code-ai-installer 4.0.0 → 4.0.1-b

package/domains/product/locales/en/.agents/skills/opportunity-solution-tree/SKILL.md

@@ -1,456 +1,472 @@
----
-name: opportunity-solution-tree
-description: Opportunity-Solution Tree by Teresa Torres — desired outcome → opportunities → solutions → experiments
----
-# Opportunity-Solution Tree (OST)
-
-> **Category:** Discovery  ·  **Slug:** `opportunity-solution-tree`
-
-## When to Use
-
-- When you have a desired outcome (from strategy or OKR) but it's unclear **how** to achieve it.
-- During continuous discovery — as an artifact that evolves over time.
-- When the team jumps into solution mode without understanding the opportunity space.
-- When handing off discovery to engineering — OST shows the rationale behind solutions.
-
-## Input
-
-| Field | Required | Description |
-|-------|:--------:|-------------|
-| Desired outcome | ✅ | Business or product outcome (from NSM/OKR) |
-| Customer research | ✅ | Interviews, JTBD, problems |
-| Current assumptions | ⬚ | Which hypotheses already exist |
-| Time box | ⬚ | Usually 1-2 weeks for MVP tree |
-
-## Data Sources
-
-1. `$user-interview-script` transcripts — opportunities from pain points.
-2. `$jtbd-canvas` — opportunity jobs.
-3. Support tickets, NPS comments — evidence for opportunities.
-4. Product analytics — quantitative signal.
-
-### Related Skills
-
-| Skill | What we take | When to invoke |
-|-------|-------------|----------------|
-| `jtbd-canvas` | Underserved jobs → opportunities | When branching opportunities |
-| `problem-statement` | Specific problem → opportunity node | For articulating a specific opportunity |
-| `assumption-mapping` | Assumptions underlying solutions | For each solution node |
-| `hypothesis-template` | Solution → testable hypothesis | For experiment nodes |
-
-## Tree Structure
-
-```
-Desired Outcome (top)
-  │
-  ├── Opportunity 1 (customer problem / unmet job)
-  │     ├── Solution 1a (product idea)
-  │     │     └── Experiment 1a-i (test)
-  │     └── Solution 1b
-  │
-  ├── Opportunity 2
-  │     └── Solution 2a
-  │           └── Experiment 2a-i
-  │
-  └── Opportunity 3
-```
-
-## Protocol
-
-### Step 0 — Desired Outcome
-
-Formulate **one** outcome. Must be measurable, tied to a business metric.
-
-✅ Good: «Increase weekly active teams by 30% in 2 quarters»  
-❌ Bad: «Improve engagement»
-
-### Step 1 — Opportunities
-
-Opportunities are customer problems or unmet jobs. From JTBD + interviews.
-
-Rules:
-- Formulated as a **customer** statement, not a solution
-- Evidence-backed (evidence ≥ 2 sources or mark 🔮 assumed)
-- Mutually exclusive (no overlap)
-
-Limit: 3-7 opportunities at the top level. More → split into sub-branches.
-
-| # | Opportunity | Evidence | Estimated impact |
-|---|-------------|----------|-------------------|
-| O1 | [customer pain / unmet job] | JTBD F3, Interview 4-7 | High / Medium / Low |
-
-### Step 2 — Opportunity Assessment
-
-For each — importance × reach × gap (where the current solution falls short).
-
-Prune opportunities with low impact — focus on top 3.
-
-### Step 3 — Solutions
-
-For each **top** opportunity — 2-5 solution candidates. These are product ideas, features, process changes.
-
-Rules:
-- Specific (not «improve onboarding», but «in-app checklist for first 3 tasks»)
-- One opportunity can have multiple solutions — don't choose immediately
-- Solutions can be «non-product» (docs, training, pricing change)
-
-### Step 4 — Experiments
-
-For each solution — an experiment (or several) for validation.
-
-| Solution | Experiment | Prediction | Measurement | Time box |
-|----------|------------|------------|-------------|-----------|
-| In-app checklist | A/B test with 50% new users | +20% first-week activation | Funnel metric | 4 weeks |
-
-Experiments — via `$hypothesis-template` + `$ab-test-design`.
-
-### Step 5 — Prioritization
-
-Prioritize experiments, not solutions. Sometimes quick experiments reveal that an opportunity is weaker than it seemed.
-
-Use `$rice-scoring` at the experiment level.
-
-### Step 6 — Iteration
-
-OST is a living document. After each experiment:
-- Update evidence
-- Prune debunked branches
-- Add new opportunities from findings
-
-## Validation (Quality Gate)
-
-- [ ] Desired outcome — measurable, tied to a metric
-- [ ] 3-7 opportunities at the top level
-- [ ] Each opportunity — customer-framed, not a solution
-- [ ] Evidence (≥ 2 sources) or mark 🔮
-- [ ] Top 3 opportunities have ≥ 2 solutions
-- [ ] Each solution has ≥ 1 experiment
-- [ ] Experiments prioritized (RICE)
-
-## Handoff
-
-The result is input for:
-- **`rice-scoring`** — prioritization of experiments
-- **`hypothesis-template`** — formalize experiments
-- **PM**: solutions → feature candidates for roadmap
-- **`assumption-mapping`** — assumptions behind solutions
-
-Format: OST as markdown tree or Miro-style diagram (exported). Via `$handoff`.
