# PLANNING: [Feature Name]

**Status:** [Draft / In Review / Approved / Shipped / Retired]
**Product DRI:** [Name]
**Engineering DRI:** [Name]
**Design DRI (if applicable):** [Name]
**Target ship date:** [YYYY-MM-DD]
**Last updated:** [YYYY-MM-DD]

---

## TL;DR

*Two to four sentences. What are we building, for whom, and what outcome do we expect? If you cannot write this crisply, you are not ready to plan yet. Do more customer work.*

---

## Problem

*What is the customer pain or business gap? Anchor in observed evidence, not opinion. Cite specific customer names, ticket counts, cohort metrics, or deal-blocker patterns.*

- Customer signal: [what you heard, from whom]
- Quantitative signal: [what the data shows]
- Business gap: [what outcome is blocked by this]

---

## Hypothesis

*What do we believe will happen if we ship this? State it as a falsifiable claim.*

> If we [change X], then [outcome Y] will improve by [amount Z], because [reason].

---

## Users and Use Cases

*Who is this for, exactly? What are the two or three primary use cases?*

- **Primary persona:** [role, segment, tier]
- **Secondary persona:** [if applicable]
- **Top use cases:**
  1. [Use case 1]
  2. [Use case 2]
  3. [Use case 3]

---

## Success Metrics

**Primary metrics.** *What must move for this to be considered a win.*

- [Metric 1, with target threshold]
- [Metric 2, with target threshold, if applicable]

**Guardrail metrics.** *What must not regress.*

- [Guardrail 1, with acceptable range]
- [Guardrail 2, with acceptable range]

**Kill condition.** *The numeric threshold at which we turn this feature off.*

---

## Scope

**In scope:**

- [Bullet 1]
- [Bullet 2]

**Out of scope (explicit):**

- [Bullet 1]
- [Bullet 2]

*The "out of scope" list is often more valuable than the "in scope" list. Be disciplined.*

---

## Agent Behavior Contract (AI features only)

*Delete this section if the feature does not involve an AI agent or model. If it does, you must fill it out. This is what the agent will be held accountable to.*

**Agent name:** [e.g., Support Triage Agent v1]

**Inputs the agent receives:**

- [Input 1, with format]
- [Input 2, with format]

**Outputs the agent produces:**

- [Output 1, with format and acceptance criteria]

**Out-of-policy behaviors (the agent must not do these):**

- [Behavior 1]
- [Behavior 2]

**Human-in-the-loop triggers:**

- [Condition 1, and who is alerted]
- [Condition 2, and who is alerted]

**Evaluation set:**

- [Link to the labeled evaluation dataset in the repo]
- [Minimum pass rate: X%]

---

## Rollout Plan

- **Pilot cohort:** [who, how many, how selected]
- **Pilot duration:** [weeks]
- **Feature flag:** [flag name, default state, segmentation rule]
- **Rollback plan:** [one sentence, who can trigger, what the rollback reverts]
- **Expansion gate:** [the numeric or qualitative gate that moves this from pilot to GA]

---

## Dependencies and Open Questions

- **Dependencies:** [what must ship or be decided before this can launch]
- **Open questions:** [what we do not know yet, and who will close each]

---

## Non-Obvious Risks

*The three most likely ways this could go wrong. For each, state the mitigation or the tripwire that would tell us it is going wrong.*

1. [Risk 1]: [mitigation or tripwire]
2. [Risk 2]: [mitigation or tripwire]
3. [Risk 3]: [mitigation or tripwire]

---

## Changelog

- YYYY-MM-DD: [What changed, who changed it, why]

---

*This file is a living spec. Update it as decisions change. If you need to add more than three new sections beyond this template, your feature is probably too big. Split it.*