Why it exists
"What does 5C mean on my Samsung washer, and how do I fix it?" is answered today by ad-riddled content farms with no structure. No developer-ready dataset maps (brand, appliance_type, code) → meaning → ranked fix with honest provenance. ApplianceDB does.
Composite code identity
Samsung washer 5C ≠ Samsung dishwasher 5C ≠ Samsung UK washer 5C. Every record is unique on (brand, market, appliance_type, code).
Ranked fixes with a stated basis
Each repair carries an explicit rank_basis (manufacturer_first or cost_ascending) and a DIY difficulty tier.
Provenance per row
Every code, fix, and part row carries its source tier and exact source URL — plus dated per-retailer price observations. No memory sourcing, ever.
Facts, not prose
Meanings and steps are paraphrased into original wording — no copyrighted manual text is reproduced.
Coverage each pair ≥ 10 codes
| Brand | Washer | Dryer | Dishwasher | Refrigerator | Oven/Range |
|---|---|---|---|---|---|
| LG | 14 | 14 | 11 | 16 | 18 |
| Samsung | 16 | 14 | 14 | 14 | — |
| GE | — | — | 13 | — | 16 |
| Whirlpool | 18 | 12 | — | — | 11 |
| Maytag | 18 | — | — | — | — |
| Bosch | 20 | — | 18 | — | — |
| Frigidaire | 19 | — | 17 | — | — |
| Samsung (UK) | 26 | — | — | — | — |
| Hotpoint (UK) | 17 | — | — | — | — |
| Indesit (UK) | 17 | — | — | — | — |
| Beko (UK) | 13 | — | — | — | — |
| Miele (UK) | 27 | — | — | — | — |
| Candy (UK) | 22 | — | — | — | — |
| Hoover (UK) | 23 | — | — | — | — |
Coverage honesty over width: GE refrigerator (8 codes) and GE combo washer (2) were dropped from v1 for falling below the ≥ 10-per-pair gate rather than padded.
Browse the code pages
Every one of the 438 verified codes has its own page with the meaning, the implicated component, ranked repair steps and the source it was derived from. Pick an appliance to see its full list.
The code pages are free to read. The paid snapshot is the same data as structured CSV and SQLite, with the ranked repairs, OEM part numbers and dated price observations joined per row.
How it's built — the honest part
- No fabricated codes, ever. Every code is re-derived from a fetched source listing — official manufacturer pages for most, disclosed secondary references for the rest — never from model memory. A
source_typecolumn flags the provenance tier of every row. - Ranks earn a basis. A rank without stated evidence is a guess, and guesses don't ship.
- NULL over guess. Costs and part numbers are left NULL wherever no verified observation exists. Where present, part numbers are corroborated against the manufacturer's own parts store (two UK-market parts rest on UK retailer listings only) and every price range ships with its dated per-retailer observations.
- Deterministic, test-gated build. Composite-uniqueness, controlled-vocabulary, rank-contiguity, provenance, and coverage gates all run in CI.
Get the data
The free developer sample is in the identical schema as the commercial corpus — CSV + Parquet.
| File | Rows | Contents |
|---|---|---|
error_codes.csv | 438 | Code registry with meaning, component, cause, severity, source type + URL |
repair_procedures.csv | 288 | Ranked fixes with difficulty, paraphrased steps, currency-aware cost ranges, and per-row provenance |
replacement_parts.csv | 20 | Exact OEM part numbers with dated street-price observations in USD and GBP |
error_codes_fixes_joined.csv | 479 | Pre-joined analytical view for one-file ingestion |
brand_recalls.csv | 69 | Appliance recalls per brand, cross-referenced from RecallDB with agency provenance |
import pandas as pd
df = pd.read_csv("error_codes_fixes_joined.csv")
lg_drain = df[(df["brand"] == "LG") & (df["cause_category"] == "drainage")]
print(lg_drain[["appliance_type", "code", "meaning", "title", "diy_difficulty"]])Pricing
| Tier | What you get | Price |
|---|---|---|
| Sample | 438 codes (this repo + Kaggle) · CSV + Parquet | Free |
| Repair Intelligence Snapshot | Commercial licence + SQLite and Parquet builds of the corpus · curated snapshot (edition 2026.09), refreshed when the corpus changes · instant download | Buy — $99 |
| Enterprise Integration License | Snapshot + commercial embedding rights (warranty / field-service) · parts cross-reference · priority coverage | Request — $2,999 |
$99 one-time · commercial licence + SQLite and Parquet builds of the corpus. The row counts are the same as the free sample (438 error codes, 288 repair procedures, 479 joined rows, 69 recalls, 17 OEM parts (20 part-to-repair links)): you are paying for the licence and the packaged builds, not for extra rows.
The Snapshot is self-serve: secure Stripe checkout (card / Apple Pay / Google Pay) with instant download after payment. Enterprise is scoped to your platform — tell us what you need and we reply with an invoice.
Licensing & custom coverage
Company invoice, custom brand or market coverage, or a question before you buy. We reply within 1 business day. Direct desk: appliancedb@dataengineered.io
License
Free sample — ODbL-1.0
This 438-code sample is free for research, education, and benchmarking with attribution and share-alike.
Full / commercial dataset
Separate commercial license included with the paid tiers above — commercial embedding, parts cross-references, and priority coverage requests.
⚠️ Repair steps are educational reference, not a substitute for the manufacturer's service manual. Disconnect power and water before servicing; gas, sealed-refrigerant, and high-voltage work is for qualified technicians only.