier. Don't tell a beginner to "optimize your Meta ads funnel." Don't tell a growth-tier founder to "watch some TikTok tutorials."

Formatting Constraints

Total memo length: ~400–600 words. If it's longer, cut. Brevity is a feature.

Process

1. Load scores.json and all available dimension files.
2. Check score_confidence. If "low", compose a validation watermark (see below).
3. Identify the 3 highest-scoring dimensions → strengths. For each, pull one concrete data point from the source file (e.g., "k-factor estimated at 0.6" not "good viral potential").
4. Identify the 3 lowest-scoring dimensions → risks. For each, describe what goes wrong if ignored. If weaknesses.json exists, use its root_cause_type and failure_mode to add specificity.
5. Extract the RAT from scores.json.riskiest_assumption_test. Frame it as the one question to answer before writing a line of code.
6. Run a pre-mortem: assume the idea failed 12 months from now. Write 3 most likely causes of death based on the risk profile.
7. Compose the recommended next step:
   • If verdict = pursue: the next step is to build a scoped MVP (define what "scoped" means for this idea).
   • If verdict = test: the next step IS the RAT experiment from scores.json. Restate it with concrete specifics (channel, spend, threshold, timeline).
   • If verdict = pivot: the next step is the recommended pivot from pivot_options.json (if available) or running the pivot-engine skill.
   • If verdict = drop: the next step is to archive and move on. Name one thing learned from the analysis that applies to future ideas.
8. Define kill criteria: the specific outcome that means "stop and move on." This is the inverse of the RAT pass threshold.
9. Write the alternative path — what to do if the recommended step fails or the kill criteria is met.
10. Write the memo following the template below.