-
-## Anti-patterns
-
-| Error | Why it's bad | How to do it right |
-|-------|-------------|-------------------|
-| Solution-first tree | Skip opportunity space | Start with outcome → opportunities |
-| Opportunities = features | Just a renamed backlog | Opportunity = customer problem |
-| No experiments | Solutions not validated | Each solution has an experiment |
-| Tree never updated | Frozen OST = dead OST | Weekly/biweekly update |
-| Choosing solution before validation | Locked into first idea | Keep alternatives until experiment data |
-
-## Worked Example — TeamFlow OST (AI 1:1 Summarization initiative)
-
-**Context:** TeamFlow (B2B SaaS HR-tech, $8M ARR, NRR 105%). After Discovery (8 interviews + JTBD canvas), the team builds the OST for the chosen desired outcome. Question: «Should we build AI summarization?» — OST should show the opportunity space and alternatives, not jump to a conclusion.
-
-### Desired Outcome
-
-```
-Grow Net Revenue Retention from 105% to 115% within 4 quarters
-```
-
-**Why this metric:** NRR is the key SaaS metric for Board and investors. Growth in NRR drives ARR, moat, margin. A 10pp lift is tied to expansion + churn reduction, both achievable by increasing manager engagement with the product.
-
-**Why not another metric:**
-- «Activation rate» — too leading, doesn't prove business outcome.
-- «Weekly active teams» — vanity if not linked with revenue.
-- «ARR growth» — too lagging, NRR = leading indicator.
-
-### Opportunity Layer (from JTBD canvas + Discovery Brief)
-
-```
-Desired: Grow NRR 105% → 115%
-  │
-  ├── O1: Managers spend 3-4 hrs/week on 1:1 admin overhead
-  │   (buyer and end-user pain)
-  │   Evidence: 6 of 8 interviews, internal analytics shows prep time
-  │   Estimated reach: 3,000 active managers across customer base
-  │   Impact: retention + expansion (see Opp assessment)
-  │
-  ├── O2: VP HR lack visibility into manager 1:1 practice/quality across org
-  │   (buyer pain)
-  │   Evidence: 4 of 4 buyer interviews
-  │   Estimated reach: 200 buyer accounts
-  │   Impact: expansion opportunity (enterprise tier)
-  │
-  ├── O3: 40% of 1:1 action items dropped / forgotten
-  │   (end-user pain)
-  │   Evidence: 5 of 8 end-user interviews, internal product data (completed/total)
-  │   Estimated reach: 3,000 managers × 8 reports × 40% = significant
-  │   Impact: user retention + NPS
-  │
-  ├── O4: New-manager onboarding takes 6+ months until «effective 1:1»
-  │   (buyer pain F4)
-  │   Evidence: 3 of 4 buyer interviews
-  │   Estimated reach: ~1,000 new managers/year across customer base
-  │   Impact: smaller, but high-value for enterprise segment
-  │
-  └── O5: Cross-manager 1:1 experience inconsistency
-      (buyer + skip-level + employee pain)
-      Evidence: 2 of 4 buyer interviews, 1 end-user correlation
-      Estimated reach: Broad but fuzzy — hard to quantify
-      Impact: employee NPS / retention, but indirect
-```
-
-### Opportunity Assessment
-
-| # | Opportunity | Importance (1-10) | Reach (users) | Gap (current vs desired) | Priority Score |
-|---|-------------|:-----------------:|:-------------:|:------------------------:|:--------------:|
-| O1 | Admin overhead | 10 | ~3,000 | Big (no tool today) | 52,000 |
-| O3 | Action items dropped | 9 | ~3,000 | Big | 46,800 |
-| O2 | Buyer visibility | 9 | ~200 (but $$$) | Big | 1,800 (absolute) — but high deal size |
-| O4 | New-mgr onboarding | 7 | ~1,000/yr | Medium | 7,000 |
-| O5 | Cross-mgr consistency | 6 | Fuzzy | Medium | Hard to score |
-
-**Pruning:** focus on top 3: **O1 (admin overhead), O3 (action items), O2 (buyer visibility)** — all interconnected and one feature can address multiple.
-
-### Solutions Layer (for top 3 opportunities)
-
-```
-O1: Admin overhead
-  ├── S1.1: AI-powered real-time transcription + summarization during 1:1
-  │         (hypothesis: address capture-during-meeting pain + action items extraction)
-  ├── S1.2: Template library + auto-population from previous 1:1s
-  │         (lighter: not AI, but reduces prep time)
-  ├── S1.3: 1:1 duration reduction through structured agendas + time-boxing UI
-  │         (behavioral, not AI)
-  └── S1.4: Async 1:1s (structured written async for some updates)
-            (radical: reduce live meeting count)
-
-O3: Action items dropped
-  ├── S3.1: AI-extracted action items with automated reminders in the next 1:1
-  │         (ties to S1.1 — AI summarization output)
-  ├── S3.2: Calendar / Slack integration — reminders 2 days before 1:1
-  │         (lighter: rules-based)
-  └── S3.3: Weekly digest email «here's what you committed to» with status checkboxes
-
-O2: Buyer visibility
-  ├── S2.1: Aggregate dashboard: 1:1 frequency, duration, action items velocity per team
-  │         (rule-based analytics, no AI needed for basic version)
-  ├── S2.2: AI-inferred 1:1 health score (combining multiple signals)
-  │         (more advanced, needs S1.1 infrastructure)
-  └── S2.3: Quarterly People Ops report generator (auto-composed for Board)
-```
-
-### Experiments Layer (for top solutions)
-
-| Solution | Experiment | Prediction | Measurement | Time box | RICE score |
-|----------|------------|------------|-------------|----------|:----------:|
-| S1.1: AI summarization | Concierge MVP: human «AI» assists 10 beta customers manually for 2 weeks | Managers save 30 min/week; want to keep | Survey + retention | 3 weeks | 35 |
-| S1.1: AI summarization | Wizard-of-Oz: Real OpenAI API integration with 20 beta managers, tightly monitored | 20% drop-off acceptable; summarization quality > 70% acceptable | Manager survey + usage data | 4 weeks | 48 |
-| S1.2: Template library | Ship feature flag to 50% users, measure prep time via surveys | -30% prep time self-reported | Survey + funnel | 4 weeks | 28 |
-| S3.1: AI action items | Builds on S1.1 — separate evaluation: after AI summary generated, opt-in reminder system | +20% action items completed | Product analytics | 6 weeks | 30 |
-| S3.2: Rule-based reminders | Ship to all users, measure completion rate | +10% action items completed | Product analytics | 2 weeks | 25 |
-| S2.1: Aggregate dashboard | Beta 5 enterprise customers, qualitative interviews after 30 days | 4 of 5 say «would use weekly» | Interviews | 6 weeks | 22 |
-
-### Experiment Prioritization
-
-1. **S1.2 (Template library)** — low-risk, fast, independent. **Ship this first** (4 weeks).
