@sppg2001/atomize 1.2.2 → 2.0.0-alpha.0

package/docs/Story-Learner.md

@@ -0,0 +1,324 @@
+# Story Learner
+
+The Story Learner analyzes existing work items with their tasks and generates a reusable Atomize template. This is the fastest way to capture your team's proven task breakdown patterns.
+
+## Table of Contents
+
+- [Overview](#overview)
+- [Learn from a Single Story](#learn-from-a-single-story)
+- [Learn from Multiple Stories](#learn-from-multiple-stories)
+- [How Pattern Detection Works](#how-pattern-detection-works)
+- [Normalization](#normalization)
+- [Output Template](#output-template)
+- [Examples](#examples)
+- [Tips and Best Practices](#tips-and-best-practices)
+- [Troubleshooting](#troubleshooting)
+
+---
+
+## Overview
+
+Instead of building a template from scratch, the Story Learner:
+
+1. Fetches one or more work items that already have child tasks
+2. Analyzes task titles, estimations, activities, and tags
+3. Detects common patterns across stories (when using multiple stories)
+4. Generates a YAML template you can use immediately
+
+This is ideal when your team has a well-established workflow and you want to replicate it consistently.
+
+---
+
+## Learn from a Single Story
+
+Pass a single story ID to capture its task breakdown as a template directly — no pattern detection, no confidence scoring. The tasks are taken as-is and normalized to 100%.
+
+```bash
+atomize template create --from-stories STORY-123 --save-as my-template
+```
+
+This is useful when one story was broken down particularly well and you want to replicate it exactly. The generated template will include a note in its metadata that it was learned from a single story.
+
+> **When to use more stories:** With a single source story, the template reflects that story's specific breakdown, including any idiosyncrasies. Use [multiple stories](#learn-from-multiple-stories) to get a template that reflects your team's general pattern rather than one instance.
+
+---
+
+## Learn from Multiple Stories
+
+Use `--from-stories` to analyze multiple stories at once. This produces higher-quality templates because it detects patterns, filters out outliers, and generates confidence scores.
+
+```bash
+atomize template create --from-stories STORY-1,STORY-2,STORY-3
+```
+
+**With options:**
+
+```bash
+atomize template create \
+  --from-stories STORY-123,STORY-456,STORY-789 \
+  --platform azure-devops \
+  --save-as backend-pattern
+```
+
+### Options
+
+| Option | Description |
+|--------|-------------|
+| `--from-stories <ids>` | Comma-separated list of story IDs |
+| `--platform <name>` | Platform to fetch from |
+| `--profile <name>` | Azure DevOps connection profile to use |
+| `--save-as <name>` | Catalog name to save the generated template under |
+
+### Minimum Requirements
+
+- At least **1 story** required (3+ recommended for pattern detection)
+- Each story must have **at least 1 child task**
+- Stories should be similar in type and complexity
+
+---
+
+## How Pattern Detection Works
+
+When analyzing multiple stories, Atomize performs several steps:
+
+### 1. Task Similarity Analysis
+
+Tasks across stories are grouped by similarity. Task titles are compared to identify tasks that represent the same kind of work (e.g., "Design API", "API Design Phase", "Design REST endpoints" are all grouped as "design" tasks).
+
+### 2. Confidence Scoring
+
+Each pattern receives a confidence score based on how consistently it appears across stories:
+
+| Confidence | Meaning | Threshold |
+|------------|---------|-----------|
+| **High** | Task appears in most stories with consistent estimation | ≥ 70% of stories |
+| **Medium** | Task appears in some stories | 40-69% of stories |
+| **Low** | Task appears in few stories | < 40% of stories |
+
+Only high and medium confidence patterns are included in the generated template by default.
+
+### 3. Outlier Detection
+
+Stories that are significantly different from the others (e.g., have many more or fewer tasks, or very different estimation ratios) are flagged as outliers. You'll see a report of which stories were considered outliers and why.
+
+### 4. Estimation Averaging
+
+For tasks that appear in multiple stories, the estimation percentage is averaged across all instances (excluding outliers).
+
+### 5. Condition Pattern Detection
+
+If certain tasks only appear when stories have specific tags or properties, Atomize may suggest conditional task logic:
+
+```yaml
+tasks:
+  - title: "Security Review"
+    estimationPercent: 10
+    condition:
+      field: "tags"
+      operator: "contains"
+      value: "security"  # Suggested based on pattern
+```
+
+### 6. Tag Pattern Analysis
+
+Tags that consistently appear on matching stories are incorporated into the template's filter criteria.
+
+---
+
+## Normalization
+
+Task percentages are adjusted to sum to exactly 100% so the generated template is ready to validate and use.
