Prisma Postgres Cost Optimization Audit

You are a senior backend engineer specializing in Prisma ORM, Postgres performance, and serverless cost optimization. Audit this codebase and produce a single actionable report on reducing Prisma + Postgres resource usage and billing.

## RULES
- **Read-only**: Do NOT modify files, run formatters, or apply fixes.
- **Single output**: Create `docs/audits/PRISMA_EFFICIENCY_AUDIT_XX.md` (increment XX if prior audits exist; never overwrite).

## OBJECTIVE
Minimize database costs by identifying: wasteful queries, missing indexes, N+1 patterns, overfetching, connection issues, and hot paths.

## SCOPE
- `schema.prisma`, migrations, indexes, relations
- All Prisma Client calls (`findMany`, `findFirst`, `findUnique`, `create`, `update`, `upsert`, `delete`, `aggregate`, `groupBy`, `transaction`, `$queryRaw`, `$executeRaw`)
- API routes, server actions, cron jobs, webhooks, auth/session code
- Pagination, search/filters, loops containing queries
- Caching layers, pooling config, connection strings

## REPORT FORMAT

### 1. Executive Summary
Top 5 cost drivers with evidence. Quick wins vs refactors. Impact: High/Med/Low.

### 2. Query Efficiency Findings
For each issue:
- **Location**: file + function/line
- **Problem**: why inefficient
- **Fix**: conceptual solution
- **Impact/Complexity**: H/M/L

**Patterns to find:**
- `findMany` without `where`/`take`/`limit`
- Missing `select` (full row fetches)
- Deep `include` chains / overfetching relations
- N+1: queries inside loops (`for...await prisma.*`)
- Repeated identical queries per request (needs memoization)
- Multiple `create`/`update` instead of `createMany`/`updateMany`
- `findFirst` before `update` (use `update` directly)
- Offset pagination on large tables (use cursor)
- `count()` on large tables without cache
- Non-sargable: `contains`, `startsWith`, `mode: 'insensitive'` on unindexed columns
- Large text/JSON/blob fields fetched unnecessarily
- Long transactions, excessive transaction round-trips
- Sorting on unindexed columns
- Missing `take` on relation includes

### 3. Schema & Index Findings
For high-traffic tables:
- Missing indexes on FKs, filtered columns, `orderBy` columns
- Compound indexes for common `where`+`orderBy`
- Unique constraints replacing app-level checks
- Large fields to lazy-load

Format: proposed index → query patterns supported → trade-offs.

### 4. Connection Management
- Prisma Client instantiation (singleton/globalThis pattern?)
- Pooled vs direct connection strings
- Edge runtime connection churn
- Concurrent query spikes

### 5. Caching Opportunities
Where caching cuts reads: request-scoped deduplication, app cache, HTTP/ISR. Include invalidation strategy.

### 6. Prioritized Actions
| Priority | Recommendation | Impact | Complexity | Location |
|----------|----------------|--------|------------|----------|

P0 = immediate wins. Focus on highest DB load reduction.

## PROCESS
1. Grep for `prisma.`, `PrismaClient`, `$queryRaw`, `transaction`
2. Map all DB touchpoints (routes, jobs, components)
3. Trace hot paths and repeated patterns
4. Cite evidence; use "likely" when uncertain

**Output**: Only the single audit markdown file.