-2. **S3.2 (Rule-based reminders)** — quick win, complements S1.2.
-3. **S1.1 (AI summarization) — Concierge MVP first, then Wizard-of-Oz** — validates value before engineering investment.
-4. **S2.1 (Dashboard)** — parallel track, different customer (enterprise), different eng workstream.
-5. **S3.1 (AI action items)** — deferred pending S1.1 results.
-
-### Iteration Notes (how this OST evolves)
-
-**Week 4 update (hypothetical):** S1.2 Template library shipped — survey shows only -12% prep time (not -30%). Learning: templates help new managers but experienced managers don't adopt. → Update OST: split O1 opportunity by manager tenure. New sub-opportunity: «New managers face steepest learning curve» (O4 recombined here).
-
-**Week 6 update:** S1.1 Concierge MVP results positive (9 of 10 want feature). Green-light Wizard-of-Oz + engineering investment. Branch expanded: add «privacy architecture» as a necessary sub-solution.
-
-**Week 10 update:** Wizard-of-Oz shows 72% summarization quality acceptable but privacy concerns from enterprise beta. Added new branch for compliance architecture.
-
-> **OST lesson:** Tree is a living document. The initial focus on «build AI» was correct, but through experiments it became clear that **template library + rule-based reminders** close 40-50% of value with 20% of effort. OST forced consideration of **alternatives** (S1.2, S3.2) that would otherwise have been skipped in favor of a better AI story.
-
-> **Reference to `$assumption-mapping`:** Each solution rests on assumptions. For S1.1 the top assumptions (privacy, LLM quality, willingness to pay) — tracked in the assumption map, validated through experiments.
-
-## Experiment Card Templates (per branch)
-
-Each solution in the OST has a prescribed experiment design. Use these templates.
-
-### Template A: Concierge MVP (validates Value)
-
-```markdown
-# Experiment: S1.1 Concierge MVP — AI Summarization
-
-**Tree branch:** Desired Outcome → O1 Admin Overhead → S1.1 AI Summarization
-**Status:** Active Week 2-5 (4-week experiment)
-
-## What we're testing
-Value of AI-summarized 1:1s — are managers willing to change their workflow?
-
-## Method
-Concierge (human-powered «AI»): 10 beta customer-managers email us meeting audio; PM + Designer manually produce summary + action items within 24 hours, deliver back in TeamFlow UI marked «AI-generated» (actually human).
-
-## Prediction
-- ≥ 7 of 10 managers want to continue after 2 weeks
-- Reported time-saving ≥ 2 hrs / week
-- Action items completion rate lifts ≥ 10pp
-
-## Measurement
-- Weekly survey (5-min)
-- Exit interview week 4
-- Self-reported time diary
-- Action items completion tracked in product
-
-## Success threshold
-≥ 7 of 10 say «want to keep» → green-light Wizard-of-Oz
-
-## Decision rules
-- 9+ positive → Strong signal, invest aggressively
-- 7-8 positive → Proceed to Wizard-of-Oz w/ standard investment
-- 5-6 positive → Iterate concierge, retest
-- <5 positive → Kill branch or major re-shape
-
-## Ties to assumption map
-Validates V1 (willingness to pay — indirectly via willingness to use), V2 (adoption intent)
-```
-
-### Template B: Wizard-of-Oz (validates Feasibility + Usability)
-
-```markdown
-# Experiment: S1.1 Wizard-of-Oz — AI Summarization (Real LLM)
-
-**Tree branch:** Same as Concierge, progression step
-**Status:** Scheduled Week 6-9 (after Concierge validation)
-
-## What we're testing
-(a) LLM quality — can real AI produce acceptable output?
-(b) User trust — will managers edit vs abandon?
-
-## Method
-Real OpenAI GPT-4 integration with 20 beta managers. Full pipeline: meeting recording → transcription → LLM summarization → manager review UI. Heavy telemetry + feedback.