+
+**Normalization behavior:**
+- Percentages are scaled proportionally to sum to 100%
+- Remainder is added to the largest task
+- Result always validates against `totalEstimationMustBe: 100`
+
+**Example:**
+
+Original tasks: Design 18%, Implementation 45%, Testing 20%, Review 8% = 91%
+
+After normalization:
+- Design: 20% (18/91 × 100)
+- Implementation: 49% (45/91 × 100)
+- Testing: 22% (20/91 × 100)
+- Review: 9% (8/91 × 100)
+
+---
+
+## Output Template
+
+The generated template includes:
+
+- **Basic info** — Name and description derived from the analyzed stories
+- **Filter criteria** — Based on story types, states, and detected tags
+- **Tasks** — One task per detected pattern, with averaged estimation
+- **Estimation settings** — Percentage strategy with nearest rounding
+- **Metadata** — Notes about the learning process, confidence scores
+
+**Example output:**
+
+```yaml
+version: "1.0"
+name: "Learned: Backend API Pattern"
+description: "Generated from 3 stories (STORY-123, STORY-456, STORY-789)"
+author: "Atomize Story Learner"
+tags: ["backend", "api", "learned"]
+
+filter:
+  workItemTypes: ["User Story"]
+  states: ["New", "Active"]
+  tags:
+    include: ["backend"]
+  excludeIfHasTasks: true
+
+tasks:
+  - title: "Design API"
+    estimationPercent: 18
+    activity: "Design"
+    # Confidence: high (3/3 stories)
+
+  - title: "Implementation"
+    estimationPercent: 47
+    activity: "Development"
+    # Confidence: high (3/3 stories)
+
+  - title: "Unit Tests"
+    estimationPercent: 20
+    activity: "Testing"
+    # Confidence: high (3/3 stories)
+
+  - title: "Integration Tests"
+    estimationPercent: 10
+    activity: "Testing"
+    # Confidence: medium (2/3 stories)
+
+  - title: "Code Review"
+    estimationPercent: 5
+    activity: "Documentation"
+    # Confidence: high (3/3 stories)
+
+estimation:
+  strategy: "percentage"
+  rounding: "nearest"
+  minimumTaskPoints: 0.5
+
+metadata:
+  category: "Learned"
+  estimationGuidelines: "Learned from stories with avg estimation of 9.3 points"
+```
+
+---
+
+## Examples
+
+### Learn from Multiple Stories
+
+```bash
+# Gather your best-structured stories
+atomize template create \
+  --from-stories STORY-100,STORY-115,STORY-132,STORY-148 \
+  --platform azure-devops \
+  --save-as backend-standard
+
+# Validate the result
+atomize validate template:backend-standard --strict
+
+# Apply to new stories
+atomize generate template:backend-standard
+```
+
+---
+
+## Tips and Best Practices
+
+### Choosing Good Source Stories
+
+- Pick stories that were broken down **after sprint planning**, not ad-hoc
+- Use stories where the task breakdown was considered **high quality** by your team
+- Choose stories of **similar complexity** (similar story point range)
+- Use **at least 5 stories** for the most reliable patterns
+
+### Reviewing the Output
+
+After generating, always:
+
+1. **Validate the template:**
+   ```bash
+	   atomize validate template:my-learned-template --strict
+   ```
+
+2. **Review the YAML** — the learner may suggest conditions or patterns you want to refine
+
+3. **Test with mock data:**
+   ```bash
+	   atomize generate template:my-learned-template --platform mock
+   ```
+
+4. **Preview against real data before executing:**
+   ```bash
+	   atomize generate template:my-learned-template
+   ```
+
+### Improving Generated Templates
+
+After generation, consider:
+- Renaming generic task titles to be more specific
+- Adding structured `condition` fields for tasks that don't always apply
+- Adding `dependsOn` to enforce task ordering
+- Adding `assignTo` patterns appropriate for your team
+- Adding `${story.title}` variable interpolation to task titles
+- Adjusting the `filter` section to be more or less restrictive
+
+---
+
+## Troubleshooting
+
+### "Story not found"
+
+```
+Error: Work item STORY-123 not found
+```
+
+- Verify the story ID exists in your project
+- Check your platform credentials are configured correctly
+- Ensure the story is accessible with your PAT permissions
+
+### "Story has no child tasks"
+
+```
+Warning: STORY-123 has no child tasks — cannot learn from it
+```
+
+- The selected story must already have child task work items
+- Add tasks to the story manually first, then run the learner
+
+### "Not enough stories with tasks"
+
+```
+Error: Only 1 of 3 provided stories had child tasks
+```
+
+- Check that all provided story IDs have child tasks
+- The learner needs stories that already have a task breakdown
+
+### "Low confidence patterns only"
+
+If all detected patterns have low confidence:
+- Try providing more stories (5+ recommended)
+- Ensure the stories have similar structures
+- Check that stories are of the same work item type
+
+### "Generated template has estimation ≠ 100%"
+
+- Manually edit the saved catalog template to balance estimations
+- Run `atomize validate template:<name> --strict` again before using it
+
+---
+
+## See Also
+
+- [CLI Reference](./Cli-Reference.md) - Full `template create` options
+- [Template Reference](./Template-Reference.md) - Template schema for refining output
+- [Validation Modes](./Validation-Modes.md) - Validate generated templates
+- [Platform Guide](./Platform-Guide.md) - Setting up Azure DevOps credentials