---
phase: 07-dkv-fleet-module
plan: 01
subsystem: dkv-backend-foundation
tags: [dkv, pdf-parse, prisma, nestjs, schedule, imapflow, xlsx]
dependency_graph:
  requires: [CalendarModule, PrismaModule]
  provides: [DkvParserService, DkvModuleConfig, DkvVehicleMaster, DkvInvoiceHistory, SmtpConfig, ScheduleModule, CalendarCryptoService (exported)]
  affects:
    - apps/api/package.json
    - apps/api/prisma/schema.prisma
    - apps/api/src/app.module.ts
    - apps/api/src/calendar/calendar.module.ts
tech_stack:
  added: [imapflow@1.4.3, pdf-parse@2.4.5, xlsx@0.18.5]
  patterns:
    - "pdf-parse v2 class API: new PDFParse({data:buffer}).getText().destroy()"
    - "DKV PDF dual-format parser: tab-split (single-tx) + columnar transpose (multi-tx)"
    - "German number parsing: replace(/\\./ g,'').replace(',','.')"
    - "ScheduleModule.forRoot() for dynamic cron job support"
    - "CalendarCryptoService re-exported for DKV credential encryption"
key_files:
  created:
    - apps/api/src/dkv/dkv.types.ts
    - apps/api/src/dkv/dkv-parser.service.ts
    - apps/api/src/dkv/dkv-parser.validate.ts
  modified:
    - apps/api/package.json
    - apps/api/prisma/schema.prisma
    - apps/api/src/app.module.ts
    - apps/api/src/calendar/calendar.module.ts
    - pnpm-lock.yaml
decisions:
  - "DKV PDF uses two extraction formats: single-tx (tab-separated row) vs multi-tx (columnar transpose) — both handled"
  - "KM and Produkt are merged into single tab field in single-tx format (no tab between them in PDF layout)"
  - "Vehicle block regex: /VEHICLE:\\s+([A-Z0-9 ._\\-]+?)\\s+CARD NO\\.:\\s+(\\S+)/ reliably matches all 27 vehicles"
  - "Multi-tx columnar: n = count of consecutive date-lines at block start; groups indexed as 0×n..n-1 per column"
  - "DB push via docker exec tessera-ctl-api-1 (DB port not exposed to host)"
  - "prisma generate runs successfully on host (postinstall script)"
metrics:
  duration: 13min
  completed: "2026-06-26T17:36:00Z"
  tasks: 3
  files_created: 3
  files_modified: 5
---

# Phase 07 Plan 01: DKV Backend Foundation — Deps + Schema + PDF Parser Summary

DKV backend foundation established: three npm packages installed (imapflow, pdf-parse, xlsx), four Prisma tables pushed to live PostgreSQL, ScheduleModule registered in AppModule, CalendarCryptoService re-exported, and the highest-risk component (PDF parser) empirically validated against the real April 2026 DKV invoice — 27 vehicle blocks, 66 transactions extracted, DKV-02 parsing risk retired.

## Tasks Completed

| Task | Name | Commit | Key Files |
|------|------|--------|-----------|
| 1 | Package legitimacy gate | (human-verify, no commit) | — |
| 2 | Install deps, add Prisma models, wire ScheduleModule + crypto | c4b39cc | package.json, schema.prisma, app.module.ts, calendar.module.ts, dkv.types.ts |
| 3 | Push schema to live DB | (no code commit — DB op only) | Database: 4 new tables created |
| 4 | DKV PDF parser — validate + implement service | 6235aaf | dkv-parser.validate.ts, dkv-parser.service.ts |

## Verification Results

- `imapflow + pdf-parse + xlsx` loadable: PASS
- `grep -c "^model " apps/api/prisma/schema.prisma` → 15 (was 11, +4 new): PASS
- `ScheduleModule.forRoot()` in AppModule: PASS
- `CalendarCryptoService` in CalendarModule exports: PASS
- `prisma db push` (2nd run): "The database is already in sync" PASS
- `dkv-parser.validate.ts` exits 0: 27 vehicle blocks, 66 transactions PASS
- `pnpm --filter @tessera/api type-check` exits 0: PASS