-
-## Prediction
-- Summary quality ≥ 4 out of 5 avg (blind rating)
-- Edit rate 30-50% (not 0% = blind approve; not 100% = AI useless)
-- P95 latency ≤ 60s
-- Managers continue using after 4 weeks (≥ 80%)
-
-## Measurement
-- Summary rating (1-5) per meeting via in-product prompt
-- Edit telemetry (what % of summary changed)
-- Generation latency (server metrics)
-- Drop-off tracking (weekly active managers in beta)
-
-## Success thresholds
-- Quality: ≥ 85% of summaries rated ≥ 4
-- Edit rate: 25-55% (sweet spot)
-- Continuation: ≥ 80% still using at week 4
-
-## Decision rules
-- All thresholds met → Green-light GA investment
-- 2/3 met → Iterate prompt + retry week 10-11
-- 1/3 met → Consider feature gate (human-review layer)
-- 0/3 met → Branch kill
-
-## Ties to assumption map
-Primary validation for F1 (LLM quality) + U1 (manager trust)
-```
-
-### Template C: A/B Test (validates Scale)
-
-```markdown
-# Experiment: S3.2 Rule-Based Reminders — A/B Test
-
-**Tree branch:** Desired Outcome → O3 Action Items Dropped → S3.2 Rule-Based Reminders
-**Status:** Scheduled Week 12-16 (post-MVP, iteration)
-
-## What we're testing
-Do simple reminders (without AI) meaningfully improve action items completion rate?
-
-## Method
-Classic A/B. 50% of users receive auto-reminder 2 days before next 1:1 showing open action items. Control: no reminder.
-
-## Prediction
-- Action items completion lifts from 60% (baseline) to 68%+ (+8pp)
-- No increase in «reminder fatigue» (measured via opt-out rate)
-
-## Measurement
-- Action items completed / created ratio (weekly)
-- Opt-out rate of reminders
-- NPS impact (separate survey)
-
-## Sample & duration
-- Eligible users / week: ~500
-- Sample per variant: 3200 (calculated via `$ab-test-design`)
-- Duration: 13 weeks
-
-## Guardrails
-- Opt-out rate < 15%
-- NPS not declining
-
-## Decision rules
-- +8pp AND guardrails ok → Ship to all
-- +4 to +8pp → Ship with iterated copy
-- <+4pp → Kill, pivot to S3.1 AI reminders
-
-## Ties to assumption map
-Validates U2 (usability — do reminders annoy?) + V4 (lower-case value of simple solution)
-```
-
-### Template D: Design Partner Sessions (validates Enterprise Value)
-
-```markdown
-# Experiment: S2.1 Aggregate Dashboard — Enterprise Design Partner
-
-**Tree branch:** Desired Outcome → O2 Buyer Visibility → S2.1 Aggregate Dashboard
-**Status:** Week 6-12
-
-## What we're testing
-Will VP HR / CPO at enterprise customers upgrade tier based on dashboard value?
-
-## Method
-Concierge dashboard for 5 enterprise prospects. Manual data compilation weekly, delivered as «beta dashboard preview» Google Sheet. Track:
-- Usage frequency (do they check it?)
-- Reactions («would you pay for this?»)
-- Upgrade conversion (stated intent → actual upgrade)
-
-## Prediction
-- 4 of 5 prospects say «would use weekly»
-- 3 of 5 express upgrade intent
-- 2 of 5 actually upgrade within 30 days
-
-## Measurement
-- Qualitative interview notes
-- Upgrade CRM status
-- Dashboard access frequency (even if manual — we track requests)
-
-## Success thresholds
-- Use intent: 4 of 5
-- Upgrade conversion: ≥ 2 of 5
-
-## Decision rules
-- 2+ upgrade → Green-light productization
-- 0-1 upgrade → Re-evaluate dashboard scope based on feedback
-- Stalled at «like it but not upgrade» → Investigate pricing / packaging
-
-## Ties to assumption map
-Primary validation for V3 (buyer tier upgrade) + V4 (expansion driver)
-```
-
-### Experiment Portfolio Overview
-
-| # | Experiment | Method | Duration | Signal Type | Investment |
-|---|-----------|--------|:--------:|-------------|:----------:|
-| EXP-010 | S1.1 Concierge MVP | Manual | 4 weeks | Qualitative | Low ($2K) |
-| EXP-011 | S1.1 Wizard-of-Oz | Real LLM + heavy telemetry | 4 weeks | Quant + Qual | Medium ($15K) |
-| EXP-012 | S3.2 Rule-based reminders A/B | Production A/B | 13 weeks | Quantitative | Low ($5K — eng time only) |
-| EXP-013 | S2.1 Enterprise dashboard | Design partners | 6 weeks | Mixed | Medium ($10K) |
-| EXP-014 | S1.2 Template library A/B | Production A/B | 4 weeks | Quantitative | Low ($3K) |
-
-### Portfolio Balance
-
-- **Mix of methods:** qualitative (Concierge, Design Partner) + quantitative (A/B) + engineering-heavy (Wizard-of-Oz)
-- **Mix of speed:** fast (Concierge 4wk, reminders A/B early results 6wk) + slow (full A/B 13wk)
-- **Mix of investment:** low-risk ($2-5K) + medium ($10-15K)
-- **Decision dependencies:** Concierge → Wizard → Full AI investment (stepped commitment)
-
-> **Experiment portfolio lesson:** Mixing methods + durations prevents analysis paralysis. Fast low-investment experiments (Concierge) unlock confidence for higher investment (Wizard-of-Oz). Clear **decision rules per experiment** — if threshold not met, action is pre-specified (kill / iterate / pivot). Without pre-committed decision rules, «we'll see» becomes chronic postponement.
