Executive-ready repair ledger

# KintsugiGuard AI Impact Report

## Executive summary
Bloated SaaS backend can recover roughly $1,670 per month and 92 kgCO2e by repairing the highest-confidence cracks before adding more cloud capacity.

## Key waste findings
- Over-provisioned cloud service tier: $410/month, 28 kgCO2e
- Bloated Docker image: $320/month, 21 kgCO2e
- Excessive duplicate AI model calls: $290/month, 4 kgCO2e
- Idle scheduled workloads stay warm all day: $210/month, 17 kgCO2e
- Redundant CI/CD jobs repeat the same work: $180/month, 14 kgCO2e

## Recommended repairs
1. Share dependency cache, split changed-package checks, and collapse duplicate transpilation steps. (priority 5.58)
2. Use a multi-stage build, prune package caches, and ship only production artifacts. (priority 4.49)
3. Right-size ECS tasks to the observed p95 CPU and memory profile, then apply autoscaling guardrails. (priority 3.45)
4. Convert daily batch workers to event-driven jobs with a shutdown policy outside processing windows. (priority 3.36)
5. Remove unused packages and consolidate overlapping date, schema, and HTTP client libraries. (priority 3.12)
6. Add prompt-hash caching, retry de-duplication, and lower-cost routing for deterministic summaries. (priority 2.06)

## Governance summary
- Repairs auto-approved: 2
- Repairs requiring human review: 1
- Blocked by policy: 0
- Audit trail entries recorded for all agent actions

## Business and sustainability impact
- Estimated monthly savings: $1,670
- Estimated annual savings: $20,040
- Estimated monthly carbon reduction: 92 kgCO2e
- Engineering time saved monthly: 44 hours
- Repair risk level: Medium
- KintsugiGuard Score: 54 -> 84

## Suggested pull request summary
Slim runtime image and production dependencies. Add CI dependency cache and selective workflow execution. Introduce budget-aware AI API caching and routing. Apply cloud right-sizing behind canary and approval gates. Attach before/after cost, carbon, and runtime checks to the repair ledger.