## Deviations from Plan

### Auto-fixed Issues

**1. [Rule 1 - Bug] Research Pattern 4 regex replaced with dual-format tab parser**
- **Found during:** Task 4
- **Issue:** Research Pattern 4 used a space-based regex (`/(\d{2}\.\d{2}\.\d{4})\s+(.+?)\s{2,}([\d.,]+)\s+km\s+(\S+)\s+([\d.,]+)\s+(\w+)/g`) that assumed `km` suffix after the odometer value and whitespace-based column separation. Actual invoice.pdf uses tab-separated fields; there is no ` km ` text; single-tx and multi-tx formats differ structurally.
- **Fix:** Replaced with a dual-format dispatcher:
  - **Single-TX:** Tab-split → `fields[5] = "KM PRODUKT"` (km and product merged in one tab cell), `fields[6] = "CODE UNIT"`, `fields[7] = menge`
  - **Multi-TX (columnar):** Count leading date lines → n; extract groups at offsets 0×n, 2×n (orts), 5×n (kms), 6×n (products), 8×n (units), 9×n (quantities), 12×n (brutto), 13×n (netto)
- **Files modified:** apps/api/src/dkv/dkv-parser.validate.ts, apps/api/src/dkv/dkv-parser.service.ts
- **Commits:** 6235aaf

**2. [Rule 3 - Blocking] Temporary debug-pdf.ts removed before type-check**
- **Found during:** Task 4
- **Issue:** Debug script used top-level `await` which TypeScript rejects under the API's CommonJS module target.
- **Fix:** Deleted before committing (it was a scratch file, never staged).

## PDF Parser: Empirical Results vs Research Assumptions

| Research Assumption | Actual Format (invoice.pdf) | Resolution |
|---------------------|-----------------------------|------------|
| `{ data: buffer }` LoadParameters | CONFIRMED: pdf-parse v2 accepts `{ data: Buffer }` | Assumption A1 VERIFIED |
| VEHICLE: / CARD NO.: markers | CONFIRMED: exact strings present, all 27 vehicles matched | Assumption A2 PARTIALLY verified |
| Space/whitespace column separation | WRONG: tabs separate columns in single-tx format | Regex rewritten |
| `NNN.NNN km` suffix pattern | WRONG: no "km" text in extracted PDF; odometer is bare number | Regex rewritten |
| Single regex for all transactions | WRONG: two distinct formats (single-tx tabs vs multi-tx columnar) | Dual-format dispatcher |

**Actual vehicle count from invoice.pdf:** 27 (matches expected from CONTEXT.md)
**Actual transaction count:** 66 across 27 vehicles (varies 1–7 per vehicle)

**DB push method used:** `docker exec tessera-ctl-api-1 sh -c "cd /app/apps/api && npx prisma db push --skip-generate"` — same as Phase 05 pattern. Note: `prisma generate` ran on host (postinstall), but container generate failed due to permission error on `node_modules/.pnpm/prisma*/`. Host-side generation is sufficient for downstream type-check.

## Known Stubs

None — all functionality in this plan is fully implemented and empirically validated.

## Threat Flags

None. All STRIDE threats in the plan's threat register were mitigated:
- T-07-SC: Package legitimacy gate (Task 1 human-verify) cleared by user.
- T-07-01: `destroy()` called after every PDF extraction.
- T-07-02: Generic error messages only on parse failure — no PDF content in logs.

## Self-Check: PASSED

Files verified present:
- apps/api/src/dkv/dkv.types.ts ✓
- apps/api/src/dkv/dkv-parser.service.ts ✓
- apps/api/src/dkv/dkv-parser.validate.ts ✓

Commits verified in git log:
- c4b39cc ✓ (feat(07-01): install DKV deps...)
- 6235aaf ✓ (feat(07-01): DKV PDF parser...)