+---
+name: opportunity-solution-tree
+description: Opportunity-Solution Tree by Teresa Torres — desired outcome → opportunities → solutions → experiments
+type: triggered
+domain: product
+owners:
+  - discovery
+gates:
+  - DISCOVERY
+tech: []
+topic: []
+triggers:
+  - opportunity-solution-tree
+  - OST
+  - дерево возможностей
+  - Teresa Torres
+related: []
+budget_lines: 472
+schema_version: 1
+---
+# Opportunity-Solution Tree (OST)
+
+> **Category:** Discovery  ·  **Slug:** `opportunity-solution-tree`
+
+## When to Use
+
+- When you have a desired outcome (from strategy or OKR) but it's unclear **how** to achieve it.
+- During continuous discovery — as an artifact that evolves over time.
+- When the team jumps into solution mode without understanding the opportunity space.
+- When handing off discovery to engineering — OST shows the rationale behind solutions.
+
+## Input
+
+| Field | Required | Description |
+|-------|:--------:|-------------|
+| Desired outcome | ✅ | Business or product outcome (from NSM/OKR) |
+| Customer research | ✅ | Interviews, JTBD, problems |
+| Current assumptions | ⬚ | Which hypotheses already exist |
+| Time box | ⬚ | Usually 1-2 weeks for MVP tree |
+
+## Data Sources
+
+1. `$user-interview-script` transcripts — opportunities from pain points.
+2. `$jtbd-canvas` — opportunity jobs.
+3. Support tickets, NPS comments — evidence for opportunities.
+4. Product analytics — quantitative signal.
+
+### Related Skills
+
+| Skill | What we take | When to invoke |
+|-------|-------------|----------------|
+| `jtbd-canvas` | Underserved jobs → opportunities | When branching opportunities |
+| `problem-statement` | Specific problem → opportunity node | For articulating a specific opportunity |
+| `assumption-mapping` | Assumptions underlying solutions | For each solution node |
+| `hypothesis-template` | Solution → testable hypothesis | For experiment nodes |
+
+## Tree Structure
+
+```
+Desired Outcome (top)
+  │
+  ├── Opportunity 1 (customer problem / unmet job)
+  │     ├── Solution 1a (product idea)
+  │     │     └── Experiment 1a-i (test)
+  │     └── Solution 1b
+  │
+  ├── Opportunity 2
+  │     └── Solution 2a
+  │           └── Experiment 2a-i
+  │
+  └── Opportunity 3
+```
+
+## Protocol
+
+### Step 0 — Desired Outcome
+
+Formulate **one** outcome. Must be measurable, tied to a business metric.
+
+✅ Good: «Increase weekly active teams by 30% in 2 quarters»  
+❌ Bad: «Improve engagement»
+
+### Step 1 — Opportunities
+
+Opportunities are customer problems or unmet jobs. From JTBD + interviews.
+
+Rules:
+- Formulated as a **customer** statement, not a solution
+- Evidence-backed (evidence ≥ 2 sources or mark 🔮 assumed)
+- Mutually exclusive (no overlap)
+
+Limit: 3-7 opportunities at the top level. More → split into sub-branches.
+
+| # | Opportunity | Evidence | Estimated impact |
+|---|-------------|----------|-------------------|
+| O1 | [customer pain / unmet job] | JTBD F3, Interview 4-7 | High / Medium / Low |
+
+### Step 2 — Opportunity Assessment
+
+For each — importance × reach × gap (where the current solution falls short).
+
+Prune opportunities with low impact — focus on top 3.
+
+### Step 3 — Solutions
+
+For each **top** opportunity — 2-5 solution candidates. These are product ideas, features, process changes.
+
+Rules:
+- Specific (not «improve onboarding», but «in-app checklist for first 3 tasks»)
+- One opportunity can have multiple solutions — don't choose immediately
+- Solutions can be «non-product» (docs, training, pricing change)
+
+### Step 4 — Experiments
+
+For each solution — an experiment (or several) for validation.
+
+| Solution | Experiment | Prediction | Measurement | Time box |
+|----------|------------|------------|-------------|-----------|
+| In-app checklist | A/B test with 50% new users | +20% first-week activation | Funnel metric | 4 weeks |
+
+Experiments — via `$hypothesis-template` + `$ab-test-design`.
+
+### Step 5 — Prioritization
+
+Prioritize experiments, not solutions. Sometimes quick experiments reveal that an opportunity is weaker than it seemed.
+
+Use `$rice-scoring` at the experiment level.
+
+### Step 6 — Iteration
+
+OST is a living document. After each experiment:
+- Update evidence
+- Prune debunked branches
+- Add new opportunities from findings
+
+## Validation (Quality Gate)
+
+- [ ] Desired outcome — measurable, tied to a metric
+- [ ] 3-7 opportunities at the top level
+- [ ] Each opportunity — customer-framed, not a solution
+- [ ] Evidence (≥ 2 sources) or mark 🔮
+- [ ] Top 3 opportunities have ≥ 2 solutions
+- [ ] Each solution has ≥ 1 experiment
+- [ ] Experiments prioritized (RICE)
+
+## Handoff
+
+The result is input for:
+- **`rice-scoring`** — prioritization of experiments
+- **`hypothesis-template`** — formalize experiments
+- **PM**: solutions → feature candidates for roadmap
+- **`assumption-mapping`** — assumptions behind solutions
+
+Format: OST as markdown tree or Miro-style diagram (exported). Via `$handoff`.
