docs: add leverage backlog items for underutilized build data

[carlos-alm]

Summary

• Add 19 backlog items (IDs 40-58) across 6 new tiers for features that leverage existing build data currently computed on every build but consumed by only 1-2 commands each
• Tier 1b — AST leverage (IDs 40-44): complexity from ast_nodes, lint predicates, triage density, dead class detection, migration patterns
• Tier 1c — CFG leverage (IDs 45-48): cyclomatic from CFG, unreachable blocks, audit summary, triage metrics
• Tier 1d — Dataflow leverage (IDs 49-52): data-dependent impact analysis, triage fan-out, check predicates, diff-impact enrichment
• Tier 1e — Co-change leverage (IDs 53-55): triage coupling, community comparison, missing partner warning
• Tier 1f — Embeddings leverage (IDs 56-58): similar functions, duplicate clusters, similarity in audit

Highest-value items are 49 and 52 (dataflow in impact analysis, problem-fit 5) — they directly address the core problem codegraph exists to solve.

Test plan

• Docs-only change, no code modified

[claude]

Claude finished @carlos-alm's task —— View job

---

PR Review: Documentation Backlog Items

Reviewing this pull request that adds 19 new backlog items for leveraging underutilized build data.

Todo List:

• Read the modified BACKLOG.md file to understand the changes
• Analyze the root cause and problem being solved
• Assess documentation quality and structure
• Provide final assessment with rating and recommendation

Analysis

This PR addresses a clear inefficiency: codegraph computes rich build data (AST nodes, CFG blocks, dataflow edges, co-change patterns, embeddings) on every build but only consumes each dataset with 1-2 commands. The new backlog items propose leveraging these existing data sources across more features.

The approach is strategically sound - instead of adding new data collection overhead, it maximizes value from data already being computed. The 6 new tiers (1b-1f) are organized by data source:

• Tier 1b (AST): Query ast_nodes for complexity, lint rules, dead code detection
• Tier 1c (CFG): Use cfg_blocks/cfg_edges for complexity, unreachable code detection
• Tier 1d (Dataflow): Integrate dataflow table into impact analysis, triage scoring
• Tier 1e (Co-change): Leverage co_changes for triage coupling, missing partner warnings
• Tier 1f (Embeddings): Use existing embeddings for similarity detection, duplicate clusters

Items 49 and 52 (dataflow in impact analysis) are correctly flagged as problem-fit 5 - they directly address codegraph's core purpose of preventing "missed blast radius" by expanding beyond call-chain analysis to data-flow dependencies.

Documentation Quality Assessment

Strengths:

• Consistent format: All new items follow the established 8-column table structure
• Clear organization: Logical grouping by data source with explanatory headers
• Appropriate prioritization: Items 49/52 correctly rated as highest value
• Implementation awareness: References existing tables (ast_nodes, cfg_blocks, etc.) show technical understanding
• Actionable descriptions: Each item clearly explains what would be built and why

Areas for improvement:

• Redundant language: Some descriptions repeat concepts ("agents can..." appears 12+ times)
• Minor inconsistencies: Mix of technical terms ("BFS pathfinding") vs plain language
• Category alignment: Some items span multiple categories but are placed in just one

Root Cause Analysis

Problem Being Solved: Resource inefficiency where expensive build-time computation is underutilized
Why This Approach: Maximizes ROI on existing data rather than adding new collection overhead
Risk Assessment: Very low - this is documentation-only with no implementation changes

Critical Concerns

None. This is a well-structured documentation enhancement that adds strategic clarity to the backlog.

Final Recommendation

• Rating: ⭐⭐⭐⭐☆ (4/5)
• Action: APPROVE
• Reasoning: High-quality documentation that identifies genuine value opportunities with clear technical understanding. The strategic focus on leveraging existing build data is sound, and the prioritization correctly emphasizes dataflow integration for core impact analysis use cases.

---