# Seed and demo scenarios

## Dataset

`seed/generate_seed.py` creates a deterministic snapshot for a chosen anchor date and random seed. The committed CSV files are generated with:

```bash
python seed/generate_seed.py --anchor 2026-08-01 --seed 20260801
```

Target scale:

- 50 vehicles;
- 180 customers including three duplicate pairs;
- 220 historical and 25 current/future bookings;
- realistic inspections and maintenance records;
- fixed quality and workflow scenarios.

## Required named scenarios

### S1 — Odometer regression return

Booking `BK-DEMO-RETURN` for vehicle `MO-024` is active. Submit a return reading below the canonical odometer. Expected: inspection saved, canonical odometer unchanged, vehicle blocked or cleaning according to other flags, quality issue created, event queued and audit visible.

### S2 — Duplicate customer merge

Customers `CUS-0012` and `CUS-0178` share normalized contact data and similar names. Expected: issue with explainable signals, merge rewires bookings, loser becomes tombstone, audit preserved.

### S3 — Missing inspection before next booking

Vehicle `MO-031` has a near-future booking and an unresolved operational attention item. Expected: dashboard links to its record.

### S4 — Legacy booking overlap

Vehicle `MO-016` has two imported overlapping reservations. Expected: visible quality issue; normal booking command would reject the same overlap.

### S5 — Failed workflow

One seeded outbox/workflow record is failed with a safe simulated connection error. Expected: dashboard and Automation page show it; Operations Manager can retry.

### S6 — Grounded damage question

Question: “What must I do when a vehicle returns with damage?” Expected: answer cites damage handling and return inspection procedures.

## Date anchoring

The committed CSVs store absolute ISO timestamps authored around a fixed anchor date
(`SEED_AUTHORED_ANCHOR = 2026-08-01` in `backend/app/seed_loader.py`, matching the
`--anchor` used to generate them). Every seed/reset shifts every seeded booking,
inspection, maintenance and outbox timestamp by `today − SEED_AUTHORED_ANCHOR`, so
"today"/"near-future"/"currently overlapping" scenarios stay true to the real moment the
environment was (re)seeded instead of decaying as real time passes between resets. Public
refs and entity relationships are untouched by the shift — only datetime columns move.
`load_seed()` returns the resolved `anchor_date`/`seeded_at`, and records a
`demo_data_seeded` audit event carrying both the resolved anchor and the original
authoring anchor, so the shift applied on any given reset stays traceable.

## Demo reset

Reset must:

- require Operations Manager;
- rebuild the deterministic dataset;
- re-establish scenario references;
- re-anchor scenario dates to the real reset moment (see above);
- clear non-seed audit/workflow state;
- complete safely and visibly;
- be covered by a test.