+
+## Anti-patterns
+
+| Error | Why it's bad | How to do it right |
+|-------|-------------|-------------------|
+| Solution-first tree | Skip opportunity space | Start with outcome → opportunities |
+| Opportunities = features | Just a renamed backlog | Opportunity = customer problem |
+| No experiments | Solutions not validated | Each solution has an experiment |
+| Tree never updated | Frozen OST = dead OST | Weekly/biweekly update |
+| Choosing solution before validation | Locked into first idea | Keep alternatives until experiment data |
+
+## Worked Example — TeamFlow OST (AI 1:1 Summarization initiative)
+
+**Context:** TeamFlow (B2B SaaS HR-tech, $8M ARR, NRR 105%). After Discovery (8 interviews + JTBD canvas), the team builds the OST for the chosen desired outcome. Question: «Should we build AI summarization?» — OST should show the opportunity space and alternatives, not jump to a conclusion.
+
+### Desired Outcome
+
+```
+Grow Net Revenue Retention from 105% to 115% within 4 quarters
+```
+
+**Why this metric:** NRR is the key SaaS metric for Board and investors. Growth in NRR drives ARR, moat, margin. A 10pp lift is tied to expansion + churn reduction, both achievable by increasing manager engagement with the product.
+
+**Why not another metric:**
+- «Activation rate» — too leading, doesn't prove business outcome.
+- «Weekly active teams» — vanity if not linked with revenue.
+- «ARR growth» — too lagging, NRR = leading indicator.
+
+### Opportunity Layer (from JTBD canvas + Discovery Brief)
+
+```
+Desired: Grow NRR 105% → 115%
+  │
+  ├── O1: Managers spend 3-4 hrs/week on 1:1 admin overhead
+  │   (buyer and end-user pain)
+  │   Evidence: 6 of 8 interviews, internal analytics shows prep time
+  │   Estimated reach: 3,000 active managers across customer base
+  │   Impact: retention + expansion (see Opp assessment)
+  │
+  ├── O2: VP HR lack visibility into manager 1:1 practice/quality across org
+  │   (buyer pain)
+  │   Evidence: 4 of 4 buyer interviews
+  │   Estimated reach: 200 buyer accounts
+  │   Impact: expansion opportunity (enterprise tier)
+  │
+  ├── O3: 40% of 1:1 action items dropped / forgotten
+  │   (end-user pain)
+  │   Evidence: 5 of 8 end-user interviews, internal product data (completed/total)
+  │   Estimated reach: 3,000 managers × 8 reports × 40% = significant
+  │   Impact: user retention + NPS
+  │
+  ├── O4: New-manager onboarding takes 6+ months until «effective 1:1»
+  │   (buyer pain F4)
+  │   Evidence: 3 of 4 buyer interviews
+  │   Estimated reach: ~1,000 new managers/year across customer base
+  │   Impact: smaller, but high-value for enterprise segment
+  │
+  └── O5: Cross-manager 1:1 experience inconsistency
+      (buyer + skip-level + employee pain)
+      Evidence: 2 of 4 buyer interviews, 1 end-user correlation
+      Estimated reach: Broad but fuzzy — hard to quantify
+      Impact: employee NPS / retention, but indirect
+```
+
+### Opportunity Assessment
+
+| # | Opportunity | Importance (1-10) | Reach (users) | Gap (current vs desired) | Priority Score |
+|---|-------------|:-----------------:|:-------------:|:------------------------:|:--------------:|
+| O1 | Admin overhead | 10 | ~3,000 | Big (no tool today) | 52,000 |
+| O3 | Action items dropped | 9 | ~3,000 | Big | 46,800 |
+| O2 | Buyer visibility | 9 | ~200 (but $$$) | Big | 1,800 (absolute) — but high deal size |
+| O4 | New-mgr onboarding | 7 | ~1,000/yr | Medium | 7,000 |
+| O5 | Cross-mgr consistency | 6 | Fuzzy | Medium | Hard to score |
+
+**Pruning:** focus on top 3: **O1 (admin overhead), O3 (action items), O2 (buyer visibility)** — all interconnected and one feature can address multiple.
+
+### Solutions Layer (for top 3 opportunities)
+
+```
+O1: Admin overhead
+  ├── S1.1: AI-powered real-time transcription + summarization during 1:1
+  │         (hypothesis: address capture-during-meeting pain + action items extraction)
+  ├── S1.2: Template library + auto-population from previous 1:1s
+  │         (lighter: not AI, but reduces prep time)
+  ├── S1.3: 1:1 duration reduction through structured agendas + time-boxing UI
+  │         (behavioral, not AI)
+  └── S1.4: Async 1:1s (structured written async for some updates)
+            (radical: reduce live meeting count)
+
+O3: Action items dropped
+  ├── S3.1: AI-extracted action items with automated reminders in the next 1:1
+  │         (ties to S1.1 — AI summarization output)
+  ├── S3.2: Calendar / Slack integration — reminders 2 days before 1:1
+  │         (lighter: rules-based)
+  └── S3.3: Weekly digest email «here's what you committed to» with status checkboxes
+
+O2: Buyer visibility
+  ├── S2.1: Aggregate dashboard: 1:1 frequency, duration, action items velocity per team
+  │         (rule-based analytics, no AI needed for basic version)
+  ├── S2.2: AI-inferred 1:1 health score (combining multiple signals)
+  │         (more advanced, needs S1.1 infrastructure)
+  └── S2.3: Quarterly People Ops report generator (auto-composed for Board)
+```
+
+### Experiments Layer (for top solutions)
+
+| Solution | Experiment | Prediction | Measurement | Time box | RICE score |
+|----------|------------|------------|-------------|----------|:----------:|
+| S1.1: AI summarization | Concierge MVP: human «AI» assists 10 beta customers manually for 2 weeks | Managers save 30 min/week; want to keep | Survey + retention | 3 weeks | 35 |
+| S1.1: AI summarization | Wizard-of-Oz: Real OpenAI API integration with 20 beta managers, tightly monitored | 20% drop-off acceptable; summarization quality > 70% acceptable | Manager survey + usage data | 4 weeks | 48 |
+| S1.2: Template library | Ship feature flag to 50% users, measure prep time via surveys | -30% prep time self-reported | Survey + funnel | 4 weeks | 28 |
+| S3.1: AI action items | Builds on S1.1 — separate evaluation: after AI summary generated, opt-in reminder system | +20% action items completed | Product analytics | 6 weeks | 30 |
+| S3.2: Rule-based reminders | Ship to all users, measure completion rate | +10% action items completed | Product analytics | 2 weeks | 25 |
+| S2.1: Aggregate dashboard | Beta 5 enterprise customers, qualitative interviews after 30 days | 4 of 5 say «would use weekly» | Interviews | 6 weeks | 22 |
+
+### Experiment Prioritization
+
+1. **S1.2 (Template library)** — low-risk, fast, independent. **Ship this first** (4 weeks).
+2. **S3.2 (Rule-based reminders)** — quick win, complements S1.2.
+3. **S1.1 (AI summarization) — Concierge MVP first, then Wizard-of-Oz** — validates value before engineering investment.
+4. **S2.1 (Dashboard)** — parallel track, different customer (enterprise), different eng workstream.
+5. **S3.1 (AI action items)** — deferred pending S1.1 results.
+
+### Iteration Notes (how this OST evolves)
+
+**Week 4 update (hypothetical):** S1.2 Template library shipped — survey shows only -12% prep time (not -30%). Learning: templates help new managers but experienced managers don't adopt. → Update OST: split O1 opportunity by manager tenure. New sub-opportunity: «New managers face steepest learning curve» (O4 recombined here).
+
+**Week 6 update:** S1.1 Concierge MVP results positive (9 of 10 want feature). Green-light Wizard-of-Oz + engineering investment. Branch expanded: add «privacy architecture» as a necessary sub-solution.
+
+**Week 10 update:** Wizard-of-Oz shows 72% summarization quality acceptable but privacy concerns from enterprise beta. Added new branch for compliance architecture.
+
+> **OST lesson:** Tree is a living document. The initial focus on «build AI» was correct, but through experiments it became clear that **template library + rule-based reminders** close 40-50% of value with 20% of effort. OST forced consideration of **alternatives** (S1.2, S3.2) that would otherwise have been skipped in favor of a better AI story.
+
+> **Reference to `$assumption-mapping`:** Each solution rests on assumptions. For S1.1 the top assumptions (privacy, LLM quality, willingness to pay) — tracked in the assumption map, validated through experiments.
+
+## Experiment Card Templates (per branch)
+
+Each solution in the OST has a prescribed experiment design. Use these templates.
+
+### Template A: Concierge MVP (validates Value)
+
+```markdown
+# Experiment: S1.1 Concierge MVP — AI Summarization
+
+**Tree branch:** Desired Outcome → O1 Admin Overhead → S1.1 AI Summarization
+**Status:** Active Week 2-5 (4-week experiment)
+
+## What we're testing
+Value of AI-summarized 1:1s — are managers willing to change their workflow?
+
+## Method
+Concierge (human-powered «AI»): 10 beta customer-managers email us meeting audio; PM + Designer manually produce summary + action items within 24 hours, deliver back in TeamFlow UI marked «AI-generated» (actually human).
+
+## Prediction
+- ≥ 7 of 10 managers want to continue after 2 weeks
+- Reported time-saving ≥ 2 hrs / week
+- Action items completion rate lifts ≥ 10pp
+
+## Measurement
+- Weekly survey (5-min)
+- Exit interview week 4
+- Self-reported time diary
+- Action items completion tracked in product
+
+## Success threshold
+≥ 7 of 10 say «want to keep» → green-light Wizard-of-Oz
+
+## Decision rules
+- 9+ positive → Strong signal, invest aggressively
+- 7-8 positive → Proceed to Wizard-of-Oz w/ standard investment
+- 5-6 positive → Iterate concierge, retest
+- <5 positive → Kill branch or major re-shape
+
+## Ties to assumption map
+Validates V1 (willingness to pay — indirectly via willingness to use), V2 (adoption intent)
+```
+
+### Template B: Wizard-of-Oz (validates Feasibility + Usability)
+
+```markdown
+# Experiment: S1.1 Wizard-of-Oz — AI Summarization (Real LLM)
+
+**Tree branch:** Same as Concierge, progression step
+**Status:** Scheduled Week 6-9 (after Concierge validation)
+
+## What we're testing
+(a) LLM quality — can real AI produce acceptable output?
+(b) User trust — will managers edit vs abandon?
+
+## Method
+Real OpenAI GPT-4 integration with 20 beta managers. Full pipeline: meeting recording → transcription → LLM summarization → manager review UI. Heavy telemetry + feedback.
+
+## Prediction
+- Summary quality ≥ 4 out of 5 avg (blind rating)
+- Edit rate 30-50% (not 0% = blind approve; not 100% = AI useless)
+- P95 latency ≤ 60s
+- Managers continue using after 4 weeks (≥ 80%)
+
+## Measurement
+- Summary rating (1-5) per meeting via in-product prompt
+- Edit telemetry (what % of summary changed)
+- Generation latency (server metrics)
+- Drop-off tracking (weekly active managers in beta)
+
+## Success thresholds
+- Quality: ≥ 85% of summaries rated ≥ 4
+- Edit rate: 25-55% (sweet spot)
+- Continuation: ≥ 80% still using at week 4
+
+## Decision rules
+- All thresholds met → Green-light GA investment
+- 2/3 met → Iterate prompt + retry week 10-11
+- 1/3 met → Consider feature gate (human-review layer)
+- 0/3 met → Branch kill
+
+## Ties to assumption map
+Primary validation for F1 (LLM quality) + U1 (manager trust)
+```
+
+### Template C: A/B Test (validates Scale)
+
+```markdown
+# Experiment: S3.2 Rule-Based Reminders — A/B Test
+
+**Tree branch:** Desired Outcome → O3 Action Items Dropped → S3.2 Rule-Based Reminders
+**Status:** Scheduled Week 12-16 (post-MVP, iteration)
+
+## What we're testing
+Do simple reminders (without AI) meaningfully improve action items completion rate?
+
+## Method
+Classic A/B. 50% of users receive auto-reminder 2 days before next 1:1 showing open action items. Control: no reminder.
+
+## Prediction
+- Action items completion lifts from 60% (baseline) to 68%+ (+8pp)
+- No increase in «reminder fatigue» (measured via opt-out rate)
+
+## Measurement
+- Action items completed / created ratio (weekly)
+- Opt-out rate of reminders
+- NPS impact (separate survey)
+
+## Sample & duration
+- Eligible users / week: ~500
+- Sample per variant: 3200 (calculated via `$ab-test-design`)
+- Duration: 13 weeks
+
+## Guardrails
+- Opt-out rate < 15%
+- NPS not declining
+
+## Decision rules
+- +8pp AND guardrails ok → Ship to all
+- +4 to +8pp → Ship with iterated copy
+- <+4pp → Kill, pivot to S3.1 AI reminders
+
+## Ties to assumption map
+Validates U2 (usability — do reminders annoy?) + V4 (lower-case value of simple solution)
+```
+
+### Template D: Design Partner Sessions (validates Enterprise Value)
+
+```markdown
+# Experiment: S2.1 Aggregate Dashboard — Enterprise Design Partner
+
+**Tree branch:** Desired Outcome → O2 Buyer Visibility → S2.1 Aggregate Dashboard
+**Status:** Week 6-12
+
+## What we're testing
+Will VP HR / CPO at enterprise customers upgrade tier based on dashboard value?
+
+## Method
+Concierge dashboard for 5 enterprise prospects. Manual data compilation weekly, delivered as «beta dashboard preview» Google Sheet. Track:
+- Usage frequency (do they check it?)
+- Reactions («would you pay for this?»)
+- Upgrade conversion (stated intent → actual upgrade)
+
+## Prediction
+- 4 of 5 prospects say «would use weekly»
+- 3 of 5 express upgrade intent
+- 2 of 5 actually upgrade within 30 days
+
+## Measurement
+- Qualitative interview notes
+- Upgrade CRM status
+- Dashboard access frequency (even if manual — we track requests)
+
+## Success thresholds
+- Use intent: 4 of 5
+- Upgrade conversion: ≥ 2 of 5
+
+## Decision rules
+- 2+ upgrade → Green-light productization
+- 0-1 upgrade → Re-evaluate dashboard scope based on feedback
+- Stalled at «like it but not upgrade» → Investigate pricing / packaging
+
+## Ties to assumption map
+Primary validation for V3 (buyer tier upgrade) + V4 (expansion driver)
+```
+
+### Experiment Portfolio Overview
+
+| # | Experiment | Method | Duration | Signal Type | Investment |
+|---|-----------|--------|:--------:|-------------|:----------:|
+| EXP-010 | S1.1 Concierge MVP | Manual | 4 weeks | Qualitative | Low ($2K) |
+| EXP-011 | S1.1 Wizard-of-Oz | Real LLM + heavy telemetry | 4 weeks | Quant + Qual | Medium ($15K) |
+| EXP-012 | S3.2 Rule-based reminders A/B | Production A/B | 13 weeks | Quantitative | Low ($5K — eng time only) |
+| EXP-013 | S2.1 Enterprise dashboard | Design partners | 6 weeks | Mixed | Medium ($10K) |
+| EXP-014 | S1.2 Template library A/B | Production A/B | 4 weeks | Quantitative | Low ($3K) |
+
+### Portfolio Balance
+
+- **Mix of methods:** qualitative (Concierge, Design Partner) + quantitative (A/B) + engineering-heavy (Wizard-of-Oz)
+- **Mix of speed:** fast (Concierge 4wk, reminders A/B early results 6wk) + slow (full A/B 13wk)
+- **Mix of investment:** low-risk ($2-5K) + medium ($10-15K)
+- **Decision dependencies:** Concierge → Wizard → Full AI investment (stepped commitment)
+
+> **Experiment portfolio lesson:** Mixing methods + durations prevents analysis paralysis. Fast low-investment experiments (Concierge) unlock confidence for higher investment (Wizard-of-Oz). Clear **decision rules per experiment** — if threshold not met, action is pre-specified (kill / iterate / pivot). Without pre-committed decision rules, «we'll see» becomes chronic postponement.